Upload
vuthien
View
221
Download
0
Embed Size (px)
Citation preview
Research ArticleMathematical Modeling for Lateral DisplacementInduced by Wind Velocity Using Monitoring Data Obtainedfrom Main Girder of Sutong Cable-Stayed Bridge
Gao-Xin Wang and You-Liang Ding
The Key Laboratory of Concrete and Prestressed Concrete Structures of Ministry of EducationSoutheast University Nanjing 210096 China
Correspondence should be addressed to You-Liang Ding civilchinahotmailcom
Received 2 April 2014 Accepted 22 May 2014 Published 19 June 2014
Academic Editor Giuseppe Rega
Copyright copy 2014 G-X Wang and Y-L Ding This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited
Based on the health monitoring system installed on the main span of Sutong Cable-Stayed Bridge GPS displacement andwind field are real-time monitored and analyzed According to analytical results apparent nonlinear correlation with certaindiscreteness exists between lateral static girder displacement and lateral static wind velocity thus time series of lateral staticgirder displacement are decomposed into nonlinear correlation term and discreteness term nonlinear correlation term of whichis mathematically modeled by third-order Fourier series with intervention of lateral static wind velocity and discreteness term ofwhich ismathematicallymodeled by the combinedmodels of ARMA(7 4) and EGARCH(2 1) Additionally stable power spectrumdensity exists in time series of lateral dynamic girder displacement which can be well described by the fourth-order Gaussianseries thus time series of lateral dynamic girder displacement are mathematically modeled by harmonic superposition functionBy comparison and verification between simulative and monitoring lateral girder displacements from September 1 to September 3the presented mathematical models are effective to simulate time series of lateral girder displacement from main girder of SutongCable-Stayed Bridge
1 Introduction
Nowadays long span cable-stayed and suspension bridgestructures are commonly constructed at home and abroadOn account of their flexible structural characteristics dis-placement response from main girder of long-span bridgestructure swings obviously impacted by strong aerostatic andfluctuating wind actions According to aerostatic responseanalysis on Sutong Cable-Stayed Bridge by Xu et al thelateral displacement response frommain girder can approach12m under strong wind velocity 40ms with attack angle 0∘[1] and research results from buffeting response analysis onGolden Gate Bridge by Vincent showed that extreme buffet-ing amplitude frommain girder can reach 17m under strongwind velocity 31ms [2] Such large amplitude can definitelythreaten comfort and safety of the whole bridge structure Forexample severe wind vibration from main girder of Tacoma
Suspension Bridge in Washington state eventually broughtabout collapse of the whole bridge structure under windvelocity 19ms [3] Therefore it is of great significance toresearch displacement response impacted bywind loads frommain girder of long span bridge structures and especially thelateral displacement response as one fairly important part formain girder should be specifically valued
Theoretical exploration numerical simulation and windtunnel tests for lateral displacement response have beencarried out to some extent Cheng andXiao improved the cal-culationmethod for aerostatic stability and further concludedthat instable lateral displacement was 424m under criticalstatic wind [4] Long et al analyzed lateral displacementresponse from Sidu Suspension Bridge through ANSYS finiteelement simulation and concluded that maximum lateraldisplacement at middle span was 3226 cm which is closeto 11000 length of main span [5] Yu et al researched
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2014 Article ID 723152 16 pageshttpdxdoiorg1011552014723152
2 Mathematical Problems in Engineering
2088100 100 1001003001088
3D ultrasonic anemometer
GPS monitoring station 300
Figure 1 Longitudinal layout of monitoring equipment on the Sutong Bridge (unit m)
41000
354002
4000
90009000 23000
GPSMS GPSMS
3DUA 3DUAUpstream Downstream
354002
Figure 2 Transverse layout of monitoring equipment on the flat steel box girder (unit mm) (1)GPSMS GPSmonitoring station (2) 3DUA3D ultrasonic anemometer
lateral displacement response from Xihoumen SuspensionBridge through wind tunnel tests and concluded that lateraldisplacement at horizontal angle 10∘ was larger than that ofother angles [6]
However for mechanism complexity of lateral dis-placement response impacted by aerostatic and fluctuatingwind actions traditional methods of theoretical deductionnumerical simulation and wind tunnel tests are difficult toaccurately reflect the actual lateral displacement responseof bridge structure on account of uncertain boundarycondition imprecise assignment of initial parameters andinappropriate ignorance of subordinate factors In recentyears with development of structural healthmonitoring tech-nology it is feasible to installmonitoring sensors on long spanbridge structures monitoring data of which can authenticallyreflect bridge structural behaviors under actual environmentand load actions Although wind field of long span bridgestructures has been widely monitored in recent years [7ndash9]lateral displacement response is rarely monitored andresearched thus real correlation regularity between lateraldisplacement response and wind action is still covered Addi-tionally lateral displacement response under actual operationenvironment is also affected by other random factors whichhave never been taken into account by researchers beforeTherefore lateral displacement response from main girder isnecessarily researched upon monitoring data to reveal realstructural behavior of long span bridges
In this paper based on the health monitoring systeminstalled on the main span of Sutong Cable-Stayed BridgeGPS displacement and wind field are real-time monitoredand analyzed According to analytical results apparent non-linear correlation with certain discreteness exists between
lateral static girder displacement and lateral static windvelocity thus time series of lateral static girder displace-ment are decomposed into nonlinear correlation term anddiscreteness term nonlinear correlation term of which ismathematically modeled by 119899th-order Fourier series withintervention of lateral static wind velocity and discretenessterm of which is mathematically modeled by the combinedmodels of ARMA(119901 119902) and EGARCH(119898 119899) Additionallystable power spectrum density exists in time series of lateraldynamic girder displacement thus time series of lateraldynamic girder displacement are mathematically modeledby harmonic superposition function By comparison andverification between simulative and monitoring lateral dis-placements from September 1 to September 3 mathematicalmodels are feasible and effective to simulate time series oflateral girder displacement frommain girder of SutongCable-Stayed Bridge
2 Bridge Monitoring and Sample Analysis
The bridge monitoring object for this research is theworldwide famous Sutong Cable-Stayed Bridge (in JiangsuProvince China) Its whole structure form is single-spannedand double-hinged with the main span reaching 1088m asshown in Figure 1 and the main girder employs flat steelbox type with 363m wide and 40m high as shown inFigure 2 3D ultrasonic anemometers and GPS monitoringstation are installed on two flanks of midspan cross-sectionfrom main girder (resp shown in Figures 1 and 2) tocontinuously acquire wind data and displacement data withsample frequency of 1Hz
