Upload
others
View
7
Download
1
Embed Size (px)
Citation preview
PhD thesis defense Thursday, May 18, 2017
Energy management under uncertainty: Application to the day-ahead planning and power reserve allocation of an
urban microgrid with active photovoltaic (PV) generators and storages
Xingyu YAN
L2EP, Centrale Lille, FranceSupervisors: Bruno FRANCOIS & Dhaker ABBES
2
Wind and photovoltaic (PV) power production are highlyintermittent due to the influence of meteorological conditionson the primary energy resource.
I. Background: Problem, Goals and Methods
I.1 Problem: Variability of RES
Accurate forecasts of renewable energy generation are useful for Producers: Making bids in electricity markets, Planning maintenance of wind farms, etc.; Grid operators: Economic dispatch, planning power reserve, planning power exchanges
with interconnections and other stakeholders, etc.
Operate the power system in a secure and economic way.
0 100 200 300 400 500 600 700 800 9000
10
20
PV P
ower
(kW
)
T ime (minutes)
22/06/2010-24/06/2010
RES: renewable energy sources
3
Intermittent RES energy production is predictable thanks to weather and statistical tools.
I.1 Problem: Prediction Errors
However, forecasting errors can not be eliminated even with the best forecasting tools.
I. Background: Problem, Goals and Methods
Uncertainty caused by forecasting errors
4I.1 Problem: Power Reserve to Cover the Uncertainty
PrefPref
Pref is given by the system operators (power dispatch)
Today the consumption/production balancing and OR provision are performed by conventional generators.
UncertainVariable Reserve
Constant Reserve
To cover the risk of an unexpected RES generation losses or load increasing, operatingreserve (OR) is scheduled one day ahead.
Massive RES generators increase the system uncertainty of power production and so thedifficulties to maintain the system security level.
To cover the risk: Additional OR is needed !
How much ? How to provide it ?
I. Background: Problem, Goals and Methods
5I.2 Applied method
Active PV Generators (PV AGs): with an additional storage system (batteries and super-capacitors); MGTs: Micro gas-turbines.
I. Background: Problem, Goals and Methods
Microgrid: From a centralized network to a decentralized network.
Distributed generators (DG) MGTs: constant power sources, with
controlled power output; PV AGs: can provide ancillary services
thanks to storage devices.
Uncertainty challenges of RES integration into the Microgrid ?
6I.3 General Organization of Energy Management System (EMS)
Microgrid supervision can be analyzed in different timing scales and functions.
I. Background: Problem, Goals and Methods
Focuses on day-ahead optimization of the network operation.
7I.4 Goals and Methods for the Study Case
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimization Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
I. Background: Problem, Goals and Methods
Active PV Generators (PV AGs): with an additional storage system (batteries and super-capacitors); MGTs: Micro gas-turbines.
8Roadmap: Part II
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimization Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
9II.1 Data Management and Predictive Analysis (1)II. Uncertainty Analysis and Forecasting of PV Power and Load
Data collecting for database building
PV data collecting: (Sunways) three PV inverters (3 kW each), Centrale de Lille
Load data collecting (www.rte‐france.com)
Meteorological data collecting (www.wunderground.com)
010203040506070
05
10152025
Prec
ipita
tion
(mm
)
Tem
pera
ture
(°C
)
Month
Data from 2000 to 2012Average precipitation Average high in °C Average low in °C
10II. Operating Reserve (OR) Quantification to Cover Uncertainty
Data mining and predictive analysis (PV power)
II.1 Data Management and Predictive Analysis (2)
Mathematical Modeling of PV Generator
) )25)(( 0.0035 - 1 ( )( )( =)( tTtIAttPV rpvpvP
0 100 200 300 400 500 600 700 800 9000
10
20
PV P
ower
(kW
)
22/06/2010-24/06/2010
0 100 200 300 400 500 600 700 800 9000
0.4
0.8
1.2
Irra
diat
ion
(kW
/m2 )
0 100 200 300 400 500 600 700 800 9000
20
40
Time (every 5 minutes)
Tem
pera
ture
(°C
)
R2 is used to determine how closely the irradiance and the temperature fit PV power.