Mathematical Problems in Engineering 3
x
y
X (longitude orientation)
Y (latitude orientation)
z or Z
Along the bridge Across the bridge
Geographical north 106∘
106∘
Figure 3 Description of the defined local andWGS-84 coordinate system Notes (1) local coordinate system 119909-119910-119911 (2)WGS-84 coordinatesystem119883-119884-119885 (3) 90∘ of wind horizontal angle denotes the axis angle from 119909 to 119910 (within range of [0∘ 360∘]) (4) 90∘ of wind vertical angledenotes axis angle from 119910 to 119911 (within range of [minus90∘ 90∘]) (5) 106∘ axis angle exists between axes 119884 and 119910 of across the bridge
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Win
d ve
loci
ty al
ong y
-axi
s (m
s)
minus10
minus20
(a) Time series of original wind velocity
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Stat
ic w
ind
velo
city
(ms
)
minus10
minus20
(b) Time series of static wind velocity
0 5 10 15 20 25 30
0
5
10
15
Time (day)
Dyn
amic
win
d ve
loci
ty (m
s)
minus10
minus10
minus15
(c) Time series of dynamic wind velocity
Figure 4 Time series of wind velocity along 119910-axis in the whole August
4 Mathematical Problems in Engineering
0 5 10 15 20 25 30
005
015
025
035
Time (day)
Gird
er d
ispla
cem
ent a
long
y-a
xis (
m)
minus005
minus015
(a) Time series of girder displacement
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Stat
ic g
irder
disp
lace
men
t (m
)
minus01
(b) Time series of static girder displacement
0 5 10 15 20 25 30
0
004
008
012
Time (day)
Dyn
amic
gird
er d
ispla
cem
ent (
m)
minus012
minus008
minus004
(c) Time series of dynamic girder displacement
Figure 5 Time series of girder displacements along 119910-axis in the whole August
5 15 250
005
01
015
02
025
03
Static wind velocity (ms)
Stat
ic g
irder
disp
lace
men
t (m
)
minus15 minus5
Figure 6 Correlation scatter plots between static wind velocity and static girder displacement
Mathematical Problems in Engineering 5
Frequency (Hz)
Day 1 to day 8Day 9 to day 16 Day 17 to day 24
Day 25 to day 31
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
Figure 7 Power spectrum densities of dynamic girder displacement in different periods
0 10 20 300
006
012
018
024
03
Static wind velocity (ms)
Fitte
d lat
eral
gird
er d
ispla
cem
ent (
m)
minus20 minus10
(a) The fitting Fourier series of nonlinear correlation term
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(b) Time series of fitting static displacement
Figure 8 The fitting Fourier series and time series of fitting static displacement
Specifically the wind data from 3D ultrasonic anemome-ters embrace such three types as wind velocity horizontalangle and vertical angle in local coordinate system (Figure 3)and the girder displacement data from GPS monitoringstation contains absolute locations in WGS-84 coordinatesystem (Figure 3) which are supposed to deduct referencelocations for analysis Until now the storage amount ofmoni-toring data has increased to 93 million for each measurementpoint Such considerable monitoring data cannot be totallyapplied to actual analysis thus the monitoring data fromupstream flank in the year 2012 are specially chosen
Taking the monitoring data in the whole August forexample considering that lateral displacement effect of maingirder is primarily reflected by wind load across the SutongBridge hence time series of wind velocity and girder dis-placement are decomposed into the119910-axis in local coordinate
system as shown in Figures 4(a) and 5(a) respectively whichcan be obviously observed that either of two time seriescontain static variant trend in whole and dynamic stochasticfluctuation in part Such two kinds of variation characteristicscan be furthermore separated by 10-minute average process asshown in Figures 4(b) 4(c) 5(b) and 5(c) respectively
By comparison between Figures 4(b) and 5(b) similarvariation characteristics exist between static wind veloc-ity and static girder displacement which can be visu-ally described by correlation scatter plots as shown inFigure 6 indicating apparent nonlinear correlation similar toquadratic parabolic curve Therefore static girder displace-ment can bemathematically expressed by static wind velocitywith consideration of definite discreteness affected by otherrandom factors Moreover time series of dynamic girderdisplacement depict obvious steady stochastic fluctuation
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Mathematical Problems in Engineering
2088100 100 1001003001088
3D ultrasonic anemometer
GPS monitoring station 300
Figure 1 Longitudinal layout of monitoring equipment on the Sutong Bridge (unit m)
41000
354002
4000
90009000 23000
GPSMS GPSMS
3DUA 3DUAUpstream Downstream
354002
Figure 2 Transverse layout of monitoring equipment on the flat steel box girder (unit mm) (1)GPSMS GPSmonitoring station (2) 3DUA3D ultrasonic anemometer
lateral displacement response from Xihoumen SuspensionBridge through wind tunnel tests and concluded that lateraldisplacement at horizontal angle 10∘ was larger than that ofother angles [6]
However for mechanism complexity of lateral dis-placement response impacted by aerostatic and fluctuatingwind actions traditional methods of theoretical deductionnumerical simulation and wind tunnel tests are difficult toaccurately reflect the actual lateral displacement responseof bridge structure on account of uncertain boundarycondition imprecise assignment of initial parameters andinappropriate ignorance of subordinate factors In recentyears with development of structural healthmonitoring tech-nology it is feasible to installmonitoring sensors on long spanbridge structures monitoring data of which can authenticallyreflect bridge structural behaviors under actual environmentand load actions Although wind field of long span bridgestructures has been widely monitored in recent years [7ndash9]lateral displacement response is rarely monitored andresearched thus real correlation regularity between lateraldisplacement response and wind action is still covered Addi-tionally lateral displacement response under actual operationenvironment is also affected by other random factors whichhave never been taken into account by researchers beforeTherefore lateral displacement response from main girder isnecessarily researched upon monitoring data to reveal realstructural behavior of long span bridges
In this paper based on the health monitoring systeminstalled on the main span of Sutong Cable-Stayed BridgeGPS displacement and wind field are real-time monitoredand analyzed According to analytical results apparent non-linear correlation with certain discreteness exists between
lateral static girder displacement and lateral static windvelocity thus time series of lateral static girder displace-ment are decomposed into nonlinear correlation term anddiscreteness term nonlinear correlation term of which ismathematically modeled by 119899th-order Fourier series withintervention of lateral static wind velocity and discretenessterm of which is mathematically modeled by the combinedmodels of ARMA(119901 119902) and EGARCH(119898 119899) Additionallystable power spectrum density exists in time series of lateraldynamic girder displacement thus time series of lateraldynamic girder displacement are mathematically modeledby harmonic superposition function By comparison andverification between simulative and monitoring lateral dis-placements from September 1 to September 3 mathematicalmodels are feasible and effective to simulate time series oflateral girder displacement frommain girder of SutongCable-Stayed Bridge
2 Bridge Monitoring and Sample Analysis
The bridge monitoring object for this research is theworldwide famous Sutong Cable-Stayed Bridge (in JiangsuProvince China) Its whole structure form is single-spannedand double-hinged with the main span reaching 1088m asshown in Figure 1 and the main girder employs flat steelbox type with 363m wide and 40m high as shown inFigure 2 3D ultrasonic anemometers and GPS monitoringstation are installed on two flanks of midspan cross-sectionfrom main girder (resp shown in Figures 1 and 2) tocontinuously acquire wind data and displacement data withsample frequency of 1Hz
Mathematical Problems in Engineering 3
x
y
X (longitude orientation)
Y (latitude orientation)
z or Z
Along the bridge Across the bridge
Geographical north 106∘
106∘
Figure 3 Description of the defined local andWGS-84 coordinate system Notes (1) local coordinate system 119909-119910-119911 (2)WGS-84 coordinatesystem119883-119884-119885 (3) 90∘ of wind horizontal angle denotes the axis angle from 119909 to 119910 (within range of [0∘ 360∘]) (4) 90∘ of wind vertical angledenotes axis angle from 119910 to 119911 (within range of [minus90∘ 90∘]) (5) 106∘ axis angle exists between axes 119884 and 119910 of across the bridge
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Win
d ve
loci
ty al
ong y
-axi
s (m
s)
minus10
minus20
(a) Time series of original wind velocity
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Stat
ic w
ind
velo
city
(ms
)
minus10
minus20
(b) Time series of static wind velocity
0 5 10 15 20 25 30
0
5
10
15
Time (day)
Dyn
amic
win
d ve
loci
ty (m
s)
minus10
minus10
minus15
(c) Time series of dynamic wind velocity
Figure 4 Time series of wind velocity along 119910-axis in the whole August
4 Mathematical Problems in Engineering
0 5 10 15 20 25 30
005
015
025
035
Time (day)
Gird
er d
ispla
cem
ent a
long
y-a
xis (
m)
minus005
minus015
(a) Time series of girder displacement
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Stat
ic g
irder
disp
lace
men
t (m
)
minus01
(b) Time series of static girder displacement
0 5 10 15 20 25 30
0
004
008
012
Time (day)
Dyn
amic
gird
er d
ispla
cem
ent (
m)
minus012
minus008
minus004
(c) Time series of dynamic girder displacement
Figure 5 Time series of girder displacements along 119910-axis in the whole August
5 15 250
005
01
015
02
025
03
Static wind velocity (ms)
Stat
ic g
irder
disp
lace
men
t (m
)
minus15 minus5
Figure 6 Correlation scatter plots between static wind velocity and static girder displacement
Mathematical Problems in Engineering 5
Frequency (Hz)
Day 1 to day 8Day 9 to day 16 Day 17 to day 24
Day 25 to day 31
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
Figure 7 Power spectrum densities of dynamic girder displacement in different periods
0 10 20 300
006
012
018
024
03
Static wind velocity (ms)
Fitte
d lat
eral
gird
er d
ispla
cem
ent (
m)
minus20 minus10
(a) The fitting Fourier series of nonlinear correlation term
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(b) Time series of fitting static displacement
Figure 8 The fitting Fourier series and time series of fitting static displacement
Specifically the wind data from 3D ultrasonic anemome-ters embrace such three types as wind velocity horizontalangle and vertical angle in local coordinate system (Figure 3)and the girder displacement data from GPS monitoringstation contains absolute locations in WGS-84 coordinatesystem (Figure 3) which are supposed to deduct referencelocations for analysis Until now the storage amount ofmoni-toring data has increased to 93 million for each measurementpoint Such considerable monitoring data cannot be totallyapplied to actual analysis thus the monitoring data fromupstream flank in the year 2012 are specially chosen
Taking the monitoring data in the whole August forexample considering that lateral displacement effect of maingirder is primarily reflected by wind load across the SutongBridge hence time series of wind velocity and girder dis-placement are decomposed into the119910-axis in local coordinate
system as shown in Figures 4(a) and 5(a) respectively whichcan be obviously observed that either of two time seriescontain static variant trend in whole and dynamic stochasticfluctuation in part Such two kinds of variation characteristicscan be furthermore separated by 10-minute average process asshown in Figures 4(b) 4(c) 5(b) and 5(c) respectively
By comparison between Figures 4(b) and 5(b) similarvariation characteristics exist between static wind veloc-ity and static girder displacement which can be visu-ally described by correlation scatter plots as shown inFigure 6 indicating apparent nonlinear correlation similar toquadratic parabolic curve Therefore static girder displace-ment can bemathematically expressed by static wind velocitywith consideration of definite discreteness affected by otherrandom factors Moreover time series of dynamic girderdisplacement depict obvious steady stochastic fluctuation
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 3
x
y
X (longitude orientation)
Y (latitude orientation)
z or Z
Along the bridge Across the bridge
Geographical north 106∘
106∘
Figure 3 Description of the defined local andWGS-84 coordinate system Notes (1) local coordinate system 119909-119910-119911 (2)WGS-84 coordinatesystem119883-119884-119885 (3) 90∘ of wind horizontal angle denotes the axis angle from 119909 to 119910 (within range of [0∘ 360∘]) (4) 90∘ of wind vertical angledenotes axis angle from 119910 to 119911 (within range of [minus90∘ 90∘]) (5) 106∘ axis angle exists between axes 119884 and 119910 of across the bridge
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Win
d ve
loci
ty al
ong y
-axi
s (m
s)
minus10
minus20
(a) Time series of original wind velocity
0 5 10 15 20 25 30
0
10
20
30
Time (day)
Stat
ic w
ind
velo
city
(ms
)
minus10
minus20
(b) Time series of static wind velocity
0 5 10 15 20 25 30
0
5
10
15
Time (day)
Dyn
amic
win
d ve
loci
ty (m
s)
minus10
minus10
minus15
(c) Time series of dynamic wind velocity
Figure 4 Time series of wind velocity along 119910-axis in the whole August
4 Mathematical Problems in Engineering
0 5 10 15 20 25 30
005
015
025
035
Time (day)
Gird
er d
ispla
cem
ent a
long
y-a
xis (
m)
minus005
minus015
(a) Time series of girder displacement
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Stat
ic g
irder
disp
lace
men
t (m
)
minus01
(b) Time series of static girder displacement
0 5 10 15 20 25 30
0
004
008
012
Time (day)
Dyn
amic
gird
er d
ispla
cem
ent (
m)
minus012
minus008
minus004
(c) Time series of dynamic girder displacement
Figure 5 Time series of girder displacements along 119910-axis in the whole August
5 15 250
005
01
015
02
025
03
Static wind velocity (ms)
Stat
ic g
irder
disp
lace
men
t (m
)
minus15 minus5
Figure 6 Correlation scatter plots between static wind velocity and static girder displacement
Mathematical Problems in Engineering 5
Frequency (Hz)
Day 1 to day 8Day 9 to day 16 Day 17 to day 24