11II. Uncertainty Analysis and Forecasting of PV Power and Load
PV power and irradiance in 2010
0100200300400500600700800900
1000
Sola
r Rad
iatio
n (W
/m2 )
Time (hour)
Highly Variable-01/05
Overcasted-12/05
Low Variable-15/05
Clear Sky-23/05
II.2 PV Power variability quantification
020406080100120140160
0
500
1000
1500
2000
2500
Irrid
ianc
e (k
Wh/
m²)
PV P
ower
(kW
h)
Month
Irridiation (kWh/m²) PV Power (kWh)
Variable Index (VI) and Radiation Index (RI)
SC
N
k kkk
I
SRSRSRNVI 2
22122
1
SC
N
k k
ISR
RI 2
121
VI: calculates PV power variabilityRI: calculates daily solar radiation
: solar radiation during 5 minutes scale
12
Forecasting with Artificial Neural Networks (ANN) PV forecast database
PV power data
Meteorological dataSolar irradiation Humidity Air pressure Wind speed Temperature
PV power output AC voltage DC voltage
Dispatch data into 3 sets
Validation set
Test set
Training set
DATA BASE
Load forecast databaseLoad demand data
Meteorological dataSolar irradiation Humidity Air pressure Wind speed Temperature
Dispatch data into 3 sets
Validation set
Test set
Training set Historical load data
DATA BASE
Input layer Hidden layer Output layer
X ... ...
)1(bias
)1( )2(
. . . y
)2(bias
II.3 PV Power and Load Forecasting with ANN (1)
A three-layer ANN
Data description: PV power and Load forecasting databases
Error Computing Method
Data normalization
minmax
min:xx
xxxxf i
n
i kk yyn
nRMSE1
2)~(1 =
n
i kk yyn
nMAE1
~1 =
nRMSE: normalized Root Mean Square Error nMAE: the normalized Mean Absolute Error
II. Uncertainty Analysis and Forecasting of PV Power and Load
13II.3 PV Power and Load Forecasting with ANN (2)
PV Forecast
24~..., ,~
hh vPvP
ANN
124 ..., , hh PvPv
Last 24h sensed PV Power
Last 24h sensed irradiance
124~..., ,~
hh II 24~..., ,~hh TT
Day ahead Predicted
Temperature
D+1 D
124~..., ,~
hh vPvP
PV Power Forecasting with ANN
Input layer Hidden layer Output layer
X . . . . . .
)1(bias
)1( )2(
... y
)2(bias
A three-layer ANN
0 5 10 15 20 250
0.5
1
Time (hours)
Inpu
t Dat
a (p
.u.)
24h-before PV Power24h-before Irradiance24h-ahead Temperature
0 5 10 15 20 250
0.5
1
Time (hours)
Out
put D
ata (
p.u.
)
Real Measured PV Power24h-ahead Predicted PV Power
nRMSE [%] nMAE [%]Training Set 6.09 3.69
Validation Set 5.58 3.13Test Set 5.95 3.12
Errors of the PV power forecasting with ANN.
0 4 8 12 16 20 24-0.4
-0.2
0
0.2
0.4
Time (hours)
Erro
r (p.
u.)
II. Uncertainty Analysis and Forecasting of PV Power and Load
14
Load Forecast
24~..., ,~
hh LL
ANN
148 ..., , hh LL
Last 48h Sensed Load
D+1 D
24~..., ,~hh TT
24h ahead Predicted Temperature
124~..., ,~
hh LL
Load Forecasting with ANN
II.3 PV Power and Load Forecasting with ANN (3)
0 10 20 30 40 50
0.4
0.6
0.8
Inpu
t Dat
a (p
.u.)
48h-before Load
0 5 10 15 20 250
0.5
1
24h-ahead Temperature
0 5 10 15 20 250.2
0.4
0.6
Time (hours)
Out
put D
ata (
p.u.
)
Real LoadPredicted Load
Input layer Hidden layer Output layer
X . . . . . .
)1(bias
)1( )2(
... y
)2(bias
A three-layer ANN
nRMSE [%] nMAE [%]Training Set 3.18 2. 45
Validation Set 3.57 2.76Test Set 3.67 2.84
Errors of the Load forecasting with ANN.
0 4 8 12 16 20 24-0.3
-0.2
-0.1
0
0.1
0.2
0.3
Time (hours)
Erro
r (p.
u.)
II. Uncertainty Analysis and Forecasting of PV Power and Load
15
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimization Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
Roadmap: Part III
16III. Operating Reserve (OR) Quantification to Cover Uncertainty
Frequency control and energy balancing
t
Reserve
Automatic Automatic/Manually
Manually
fnom
Primary Secondary
Tertiaire
30 s 15 min 30 min
Band of the normal frequency variation
Detection of the problem
Deployment of the reserve
Stabilization Recovery Maintain
f
Frequency
In this thesis, the OR is defined as the real power that can be called on instantaneously for the imbalance between power generation and load demand (Primary and secondary reserve).