Day 25 to day 31
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
Figure 7 Power spectrum densities of dynamic girder displacement in different periods
0 10 20 300
006
012
018
024
03
Static wind velocity (ms)
Fitte
d lat
eral
gird
er d
ispla
cem
ent (
m)
minus20 minus10
(a) The fitting Fourier series of nonlinear correlation term
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(b) Time series of fitting static displacement
Figure 8 The fitting Fourier series and time series of fitting static displacement
Specifically the wind data from 3D ultrasonic anemome-ters embrace such three types as wind velocity horizontalangle and vertical angle in local coordinate system (Figure 3)and the girder displacement data from GPS monitoringstation contains absolute locations in WGS-84 coordinatesystem (Figure 3) which are supposed to deduct referencelocations for analysis Until now the storage amount ofmoni-toring data has increased to 93 million for each measurementpoint Such considerable monitoring data cannot be totallyapplied to actual analysis thus the monitoring data fromupstream flank in the year 2012 are specially chosen
Taking the monitoring data in the whole August forexample considering that lateral displacement effect of maingirder is primarily reflected by wind load across the SutongBridge hence time series of wind velocity and girder dis-placement are decomposed into the119910-axis in local coordinate
system as shown in Figures 4(a) and 5(a) respectively whichcan be obviously observed that either of two time seriescontain static variant trend in whole and dynamic stochasticfluctuation in part Such two kinds of variation characteristicscan be furthermore separated by 10-minute average process asshown in Figures 4(b) 4(c) 5(b) and 5(c) respectively
By comparison between Figures 4(b) and 5(b) similarvariation characteristics exist between static wind veloc-ity and static girder displacement which can be visu-ally described by correlation scatter plots as shown inFigure 6 indicating apparent nonlinear correlation similar toquadratic parabolic curve Therefore static girder displace-ment can bemathematically expressed by static wind velocitywith consideration of definite discreteness affected by otherrandom factors Moreover time series of dynamic girderdisplacement depict obvious steady stochastic fluctuation
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Mathematical Problems in Engineering
0 5 10 15 20 25 30
005
015
025
035
Time (day)
Gird
er d
ispla
cem
ent a
long
y-a
xis (
m)
minus005
minus015
(a) Time series of girder displacement
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Stat
ic g
irder
disp
lace
men
t (m
)
minus01
(b) Time series of static girder displacement
0 5 10 15 20 25 30
0
004
008
012
Time (day)
Dyn
amic
gird
er d
ispla
cem
ent (
m)
minus012
minus008
minus004
(c) Time series of dynamic girder displacement
Figure 5 Time series of girder displacements along 119910-axis in the whole August
5 15 250
005
01
015
02
025
03
Static wind velocity (ms)
Stat
ic g
irder
disp
lace
men
t (m
)
minus15 minus5
Figure 6 Correlation scatter plots between static wind velocity and static girder displacement
Mathematical Problems in Engineering 5
Frequency (Hz)
Day 1 to day 8Day 9 to day 16 Day 17 to day 24
Day 25 to day 31
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
Figure 7 Power spectrum densities of dynamic girder displacement in different periods
0 10 20 300
006
012
018
024
03
Static wind velocity (ms)
Fitte
d lat
eral
gird
er d
ispla
cem
ent (
m)
minus20 minus10
(a) The fitting Fourier series of nonlinear correlation term
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(b) Time series of fitting static displacement
Figure 8 The fitting Fourier series and time series of fitting static displacement
Specifically the wind data from 3D ultrasonic anemome-ters embrace such three types as wind velocity horizontalangle and vertical angle in local coordinate system (Figure 3)and the girder displacement data from GPS monitoringstation contains absolute locations in WGS-84 coordinatesystem (Figure 3) which are supposed to deduct referencelocations for analysis Until now the storage amount ofmoni-toring data has increased to 93 million for each measurementpoint Such considerable monitoring data cannot be totallyapplied to actual analysis thus the monitoring data fromupstream flank in the year 2012 are specially chosen
Taking the monitoring data in the whole August forexample considering that lateral displacement effect of maingirder is primarily reflected by wind load across the SutongBridge hence time series of wind velocity and girder dis-placement are decomposed into the119910-axis in local coordinate
system as shown in Figures 4(a) and 5(a) respectively whichcan be obviously observed that either of two time seriescontain static variant trend in whole and dynamic stochasticfluctuation in part Such two kinds of variation characteristicscan be furthermore separated by 10-minute average process asshown in Figures 4(b) 4(c) 5(b) and 5(c) respectively
By comparison between Figures 4(b) and 5(b) similarvariation characteristics exist between static wind veloc-ity and static girder displacement which can be visu-ally described by correlation scatter plots as shown inFigure 6 indicating apparent nonlinear correlation similar toquadratic parabolic curve Therefore static girder displace-ment can bemathematically expressed by static wind velocitywith consideration of definite discreteness affected by otherrandom factors Moreover time series of dynamic girderdisplacement depict obvious steady stochastic fluctuation
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 5
Frequency (Hz)
Day 1 to day 8Day 9 to day 16 Day 17 to day 24
Day 25 to day 31
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
Figure 7 Power spectrum densities of dynamic girder displacement in different periods
0 10 20 300
006
012
018
024
03
Static wind velocity (ms)
Fitte
d lat
eral
gird
er d
ispla
cem
ent (
m)
minus20 minus10
(a) The fitting Fourier series of nonlinear correlation term
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(b) Time series of fitting static displacement
Figure 8 The fitting Fourier series and time series of fitting static displacement
Specifically the wind data from 3D ultrasonic anemome-ters embrace such three types as wind velocity horizontalangle and vertical angle in local coordinate system (Figure 3)and the girder displacement data from GPS monitoringstation contains absolute locations in WGS-84 coordinatesystem (Figure 3) which are supposed to deduct referencelocations for analysis Until now the storage amount ofmoni-toring data has increased to 93 million for each measurementpoint Such considerable monitoring data cannot be totallyapplied to actual analysis thus the monitoring data fromupstream flank in the year 2012 are specially chosen
Taking the monitoring data in the whole August forexample considering that lateral displacement effect of maingirder is primarily reflected by wind load across the SutongBridge hence time series of wind velocity and girder dis-placement are decomposed into the119910-axis in local coordinate
system as shown in Figures 4(a) and 5(a) respectively whichcan be obviously observed that either of two time seriescontain static variant trend in whole and dynamic stochasticfluctuation in part Such two kinds of variation characteristicscan be furthermore separated by 10-minute average process asshown in Figures 4(b) 4(c) 5(b) and 5(c) respectively