III.1 Calculation of OR by considering RES uncertainties (1)
Deploying of the primary, secondary and tertiary regulation of frequency.
17III.1 Calculation of OR by considering RES uncertainties (2)III. Operating Reserve (OR) Quantification to Cover Uncertainty
The OR must be precisely quantified in order to maintain a specified level of system security with a minimum cost.
Most conventional utilities have adopted deterministic criteria by considering only two sources of uncertainty:
the possibility of multiple large generators failing: low probability but high impact; and load forecast errors: often but usually relatively small.
Deterministic methods do not match the stochastic nature of the OR quantification problem.
Probabilistic methods are adapted to the stochastic characteristics of RES based generators and loads. They can set a certain security level.
18III. Operating Reserve (OR) Quantification to Cover Uncertainty
Recent examples of probabilistic methods:
Probabilistic Model
Fixed Risk Indices
Uncertainty Analysis
Load Forecasting Uncertainty
PV Power Forecasting Uncertainty
Net Demand (ND) Uncertainty t
FtF
tF PVLND
Risk/Reserve Curve
Operating Reserve Quantification
Eastern Wind Integration and Transmission Study (EWITS): focused on the operational impacts of various wind penetrations. Load variability and wind power variability are considered independent [1].
Western Wind and Solar Integration Study (WWSIS): discussed the OR requirement that dynamically relied on both the load and wind penetration levels [1].
Other probabilistic methods
General Scheme for OR quantification
Decisions made under uncertainty must be informed by probabilistic information in order to correctly quantify the risk.
III.1 Calculation of OR by considering RES uncertainties (3)
[1]: L. E. Jones, Renewable energy integration: practical management of variability, uncertainty, and flexibility in power grids: Academic Press, 2014.
19III.2 Net Demand (ND) Uncertainty Analysis (1)
124~..., ,~
hh LL 124 ..., , hh PvPv
D+1 D
D+1D
+ _
+
Last 24h ND Last 24h ND forecast
124~..., ,~
hh DNDN
+ _
124 ..., , hh LL
ND Error Forecast ANN
NDh
NDh 124 ..., ,
Last 24h prediction errors
124~..., ,~
hh vPvP
Sensed Load Database
D+1 D
Sensed PV Database
D+1 D
Forecasted PV Database
Forecasted Load Database
_
NDh
NDh 241
~..., ,~
NDh
NDh
Mean average and Standard deviation
First Method: Day-ahead Net Demand Errors Forecast The real ND is composed of the forecasted ND and an error: ND
hhh DNND ~ =
NDhNDh
: Mean value: Standard deviation
III. Operating Reserve (OR) Quantification to Cover Uncertainty
NDh : ND forecasting errors
-10 0 100
10
20
30
40
Errors (% p.u.)C
ount
Frequency histFitted Gaussian
(a)Net Forecast Errors
20III.2 Net Demand (ND) Uncertainty Analysis (2)
Second Method: Calculation from the PV Power and the Load Forecast Errors Estimation.
Prediction Error Forecast
ANN
Lh
Lh 124 ..., ,
Lh
Lh 24
~..., ,~
Prediction Error Forecast
ANN
PVh
PVh 124 ..., ,
PVh
PVh 24
~..., ,~
Last 24h prediction errors
2)(, NDh
NDh
2)(, PVh
PVh 2)(, L
hLh
Mean average and Standard deviation
Equations
Mean average and Standard deviation
124~..., ,~
hh LL
124 ..., , hh PvPv
D+1 D
D+1D
+ + _
124 ..., , hh LL 124
~..., ,~ hh vPvP
Sensed Load Database
D+1 D
Sensed PV Database
D+1 D
Forecasted PV Database
Forecasted Load Database
_
PVh
Lh
NDh
222 )()()( PVh
Lh
NDh
The real ND is composed of the forecasted ND and an error: NDhhh DNND ~ =
III. Operating Reserve (OR) Quantification to Cover Uncertainty
NDhNDh
: Mean value: Standard deviation
NDh : ND forecasting errors
-40 0 400
15
30
45
Errors (% p.u.)
Coun
t
(a) PV Forecast Errors
0 -20 0 200
15
30
45
Errors (% p.u.)