By comparison between Figures 4(b) and 5(b) similarvariation characteristics exist between static wind veloc-ity and static girder displacement which can be visu-ally described by correlation scatter plots as shown inFigure 6 indicating apparent nonlinear correlation similar toquadratic parabolic curve Therefore static girder displace-ment can bemathematically expressed by static wind velocitywith consideration of definite discreteness affected by otherrandom factors Moreover time series of dynamic girderdisplacement depict obvious steady stochastic fluctuation
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Mathematical Problems in Engineering
0 6 12 18 24 30
003
007
011
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus005
minus001
(a) Time series of original discreteness
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
minus02
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
0 10 20 30 40 50
0
02
04
06
08
1
12
Lag phase s
minus02
Part
ial c
orre
latio
n va
lue 120588
(c) The partial correlation function 120573(119904)
Figure 9 Time series of original discreteness with its 120588(119904) 120573(119904)
as well as no variation of its power spectrum densities bytime shown in Figure 7 thus dynamic girder displacementcan be mathematically simulated by harmonic superpositionmethod
3 Modeling Theory and Procedure
31 Modeling for Static Girder Displacement
311 Modeling Theory and Method Correlation scatter plotsin Figure 6 can be acquired by combination of nonlinear cor-relation term and discreteness term Furthermore nonlinear
correlation term can be mathematically modeled by the 119899-order Fourier series that is
1199061(119905) = 119886
0+
119899
sum119896=1
(119886119896cos (119896120596V (119905)) + 119887
119896sin (119896120596V (119905))) (1)
where 1199061(119905) denotes time series of fitting static displacement
for nonlinear correlation term V(119905) denotes time series ofstatic wind velocity (Figure 4(b)) and 119886
119896 119887119896(119896 = 0 1 119899)
and 120596 are fitting parameters of 119899th-order Fourier seriesWith denotation of original static girder displacement as
119906(119905) (Figure 5(b)) 119906(119905) minus 1199061(119905) acquires time series of
discreteness 1199062(119905) which contains three types of stochastic
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 7St
atic
disp
lace
men
t of d
iscre
tene
ss (m
)
0 6 12 18 24 30
001
003
005
Time (day)
minus005
minus003
minus001
(a) Time series of processed discreteness
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s 0 10 20 30 40 50
0
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588(b) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
Part
ial c
orre
latio
n va
lue 120588
minus02
minus04
(c) The partial correlation function 120573(119904)
Figure 10 Time series of processed discreteness with its 120588(119904) 120573(119904)
characteristics (autoregression moving average and het-eroscedasticity) and can be mathematically described by thecombined models of ARMA(119901 119902) and EGARCH(119898 119899) Indetail the ARMA(119901 119902) model defines the stochastic charac-teristics of autoregression and moving average as follows
1199062(119905) = 119888 +
119901
sum119895=1
1198871198951199062(119905 minus 119895) +
119902
sum119895=0
119888119895120576119905minus119895 (2)
where 119888 is the constant term 119901 and 119902 respectively denotethe orders of autoregression or moving average of 119906
2(119905) 119887119895
and 119888119895 respectively denote the coefficients of autoregression
or moving average of 1199062(119905) with 119888
0= 1 and 120576
119905minus119895denotes
the innovations process with time delay of 119895 Meanwhile theother EGARCH(119898 119899)model defines the stochastic character-istic of heteroscedasticity as follows [9 10]
log1205902119905
= 120581 +
119898
sum119896=1
119892119896log1205902119905minus119896
+
119899
sum119896=1
V119896[
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
minus 119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816]
+
119899
sum119896=1
119897119896(120576119905minus119896
120590119905minus119896
)
(3)
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Mathematical Problems in Engineering
1 2 3 4 5 6 7 8minus33615
minus33605
minus33595
minus33585
minus33575
33565
The value of m
The s
tatis
tical
val
ue o
f AIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
(a) The statistical values of AIC
The s
tatis
tical
val
ue o
f BIC
n = 1n = 2n = 3n = 4
n = 5n = 6n = 7n = 8
1 2 3 4 5 6 7 8minus33530
minus33496
minus33458
minus33422
minus33386
minus33350
The value of m
(b) The statistical values of BIC
Figure 11 The statistical values of AIC and BIC
where 120590119905denotes the conditional variance of the innovations
process 120576119905 120581 is the constant term119898 and 119899 respectively denote
the orders of the EGARCH(119898 119899)model119892119896 V119896 and 119897
119896 respec-
tively denote the coefficients of the EGARCH(119898 119899) model119911119905is a standard independent and identically distributed
random draw from some specified probability distributionsuch as Gaussian or Studentrsquos 119905 and
119864 1003816100381610038161003816119911119905minus119896
1003816100381610038161003816 = 119864(
1003816100381610038161003816120576119905minus1198961003816100381610038161003816
120590119905minus119896
)
=
radic2
120587Gaussian
radic] minus 2120587
sdotΓ ((] minus 1) 2)Γ (]2)
Studentrsquos 119905
(4)
with degrees of freedom V gt 2 Considering that theEGARCH(119898 119899) model is treated as ARMA(119901 119902) models forlog1205902119905
thus the stationarity constraint for the EGARCH(119898 119899)model is included by ensuring that the eigenvalues of thecharacteristic polynomial
120582119898
minus 1198661120582119898minus1
minus 1198662120582119898minus2
minus sdot sdot sdot minus 119866119898 (5)
are inside the unit circle where 120582 and 119866119895are the variable and
coefficients of the characteristic polynomial respectivelyDuring modeling process for time series of discreteness
1199062(119905) the autocorrelation function 120588(119904) and partial corre-
lation function 120573(119904) with their lag phase 119904 are introduced
for stationary test of 1199062(119905) Specifically the autocorrelation
function 120588(119904) can be calculated as follows [11]
120588 (119904) =(1119873)sum
119873minus119904
119905=1
(1199062(119905) minus 119906
2) (1199062(119905 + 119904) minus 119906
2)
1205902120588
119904 = 0 1 2 119873 minus 1
(6)
where
1199062=1
119873
119873
sum119905=1
1199062(119905) 120590
2
120588
=1
119873
119873
sum119905=1
(1199062(119905) minus 119906
2)2 (7)
with 119873 denoting the amount of 1199062(119905) And the other partial
correlation function 120573(119904) can be calculated through fittingsuccessive autoregressive models of orders by ordinary leastsquares retaining the last coefficient of each regression [12]Besides the AIC and BIC delimitation criteria are applied todetermine model orders with their statistical values aic andbic being respectively calculated as follows [13 14]
119886119894119888 = minus2 ln (119871) + 21198701
bic = minus2 ln (119871) + 1198701sdot ln (119870
2)
(8)
where 119871 denotes the optimized log-likelihood objectivefunction (LLF) values associated with parameter estimates ofthe combined models 119870
1denotes the number of estimated
parameters associated with each value in LLF and1198702denotes
the sample size of the observed 1199062(119905) associatedwith each LLF
value
312 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeling
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 9
0
0
10 20 30 40 50
Autocorrelation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
02
04
06
08
1
minus02
minus04
Auto
corr
elat
ion
valu
e120588
(a) The autocorrelation function 120588(119904)
0 10 20 30 40 50
0
02
04
06
08
1
Partial correlation valueUpper limit with 95 confidence intervalsLower limit with 95 confidence intervals
Lag phase s
minus02
minus04
Part
ial c
orre
latio
n va
lue 120588
(b) The partial correlation function 120573(119904)
Figure 12 The correlation functions 120588(119904) and 120573(119904) of residuals
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
r times10
minus3
(a) The standard errors when119898 = 1 119899 = 1
1 2 3 4 5 6 7590
591
592
593
594
595
596
The value of order p
q = 1q = 2
q = 3q = 4
The s
tand
ard
erro
rtimes10
minus3
(b) The standard errors when119898 = 2 119899 = 1
Figure 13 The standard errors between processed discreteness and its models with lower orders
time series of static girder displacement is illustrated takingthe correlation scatter plots in the whole August in Figure 6for example as follows
Step 1 Fitting Fourier series for nonlinear correlation termBy means of the MATLAB fitting tools (utilizing the third-order Fourier series (1) to fit the correlation scatter plots)[10] the mathematical model of nonlinear correlation termis straightforward and acquired as shown in Figure 8(a)together with the estimated values of Fourier parameterspresented in Table 1 By substitution of estimated valuestogether with V(119905) into formula (1) time series of fitting static
displacement1199061(119905) in thewholeAugust are acquired as shown
in Figure 8(b)
Step 2 Stationary test for time series of discreteness Timeseries of discreteness 119906
2(119905) can be acquired by 119906(119905) minus
1199061(119905) as shown in Figure 9(a) Autocorrelation function 120588(119904)
and partial correlation function 120573(119904) of 1199062(119905) are calcu-
lated with 50 lag phases shown in Figures 9(b) and 9(c)respectively presenting that 120588(119904) is slowly converging intothe 95 confidence intervals as the lag phase 119904 increaseswhich indicates bad stationarity of 119906
2(119905) for mathematical
modeling Due to this process of first-order difference for
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Mathematical Problems in Engineering
0 6 12 18 24 30
minus001
001
003
005
Time (day)
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
minus003
minus005
(a) Time series of simulative processed discreteness
minus001
Stat
ic d
ispla
cem
ent o
f disc
rete
ness
(m)
0 6 12 18 24 30
003
007
011
Time (day)
minus005
(b) Time series of simulative original discreteness 1199062(119905)
0 5 10 15 20 25 30
0
01
02
03
Time (day)
Fitti
ng st
atic
disp
lace
men
t (m
)
minus01
(c) Time series of simulative static displacement 119906(119905)
Figure 14 Simulation for time series of discreteness 1199062
(119905) and static displacement 119906(119905)
Table 1 Estimated values of fitting parameters (minus12 lt V(119905) lt 21)
Fitting parameters 1198860
1198861
1198871
1198862
1198872
1198863
1198873
119908
Estimated values 01102 minus01026 00218 001214 0006845 minus000618 minus0003763 01374
1199062(119905) is carried out as shown in Figure 10(a) with its 120588(119904) and
120573(119904) shown in Figures 10(b) and 10(c) both presenting rapidconvergence into the 95 confidence intervals and verifyinggood stationarity for processed discreteness
Step 3 Order determination of ARMA(119901 119902) andEGARCH(119898 119899) The orders of 119901 and 119902 are relative tothe convergent forms of 120588(119904) and 120573(119904) That is the 120588(119904)
and 120573(119904) clearly present the trailing property in Figures10(b) and 10(c) with 4th and 7th of lag phase 119904 initiallyconverging into the 95 confidence intervals inferring that7 and 4 are appropriately assigned to the orders of 119901 and 119902respectively [11 12] Furthermore the orders of 119898 and 119899 aredetermined utilizing the AIC and BIC delimitation criterion
In detail the statistical values of AIC and BIC from timeseries of processed discreteness (Figure 10(a)) are calculatedrespectively under integer assignment of119898 and 119899 between 0and 8 as shown in Figure 11 presenting that the most suitableorders of 119898 and 119899 are 2 and 1 respectively corresponding tothe minimum statistical values [13 14]
Step 4 Parameter estimation and residual test of orderedmodels Based on the specified orders above model param-eters (119887
119895 119888119895 119892119896 V119896 119888 and 120581) are further estimated to fit
time series of processed discreteness as shown in Tables2 and 3 respectively For testing the fitting effectiveness ofestimated parameters residuals between processed discrete-ness (Figure 10(a)) and its defined models with estimated
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 11
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus5
10minus6
10minus610minus7
(a) Original power spectrum density
August 1 to August 8
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Processed power spectrum density
Figure 15 Original and processed power spectrum densities
Table 2 Estimating model parameters of the ARMA(7 4) model
The combined models The ARMA(7 4) modelThe fitting parameters 119888 119887
1
1198872
1198873
1198874
1198875
1198876
1198877
1198881
1198882
1198883
1198884
The values 000 minus049 094 075 minus032 minus001 004 002 006 minus126 minus048 068
Table 3 Estimatingmodel parameters of the EGARCH(2 1) model
The combined models The EGARCH(2 1) modelThe fitting parameters 120581 119892
1
1198922
V1
1198971
The values minus076 027 065 028 006
parameters are analyzed using 120588(119904) and 120573(119904) as shownin Figure 12 which presents that both 120588(119904) and 120573(119904) areconsistent in the 95 confidence intervals (except 119904 is 0)and thus verifies good availability of estimated parametersfor ordered models Moreover the standard errors betweenprocessed discreteness (Figure 10(a)) and its models withlower orders are shown in Figure 13
Step 5 Simulation for time series of discreteness 1199062(119905)
and static displacement 119906(119905) Based on the mathematicalmodels of ARMA(7 4) and EGARCH(2 1) with estimatedparameters time series of simulative processed discretenessare shown in Figure 14(a) and through inverse calculationof first-order difference time series of simulative originaldiscreteness 119906
2(119905) are obtained as shown in Figure 14(b)
Together with time series of fitting static displacement 1199061(119905)
in Figure 8(b) time series of simulative static displacement119906(119905) are ultimately shown in Figure 14(c) definitely similar totime series of monitoring static displacement in Figure 5(b)
32 Modeling for Dynamic Girder Displacement