Coun
t
(b) Load Forecast Errors
Frequency histFitted Gaussian
21III.3 Uncertainty Assessment
Lower bound (Blower)
Normal Inverse Cumulative Distribution
Function (F-1)
Normal Probability
Density Function (F(B|µND,σND))
Upper bound (Bup)
+ +
_ +
NDh
NDh
Net Demand Forecast
241~..., ,~
hh DNDN max_
~hDN
min_~
hDN
x: Desired probability
NDhpdf
deBxB
NDh
NDh
NDh
NDh
NDh
2
2
)(2)(
21),(F
xBFBxB NDh
NDh
NDh
NDh ),(:),(F 1-
Hourly ND uncertainty assessment
0 4 8 12 16 20 240
5
10
15
20
Time [hours]
PV u
ncer
tain
ty [k
W]
90% probability
80% probability
70% probability
60% probability
Forecasted PV
0 4 8 12 16 20 240
20
40
60
80
100
120
Time [hours]
Loa
d D
eman
d U
ncer
tain
ty [k
W]
90% probability80% probability70% probability60% probabilityForecasted Load
0 4 8 12 16 20 240
20
40
60
80
100
Time [hours]
Net
Dem
and
Unc
erta
inty
[kW
]
90% probability80% probability70% probability60% probabilityNet Forecasted
III. Operating Reserve (OR) Quantification to Cover Uncertainty
Bound margin:
Probability:
22III.3 Uncertainty Assessment
Lower bound (Blower)
Normal Inverse Cumulative Distribution
Function (F-1)
Normal Probability
Density Function (F(B|µND,σND))
Upper bound (Bup)
+ +
_ +
NDh
NDh
Net Demand Forecast
241~..., ,~
hh DNDN max_
~hDN
min_~
hDN
x: Desired probability
NDhpdf
deBxB
NDh
NDh
NDh
NDh
NDh
2
2
)(2)(
21),(F
xBFBxB NDh
NDh
NDh
NDh ),(:),(F 1-
Hourly ND uncertainty assessment
0 4 8 12 16 20 240
5
10
15
20
Time [hours]
PV u
ncer
tain
ty [k
W]
90% probability
80% probability
70% probability
60% probability
Forecasted PV
0 4 8 12 16 20 240
20
40
60
80
100
120
Time [hours]
Loa
d D
eman
d U
ncer
tain
ty [k
W]
90% probability80% probability70% probability60% probabilityForecasted Load
0 4 8 12 16 20 240
20
40
60
80
100
Time [hours]
Net
Dem
and
Unc
erta
inty
[kW
]
90% probability80% probability70% probability60% probabilityNet Forecasted
III. Operating Reserve (OR) Quantification to Cover Uncertainty
Bound margin:
Probability:
High uncertainty
23III.4 OR Quantification Power reserve quantification
PRhhh pdfPLprobLOLP )d()0( =LOLP represents the probability that load exceeds PV power.
III. Operating Reserve (OR) Quantification to Cover Uncertainty
1
PV power forecast errors database
tPVcdf CDF CDF CDF
CDF CDF CDF
CDF CDF CDF
tLcdf
tNDcdf
2 T …
Load forecast errors database
Net load forecast errors database
…
…
…
0 0
0.02
0.04
0.06
0.08
0.1
Prob
abili
ty
Accepted risk
LOLP= x %
x %
RPower DeficitPower Surplus
0 1 2 3 4 5 6 7 8 9 10
0510
1520
250
5
10
15
x % of LOLPTime [Hour]
Pow
er R
eser
ve [k
W]
0 4 8 12 16 20 24 0
6
12
18Hourly power reserve with 1 % LOLP
Time (Hour)
Hou
rly P
ower
Res
erve
(kW
h)
First Method Second Method
CDF: cumulative distribution function
24
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimization Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
Roadmap : Part IV
25IV.1 Optimal OR Dispatching Strategies
8h 10h 13h 16h 19h 22h 1h 4h 7h0
30
60
Fore
cast
PV (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
60
120
Fore
cast
Loa
d (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
10
20
Time (half hour)
PR (k
W)
With 1 % of LOLP
Forecasted PV, load and OR quantification
IV. Optimal Operating Reserve (OR) Dispatching Strategies
Day-ahead PV power forecasting Day-ahead load forecasting Statistic OR quantification
One Day-ahead OR dispatching
Collected data for system
uncertainty analysis
PV power and load forecasting
OR quantification
Control System of the nth PVAG
(fig. 1-7)
E-Box t
FPV tFL
tORP
tnAGP _
Forecasting errors
Central EMS
OR Dispatching
And AG planning Optimal
MGT planning
tmMGTP _
tmMGT
M
mP _
1
Local
controller mth MGT
The flowchart of the energy management
26IV.2 Non-linear Constraints
Security: OR assessment with x % of LOLP;
Power balancing:
Maximization of RES usage: considering the battery capacity limitation (more PV power, larger battery storage !)
0))()(()()()( =)( _
N
1n 1_
tPttPtPtPt iMGT
M
iinAGresL
Non-linear characteristic of MGTs: )(%100)()(%50 max___max__ tPtPtP iMiMiM
Power output (p.u.)
CO
2eq
uiva
lent
(kg)
Power output (%)
Fuel
con
sum
ptio
n (k
Wh)
IV. Operating Power Reserve Dispatching
Load and OR
27IV.3 Management Strategy of OR Dispatching IV. Operating Power Reserve Dispatching
Two considered scenarios
OR dispatching strategies according to power sources: First strategy: OR on three MGTs only Second strategy: OR on three MGTs and PV AGs including batteries
According to PV energy during the day Uncertainty from
Scenario H ∗ _ Used PV power and load forecasting
Scenario L ∗ _ Stored in the batteries Only load forecasting
-20 -10 0 10 200
0.05
0.1
0.15
0.2
Generation Margin (kW)
Prob
abili
ty
2 p.m.
-30 -15 0 15 300
0.02
0.04
0.06
0.08
Generation Margin (kW)
Prob
abili
ty
2 p.m.(a) (b)
Scenario L Scenario H
Pre-processing of the uncertainty by energy management of batteries
OR quantification for the electrical system
28IV.4 OR Dispatching StrategiesIV. Operating Power Reserve Dispatching
More details can be found: X. Yan, D. Abbes, B. Francois, and Hassan Bevrani “Day-ahead Optimal Operational and Reserve Power Dispatching in a PV-based Urban Microgrid,” EPE 2016, ECCE Europe, Karlsruhe/ Germany.
Starting
PV energy is more than battery energy sizing
1<t<tnight
tF
N
n
tnAG LP
1_
tOR
M
m
tmMGT PP
1_
At time step t, if tF
tF LPV
Day Night
tF
N
n
tnAG PVP
1_
tOR
tF
tF
M
m
tmMGT PPVLP
1_
H L
1<t<tnightDay
tOR
tF
M
m
tmMGT PLP
1_
End
•
• •
•
•
0_ tnAGPt
ORP
0tMGTP
1st strategy
2nd strategy
Autonomous PV AG mode
1st strategy
2nd strategy
PV sure PV Not sure
tOR
tF
tF
M
m
tmMGT PPVLP
1_
Autonomous PV mode
tF
N
n
tnAG LP
1_
tOR
M
m
tmMGT PP
1_
0tMGTP
1st strategy
2nd strategy
PV AG sure
Full MGTs mode
PV AG sure
Forced PV AG to be sure But costly
•).( t
FtF
tBat PVLE t
FtBat LE
).( tOR
tF
tF
tBat PPVLE ).( t
ORtF
tBat PLE
1st strategy
2nd strategy
1st strategy
2nd strategy
Both strategies
Case 1.2 Case 1.1 Case 2.2 Case 2.1 Case 3.2 Case 3.1 Case 4
tOR
tF
N
n
tnAG PPVP
1_
tOR
tF
N
n
tnAG PLP
1_
29
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimal Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
Roadmap : Part V
30V.1 Day-ahead Optimal OR Planning on MGTsV. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
Technological features
SoC(t)
Dynamic Programming
Optimization
Operational Planning of MGT
J
Day-ahead Optimal Operational Planning Constraints Power Balance MGT Efficiency Battery Power
& Energy Limitation
Day-ahead PV
Power, Load and
Operating R
eserve References
Criteria Objective Function
? (t)
u(t)
z(t)
tORP
Electrical SystemM odeling
Loads
PV AG
Economic Cost
CO2 EmissionsMGT
tAGP t
MGTP
...... _t
mMGTP
Focus on the design of the microgrid central EMS. Unit commitment (UC) problem with dynamic programming (DP) is developed in order to
reduce the economic cost and CO2 equivalent emissions.
Operational and Reserve Power Dispatching
31
)(),...,(),( =(t) _2_1_ tPtPtPx iMGTMGTMGT
)(),...,(),( =(t) 21 tttu i
UC Problem: is an operation scheduling function.
V.2 Unit Commitment (UC) Problem
Problem: optimal operation of a cluster of MGTs (three in our case)
MGT1(30kW)
MGT2(30kW)
MGT3(60kW)
Three MGTs with different characteristics
Number of states Generators states
1 δ1=1; δ2=1; δ3=1
2 δ1=1; δ2=1; δ3=0
3 δ1=1; δ2=0; δ3=1
4 δ1=1; δ2=0; δ3=0
5 δ1=0; δ2=1; δ3=1
6 δ1=0; δ2=1; δ3=0
7 δ1=0; δ2=0; δ3=1
8 δ1=0; δ2=0; δ3=0
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
32V.3 Optimization Goals
Economic criteria: minimize the total fuel cost;
Environmental criteria: minimize the equivalent CO2 emission;
Best compromise criteria: make a compromise of economic and environmental criteria.