321 Modeling Theory and Method Time series of dynamicgirder displacement show consistent power spectrum den-sity (Figure 7) which can be mathematically simulated byharmonic superposition method [15ndash17] Primarily functionexpression of power spectrum density should be confirmedto lay foundation for harmonic superposition Consideringthat straightforward function fitting will ignore local featherof acute peak at 10minus1Hz around of frequency thus the powerspectrum density are decomposed into two parts part one tofit the whole trend with ignorance of acute peak part two tospecifically fit the acute peak Furthermore each part can beexpressed by the 119899-order Gaussian series in logarithmic formthat is
lg1199011(119891) =
119899
sum119896=1
119886119896119890minus((lg119891minus119887119896)119888119896)
2
lg1199012(119891) =
119899
sum119896=1
119889119896119890minus((lg119891minus119903119896)ℎ119896)
2
(9)
where 1199011(119891) denotes function expression of part one 119901
2(119891)
denotes function expression of part two and 119886119896 119887119896 119888119896 119889119896 ℎ119896
and 119903119896
are fitting parameters of 1199011(119891) and 119901
2(119891)
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Mathematical Problems in Engineering
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p1(f)
(a) The first part and its fitting curve 1199011(119891)
Monitoring curve
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
102
100
101
10minus110minus210minus310minus410minus5100
Fitting curve p2(f)
(b) The second part and its fitting curve 1199012(119891)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
Fitting curve p(f)
(c) The combination 119901(119891) of 1199011(119891) and 119901
2(119891)
Figure 16 Two parts of processed power spectrum densities and their fitting curves
119896 = 0 1 119899 Accordingly the function expression ofpower spectrum density 119901(119891) can be formulized as follows
119901 (119891) = 1199011(119891) + 119901
2(119891) (10)
Based on power spectrum density119901(119891) above variance 1205902within frequency band of [119891
11198912] can be calculated utilizing
the expanding characteristics of average power spectrum forstationary and stochastic process as follows
1205902
= int1198912
1198911
119901 (119891) 119889119891 (11)
Then the frequency band of [11989111198912] is divided into 119899 subin-
tervals and the whole values 119901(119891) in the 119894th subintervalcan be represented by the corresponding value 119901(119891mid119894) at
the middle frequency 119891mid119894 thus formula (11) is furthermorediscretized as follows
1205902
asymp
119899
sum119894=1
119901 (119891mid119894) sdot Δ119891 (12)
where
Δ119891 =1198912minus 1198911
119899 (13)
As for the 119894th subinterval its sinusoidal function 119908(119905 119894)
containing middle frequency 119891mid119894 and standard deviationradic119901(119891mid119894) sdot Δ119891 can be expressed as follows
119908 (119905 119894) = radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (14)
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 13
0 14400 28800 43200 57600 72000 86400
0
002
004
Time (s)
Sim
ulat
ive d
ynam
ic d
ispla
cem
ent (
m)
minus004
minus002
(a) Time series of simulative dynamic displacement 119908(119905)
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
100
100
10minus1
10minus1
10minus2
10minus2
10minus3
10minus3
10minus4
10minus4
10minus5
10minus6
10minus510minus7
(b) Power spectrum density of 119908(119905)
Figure 17 Time series of simulative dynamic displacement 119908(119905) and its power spectrum density
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
minus005
minus01
(a) Time series of simulative lateral displacement
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
0
005
01
015
02
Time (day)
Mon
itorin
g lat
eral
disp
lace
men
t (m
)
minus005
minus01
(b) Time series of monitoring lateral displacement
Figure 18 Time series of simulative and monitoring lateral girder displacement
where 120579119894is the random variable of uniform distribution
within [0 2120587] With superposition of119908(119905 119894) from each subin-terval the time series of dynamic girder displacement 119908(119905)can be mathematically modeled by harmonic superpositionfunction as follows
119908 (119905) =
119899
sum119894=1
radic2119901 (119891mid119894) sdot Δ119891 sin (2120587119891mid119894119905 + 120579119894) (15)
322 Detailed Procedure Based on the theory and methodabove the detailed procedure for mathematically modeledtime series of dynamic girder displacement is illustrated
taking the correlation scatter plots from August 1 to August8 in Figure 15(a) for example as follows
Step 1 Decreasing discreteness for original power spectrumdensity Considering that certain discreteness existing inoriginal power spectrum density can cover the acute peakat 10minus1Hz adverse to fitting the 119899-order Gaussian seriestherefore average process with double frequency scales iscarried out to decrease discreteness specifically by 10minus3Hzaverage process within frequency band [10minus15Hz 10minus05Hz]and 10minus2Hz average process within other frequency bandsProcess result is shown in Figure 15(b) showing distinct acutepeak for fitting the 119899-order Gaussian series
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
14 Mathematical Problems in Engineering
0
005
01
015
02
Sim
ulat
ive l
ater
al d
ispla
cem
ent (
m)
Simulative lateral displacement Monitoring lateral displacement
minus005
minus01
Sept
embe
r 1
Sept
embe
r 2
Sept
embe
r 3
Sept
embe
r 4
Time (day)
(a) Simulative and monitoring results
0 005 01 015
0
005
01
015
Simulative lateral displacement (m)M
onito
ring
later
al d
ispla
cem
ent (
m)
Scatter plotsFitting curve
minus005minus005
ds(t) = 1008dm(t) + 00016
(b) Correlation scatter plots with fitting curve
Figure 19 Simulative and monitoring results and their correlation scatter plots with fitting curve
Simulative resultMonitoring result
Frequency (Hz)
Pow
er sp
ectr
um d
ensit
y (m
2s)
10minus1
10minus110minus2
10minus3
10minus3
10minus5
10minus410minus9
10minus7
Figure 20 Power spectrum densities of simulative and monitoring results
Step 2 Fitting Gaussian series for two parts of processedpower spectrum density The processed power spectrumdensity can be divided into two parts the whole trendwith ignorance of acute peak as shown in Figure 16(a) thespecific acute peak as shown in Figure 16(b) By means of theMATLAB fitting tools (utilizing the fourth-order Gaussianseries (9) to fit two parts) [10] themathematicalmodels119901
1(119891)
and 1199012(119891) of two parts are respectively shown in Figures
15(a) and 15(b) together with estimated parameters presentedin Table 4The combination119901(119891) of119901
1(119891) and119901
2(119891) is shown
in Figure 16(c)
Step 3 Harmonic superposition for fitting power spectrumdensity 119901(119891) sdot 119901(119891) is divided into 25000 subintervals withinfrequency bands [10minus5Hz 10minus05Hz] In each subintervalone middle frequency 119891mid119894 and its corresponding value119901(119891mid119894) are existent and then substituted into (15) forsummation thus time series of dynamic girder displacement119908(119905) can be acquired as shown in Figure 17(a) (simulated for86400 s) By comparing its power spectrum density shown inFigure 17(b) with the monitoring one shown in Figure 15(a)good similarity of the whole trend and the acute peak verifies
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 15