),( = 1___
ti
ti
CostiP
tiM
ti
tiC CCJ
T M
i
tiCJJ
1t 1_Cost =
),(2 = 12___2
ti
ti
COiP
tiM
ti
tiCO CCOJ
T M
i
tiCOJJ
1t 1_2CO2
)),(2)(1()),( ( = 12__
1___
ti
ti
COiP
tiM
ti
ti
ti
CostiP
tiM
ti
tiBC CCOCCJ
T M
i
tiBCJJ
1t 1_BC =
M: the number of MGTs;T: the number of operational steps.
the proportion rate, from 0 to 1
Fuel cost Start/stop penalty
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
33
DP: Systematically evaluates a large number of possible decisions in a multi-step problem considering the "transition costs".
Start
t=1
t=T
Stop
t=t-1
Optimization: calculate the minimum J(t) with vectors x(t) and u(t), using recursive equation
F(t, x(t), u(t)).
Calculate F(T, x(T), u(T)) with x(T) and u(T) to minimize J(T).
F1(1) F1(2) … F1(T-1) F1(T) F2(1) F2(2)… F2(T-1) F2(T) F3(1) F3(2)… F3(T-1) F3(T)F4(1) F4(2)… F4(T-1) F4(T) … . … . … . … .
F1(t) F1(T) F2(t) F2(T) F3(t) F3(T) F4(t) F4(T) … . … .
F1(T) F2(T) F3(T)F4(T) … .
No
Yes
V.3 Dynamic Programming (DP)
Multistage decision process formulation with backward recursion:
State
back-track optimal path
An evaluation of all possible configurations in each time step (Stages and States); A “back-track” operation from the end back to the beginning (Recursive Optimization).
Stage
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
34V.4 Case Study and Simulation Results (1)
In this case: rated load (110 kW), rated PV power (55 kW) and the OR (with 1 % of LOLP) coming from the net demand uncertainty assessment.
8h 10h 13h 16h 19h 22h 1h 4h 7h0
30
60
Fore
cast
PV (k
W)
Scenario HScenario L
8h 10h 13h 16h 19h 22h 1h 4h 7h0
60
120
Fore
cast
Load
(kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
10
20
Time (half hour)
OR
(kW
) With 1 % of LOLP
Scenario HScenario L
Scenario H, sunny day, 269.5 kWh; Scenario L, cloudy day, 128.4 kWh; The total battery capacity (150 kWh).
The daily load is 1082 kWh.
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
OR calculation with two scenarios.
35
Scenarios Optimization Criteria Cost (€) Pollution (kg) OR on AG (%) Ebat_Max (kWh)
H 1st strategy: Only MGTs
None 173 1224 0 78.6Environmental 169 1141 0 78.6
Economic 167 1167 0 78.6Best compromise 171 1156 0 78.6
H 2nd strategy: MGTs and AGs
None 173 1027 35.4 52.7Environmental 171 937 35.4 52.7
Economic 168 976 35.4 52.7Best compromise 170 952 35.4 52.7
None 185 1290 0 132.4L 1st Environmental 182 1181 0 132.4
strategy Economic 180 1245 0 132.4Best compromise 183 1195 0 132.4
None 188 1119 11.8 132.4L 2nd Environmental 184 993 11.8 132.4
strategy Economic 181 1063 11.8 132.4Best compromise 185 1006 11.8 132.4
V.4 Case Study and Simulation Results (2) Day-ahead Operational Planning Results.
71%
10%
19%
H 1st Strategy
MGT1MGT2MGT3AG=0
8h 10h 13h 16h 19h 22h 1h 4h 7h0
10
20
OR
(kW
)
PR in MGT1PR in MGT2PR in MGT3
20
(a) H1
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
36
Scenarios Optimization Criteria Cost (€) Pollution (kg) OR on AG (%) Ebat_Max (kWh)
H 1st strategy: Only MGTs
None 173 1224 0 78.6Environmental 169 1141 0 78.6
Economic 167 1167 0 78.6Best compromise 171 1156 0 78.6
H 2nd strategy: MGTs and AGs
None 173 1027 35.4 52.7Environmental 171 937 35.4 52.7
Economic 168 976 35.4 52.7Best compromise 170 952 35.4 52.7
None 185 1290 0 132.4L 1st Environmental 182 1181 0 132.4
strategy Economic 180 1245 0 132.4Best compromise 183 1195 0 132.4
None 188 1119 11.8 132.4L 2nd Environmental 184 993 11.8 132.4
strategy Economic 181 1063 11.8 132.4Best compromise 185 1006 11.8 132.4
V.4 Case Study and Simulation Results (2)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
5
10
15
20
Time (half hour)
OR
(kW
)
PR in MGT1PR in MGT2PR in MGT3PR in AG
(b) H2
30%
12%13.4%
35.4%
H 2nd strategy
MGT1
MGT2
MGT3
AG
35.4% of OR ison PV AGs
Day-ahead Operational Planning Results.
Each PV AG equally contributed to the OR
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
37V.4 Case Study and Simulation Results (2)
Day-ahead Operational Planning Results.
Scenarios Optimization Criteria Cost (€) Pollution (kg) OR on AG (%) Ebat_Max (kWh)
H 1st strategy: Only MGTs
None 173 1224 0 78.6Environmental 169 1141 0 78.6
Economic 167 1167 0 78.6Best compromise 171 1156 0 78.6
H 2nd strategy: MGTs and AGs
None 173 1027 35.4 52.7Environmental 171 937 35.4 52.7
Economic 168 976 35.4 52.7Best compromise 170 952 35.4 52.7
None 185 1290 0 132.4L Environmental 182 1181 0 132.4
1st strategy Economic 180 1245 0 132.4Best compromise 183 1195 0 132.4
None 188 1119 11.8 132.4L Environmental 184 993 11.8 132.4
2nd strategy Economic 181 1063 11.8 132.4Best compromise 185 1006 11.8 132.4
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
Cost and pollution are increased
Battery size is larger
38V.4 Results (3): MGTs Load Ratio H 1st Strategy: OR on MGTs
00,10,20,30,40,50,60,70,80,9
1
MGT1 MGT2 MGT3
MG
T P
ower
Rat
io
MGT Power Ratio with Strategy 2
None Environmental Economic
00,10,20,30,40,50,60,70,80,9
1
MGT1 MGT2 MGT3
MG
T P
ower
Rat
io
MGT Power Ratio with Strategy 1
None Environmental Economic
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
1 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
2 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T1 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T2 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
H 2nd Strategy: OR on MGTs and PV AGs
0.5
0.8
Better use of MGTs in their higher efficiency operation domain.
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
Turnoff all the MGTs
39V.4 Results (3): Obtained system security H 1st Strategy: OR on MGTs
7h 10h 13h 16h 19h 22h 1h 4h 7h0
0.5
1
LO
LP
(%) Strategy 1
7h 10h 13h 16h 19h 22h 1h 4h 7h0
1
2Strategy 2
LO
LP
(%)
Time (half hour)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
1 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
2 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T1 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T2 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
H 2nd Strategy: OR on MGTs and PV AGs
This is the risk we need to face with.
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
40V.4 Results (4): Battery State of Charge
8h 10h 13h 16h 19h 22h 1h 4h 7h
8h 10h 13h 16h 19h 22h 1h 4h 7h-40
0
40
P Batt
ery
(kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
20
40
60E Ba
tter
y (k
Wh)
Time (half hour)
9 7
8h 10h 13h 16h 19h 22h 1h 4h 7h-20
0
20
P Batt
ery
(kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
100
E Batt
ery
(kW
h)
T ime (half hour)
H 1st Strategy: OR on MGTs
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
1 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MGT
2 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T1 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
102030
P MG
T2 (k
W)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P MGT
3 (kW
)
8h 10h 13h 16h 19h 22h 1h 4h 7h0
50
P AG (k
W)
T ime (half hour)
H 2nd Strategy: OR on MGTs and PV AGs78.6 kWh
52.7 kWh
V. Day-ahead Unit Commitment (UC) Problem with Dynamic Programming (DP)
41
Predictive Analysis for Uncertainty: PV power and load forecasting
Operating Reserve Quantification:Loss of load probability (LOLP)
Day-ahead Optimal Planning:Unit commitment problem with
dynamic programming
OR Dispatching Strategies on Generators
A User-friendly EMS and Operational Supervisor
Roadmap : Part VI
42VI.1 Application VI. A User-friendly EMS and Operational Supervisor
….