Table 4 Estimated parameter values of the fourth-order Gaussian series
Fitting curves Estimated parameter values of the fourth-order Gaussian Series1198861
1198862
1198863
1198864
1198871
1198872
1198873
1198874
1198881
1198882
1198883
1198884
1199011
(119891) minus536 045 minus486 minus073 minus498 minus298 minus075 minus022 169 038 173 0361199012
(119891) 058 085 014 070 minus096 minus096 minus120 minus110 001 005 004 015
appropriateness of mathematical models for dynamic girderdisplacement
4 Model Test and Evaluation
According to mathematical modeling process above timeseries of lateral girder displacement are expressed by com-bination of third-order Fourier series 119906
1(119905) ARMA(74)
EGARCH(2 1) and harmonic superposition function 119908(119905)For verifying feasibility and effectiveness of the whole math-ematical models time series from September 1 to September3 are simulated with intervention of monitoring static wind(Figure 18(a)) and then compared with monitoring onesduring same period (Figure 18(b))
According to mathematical modeling theory and proce-dure above comparison is divided into two parts (1) timeseries of static girder displacement (2) time series of dynamicgirder displacement As for the first part its simulative andmonitoring results are shown in Figure 19(a) and linear fittingcurve of correlation scatter plots is shown in Figure 19(b)presenting consistent variation tendency in Figure 19(a) and119889119904(119905) approximating to 119889
119898(119905) in Figure 19(b) (where 119889
119904(119905)
and 119889119898(119905) respectively denote simulative and monitoring
results) which verifies good feasibility and effectiveness ofmathematical models for static girder displacement As forthe second part power spectrum densities of simulativeand monitoring results are shown in Figure 20 present-ing uniform variation tendency of both whole trends andacute peaks which verifies good feasibility and effectivenessof mathematical models for dynamic girder displacementTherefore mathematical models above can be reasonablyutilized to simulate time series of lateral girder displacementfrom main girder of Sutong Cable-Stayed Bridge
5 Conclusions
Based onmonitoring data frommain girder of Sutong Cable-Stayed Bridge time series of lateral girder displacement effectare mathematically modeled by methods of fitting Fourierseries and Gaussian series combined models of ARMA(74)and EGARCH(21) and harmonic superposition functionAnd conclusions can be drawn as follows
(1) Scatter plots between lateral static wind velocity andlateral static displacement present apparent nonlinearcorrelation which is similar to quadratic paraboliccurve and time series of lateral dynamic displacementcontain obvious stable power spectrum density withno variation by time
(2) Time series of lateral static displacement can bedecomposed into nonlinear correlation term and
discreteness term Moreover nonlinear correlationterm can be mathematically modeled by third-orderFourier series with intervention of lateral static windvelocity and discreteness term can bemathematicallymodeled by the combinedmodels of ARMA(7 4) andEGARCH(2 1)
(3) Through decreasing discreteness in double frequencyscales and division in double frequency bands powerspectrum density of lateral dynamic displacementcan be mathematically modeled by the fourth-orderGaussian series and time series of lateral dynamicdisplacement can be further mathematically modeledby harmonic superposition function
(4) By the comparison between simulative and moni-toring lateral displacement effect from September 1to September 3 mathematical models are feasibleand effective to simulate time series of lateral girderdisplacement from main girder of Sutong Cable-Stayed Bridge
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The authors gratefully acknowledge the National Scienceand Technology Support Program (no 2014BAG07B01) theNational Natural Science Foundation (no 51178100) and KeyProgram of Ministry of Transport (no 2011318223190 and no2013319223120)
References
[1] F Xu A Chen and Z Zhang ldquoAerostatic wind effects on theSutong Bridgerdquo in Proceedings of the 3rd International Confer-ence on Intelligent System Design and Engineering Applications(ISDEA 2013) pp 247ndash256 January 2013
[2] M Gu and H F Xiang ldquoAnalysis of buffeting response and itscontrol of Yangpu Bridgerdquo Journal of Tongji University vol 21no 3 pp 307ndash314 1993 (Chinese)
[3] D Green andW G Unruh ldquoThe failure of the Tacoma bridge aphysical modelrdquo American Journal of Physics vol 74 no 8 pp706ndash716 2006
[4] J Cheng and R-C Xiao ldquoA simplified method for lateralresponse analysis of suspension bridges under wind loadsrdquoCommunications in Numerical Methods in Engineering vol 22no 8 pp 861ndash874 2006
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
16 Mathematical Problems in Engineering
[5] X-H Long L Li and LHu ldquoTime domain analysis of buffetingresponses of sidu river suspension bridgerdquoEngineeringMechan-ics vol 27 no 1 pp 113ndash117 2010 (Chinese)
[6] M Yu H-L Liao M-S Li C-M Ma and M Liu ldquoFieldmeasurement and wind tunnel test of buffeting response oflong-span bridge under skew windrdquo Journal of Experiments inFluid Mechanics vol 27 no 3 pp 51ndash55 2013 (Chinese)
[7] H Wang A Li J Niu Z Zong and J Li ldquoLong-term moni-toring of wind characteristics at Sutong Bridge siterdquo Journal ofWind Engineering and Industrial Aerodynamics vol 115 pp 39ndash47 2013
[8] F T Lombardo J AMain and E Simiu ldquoAutomated extractionand classification of thunderstorm and non-thunderstormwinddata for extreme-value analysisrdquo Journal of Wind Engineeringand Industrial Aerodynamics vol 97 no 3-4 pp 120ndash131 2009
[9] H Wang A-Q Li C-K Jiao and X-P Li ldquoCharacteristics ofstrong winds at the Runyang Suspension Bridge based on fieldtests from 2005 to 2008rdquo Journal of Zhejiang University ScienceA vol 11 no 7 pp 465ndash476 2010
[10] MATLAB MATLAB online manual Natick Mass USA 2010[11] E Castano and J Martınez ldquoUse of the crosscorrelation func-
tion in the identification ofARMAmodelsrdquoRevistaColombianade Estadistica vol 31 no 2 pp 293ndash310 2008
[12] T I Lin and H J Ho ldquoA simplified approach to invertingthe autocovariance matrix of a general ARMA(p q) processrdquoStatistics and Probability Letters vol 78 no 1 pp 36ndash41 2008
[13] B Clarke J Clarke and C W Yu ldquoStatistical problem classesand their links to information theoryrdquoEconometric Reviews vol33 no 1ndash4 pp 337ndash371 2014
[14] J C Escanciano I N Lobato and L Zhu ldquoAutomatic spec-ification testing for vector autoregressions and multivariatenonlinear time series modelsrdquo Journal of Business amp EconomicStatistics vol 31 no 4 pp 426ndash437 2013
[15] J A Munoz J R Espinoza and C R Baier ldquoDecoupled andmodular harmonic compensation formultilevel statcomsrdquo IEEETransactions on Industrial Electronics vol 61 no 6 pp 2743ndash2753 2014
[16] S Yan and W Zheng ldquoWind load simulation by superpositionof harmonicrdquo Journal of Shenyang Architectural and CivilEngineering Institute vol 21 no 1 pp 1ndash4 2005 (Chinese)
[17] C-X Li and C-Z Liu ldquoRBF-neural-network-based harmonysuperposition methodrdquo Journal of Vibration and Shock vol 29no 1 pp 112ndash116 2010
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of