Adopted Methods for the Urban
Microgrid Simulator
Environmental, Economic, Best Compromise
LOLP
Scenarios H&L
OR Quantify
Mining Collecting
Test Validation Training
Big Data
Predictive Analysis
Forecast with ANN
LOLP
PV Uncertainty
Load Uncertaitny
ND Uncertainty Assessment
Uncertainties Assessment for OR
Quantification
Constrains
Unit Commitment
Dispatching Strategies
Dynamic Programming
MGT & PV AG
Optimization Ceriteria
Data Collection and System Uncertainty
Analysis
OR Dispatching for UC Problem
with DP
Objective: to provide a complete set of user-friendly GUI to properly model uncertainties and optimal manage the details of PV AGs, loads, and MGTs one day-ahead.
43VI.2 Microgrid EMS Supervisor Frame Design
Main Interfaces Individual Modules
Data Collect and System Uncertainty
Analysis
Historical Data Collect for ANN Training Day-ahead Data Download: Weather
Information, Load, and PV Power Data PV Power and Load Demand Forecast by
Using Well Trained ANN
System Uncertainties Assessment and OR
Power Quantification
PV Power Uncertainty Load Demand Uncertainty Net Demand Uncertainty OR Quantification
Operational and OR Dispatching
Dispatching Strategies PV AGs and MGTs Power References
VI. A User-friendly EMS and Operational Supervisor
44
Main Interfaces Individual Modules
Data Collect and System Uncertainty
Analysis
Historical Data Collect for ANN Training Day-ahead Data Download: Weather
Information, Load, and PV Power Data PV Power and Load Demand Forecast by
Using Well Trained ANN
System Uncertainties Assessment and OR
Power Quantification
PV Power Uncertainty Load Demand Uncertainty Net Demand Uncertainty OR Quantification
Operational and OR Dispatching
Dispatching Strategies PV AGs and MGTs Power References
VI. A User-friendly EMS and Operational Supervisor VI.2 Development software
45
Main Interfaces Individual Modules
Data Collect and System Uncertainty
Analysis
Historical Data Collect for ANN Training Day-ahead Data Download: Weather
Information, Load, and PV Power Data PV Power and Load Demand Forecast by
Using Well Trained ANN
System Uncertainties Assessment and OR
Power Quantification
PV Power Uncertainty Load Demand Uncertainty Net Demand Uncertainty OR Quantification
Operational and OR Dispatching
Dispatching Strategies PV AGs and MGTs Power References
VI. A User-friendly EMS and Operational Supervisor VI.2 Microgrid EMS Supervisor Frame Design
46VI.3 DemonstrationVI. A User-friendly EMS and Operational Supervisor
47VII. Conclusions and perspectives
PV power variability and load demand variability are analyzed; A probabilistic method for the OR calculation is proposed (with two different kinds of
ND uncertainty assessment methods); The dynamic joint operational and OR dispatching strategies are developed; Day-ahead operational and OR planning with DP is proposed by considering different
optimization strategies; A User-friendly EMS and Operational Supervisor is developed.
“Big data” for distributed RES uncertainty analysis, prediction for more then one day and with multiple time steps, and intraday adjustment;
Optimization method to improve the battery efficiency and its lifespan;
Considering load shedding;
How the uncertainties are propagated in the electrical system.
Prospectives
Contributions
Thank you for your attention !
1. X. Yan, B. Francois, and D. Abbes, “Uncertainty Analysis for Power Reserve Quantification in an Urban Microgrid Including PV Generators”, Elsevier, Renewable Energy, Vol (106), June 2017, pp. 288–297. (Accepted 9 January, 2017)
2. X. Yan, B. Francois, and D. Abbes, “Operating Reserve Quantification and Day-ahead Optimal Dispatching of a Microgrid with Active PV Generators,” Elsevier, Sustainable Energy, Grids and Networks, under review.
3. X. Yan, B. Francois, and D. Abbes, ”Solar radiation forecasting using artificial neural network for local power reserve,” in Electrical Sciences and Technologies in Maghreb (CISTEM), 2014 International Conference, pp. 1-6.
4. X. Yan, B. Francois, and D. Abbes, “Operating power reserve quantification through PV generation uncertainty analysis of a microgrid,” in PowerTech, 2015 IEEE Eindhoven, 2015, pp. 1-6.
5. Yan, X., Abbes, D., Francois, B. and Bevrani, H., 2016, October. Day-ahead optimal operational and reserve power dispatching in a PV-based urban microgrid. In Power Electronics and Applications (EPE'16 ECCE Europe), 2016 18th European Conference on (pp. 1-10). IEEE.
Related Publications (https://www.researchgate.net/profile/Xingyu_Yan)