Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Junhua Wu,Slava Shekh,NataliiaSergiienko,BenjaminCazzolato,Boyin Ding,
FrankNeumann,andMarkusWagner
PresentedbyMarkus
§Waveenergyisawidelyavailablebutlargelyunexploitedsourceofrenewableenergy
§ Therearedozensofactivewaveenergyconverter(WEC)projectsexploringavarietyoftechniquesforharnessingwaveenergy
§ InpartnershipwiththeSchoolofMechanicalEngineering,weareconsideringawaveenergyconverter(WEC)calledCETO
§ TheCETOsystemconsistsofoneormorefullysubmergedbuoys
SingleWECoptimisation§ Ringwood(2004),McCabe(2010)andHals(2011)optimisevariousaspectsofsemi-submergedbuoys,suchasgeometryandcontrol
§ Korde (2015)investigatesdifferentcontrolstrategies formaximisingpowerabsorptionoftwobuoys,oneofwhichisfullysubmerged
WECarraysandtheiroptimisation§ Cruz(2009)andWeller(2010)exploretheeffectofvariousfactorsonarrayperformance,includingdevicespacingandarraylayout
§ Fitzgerald(2007),Child(2010)andSnyder(2014)optimisearraysofsemi-submergedWECs
ThereisalackofresearchonoptimisingarraysoffullysubmergedWECs
Submerged buoy
Tether
Power take-off system
Sea floor
Advantages• Invisible from the shore• Higher survival in storm conditions• Hydrodynamics allow 2 times more
power to be absorbed from surge motion (e.g. via three-tether or asymmetric mass)
§ ThevariablesoftheCETOmodelleadtoanoptimisation problem:Whatisthebestcombinationofbuoyradiitousefordifferentarraysizes?
§ Asolution(configuration)canberepresentedas:(r1,…,rn)e.g.thelayoutshown is(2,2.5,4,5)
§ Asolutioncanbeevaluatedusingtheq-factor,whichistheratioofthepowerabsorptionofabuoyarraycomparedtothepowerabsorptionofthesamebuoysinisolation
2x2Array 3x3Array
Best(q-Factor) 0.999 0.996
Worst (q-Factor) 0.965 0.933
Best2x2 Best3x3
waves
Hiddenbiasofqfactor:smallerbuoysarealwaysmoreefficientthanlargeronesàWidth(considersbuoydimensions)
1.10E+07'
1.11E+07'
1.12E+07'
1.13E+07'
1.14E+07'
1.15E+07'
1.16E+07'
1.17E+07'
1' 22'
43'
64'
85'
106'
127'
148'
169'
190'
211'
232'
253'
274'
295'
316'
337'
358'
379'
400'
Power'out'(W
a9)'
Genera=ons'
Best'Power'
Mutated'Power'
0.65%
0.7%
0.75%
0.8%
0.85%
0.9%
50v10% 50v5% 50v4% 50v3% 50v2% 50v1%
Accuracy%
Power%
RCW%
Speed-upbyfrequencyreductionfrom2100minutesto42minutes(50buoys).
Anoldcomputersciencetrick…caching!!!Matlab mostfrequentlycalls:integral,factorial,bessel.Fora50-WEC-array,1millioncallstointegral aremade(90%duplicates).èCachingreducestheruntimeby85%.
Now:runtime6minutes(factor350).
(400g) (200g)1+1 EA CMA-ES
Po
we
r O
ut
(Wa
tt)
×107
1.2
1.21
1.22
1.23
1.24
1.2525 Buoys
(400g) (200g)1+1 EA CMA-ES
Po
we
r O
ut
(Wa
tt)
×107
2.14
2.16
2.18
2.2
2.22
2.2450 Buoys
80070060050040030020010000
100
200
300
400
500
600
700
800×10
5
3.8
4
4.2
4.4
4.6
4.8
5
5.2
5.4
5.6
5.8
à TuningintheendwithCMA-ESispossible,though.
0 100 200 300 400 500 600 700 800 900 10000
100
200
300
400
500
600
700
800
900
1000×10
5
3
3.5
4
4.5
5
5.5
6
010020030040050060070080090010000
100
200
300
400
500
600
700
800
900
1000×10
5
3
3.5
4
4.5
5
5.5
6
1800160014001200100080060040020000
200
400
600
800
1000
1200
1400
1600
1800×10
5
2
2.5
3
3.5
4
4.5
5
5.5
6
Algorithms:localoptimanotexploitedSpeed-up:simplificationnotadequate
Computationtime:8CPUdaysvs.7CPUyears
§ TranslationtoC§ Parallelisation§ Increaseinaccuracy§Multi-objectiveoptimisationà PPSN2016:142-foldspeed-upwhilestillusing25frequencies.
Actualnextsteps§Wavedirections:distribution(happeningnow)§ “mechanicalengineering”-analysisofresults(happeningnow)§ CarnegietosetuparraysofWECsaroundAustralia (jointARCgranthappeningnow)
§ Power-takeoff-controlleroptimisation§Machinelearningmodelstolearntheinteraction(happeningnow)
§Over100peopleplayedthe“optimisepoweroutput”gameusingAndroidtabletsatOpenDay 2017
§ Leaderboard:http://od.mewx.org/
§RefinedversiontobeusedatIngenuity2017(31Oct,thousandsofattendees)
MSSoftwareEngineeringstudents:Chenwei Feng,Mengyu Li,Yuanzhong XiaSupervisor:Dr. MarkusWagner
§ Goal:model&predictpoweroutputbasedonfarmlayout§ MachinelearningtechnologyasquickandprecisesurrogatesforNataliia’s analyticalmodel(frequencydomain)
§ https://mse.mewx.org§ 2buoysdoable,4buoysdifficult(imprecise)
MSSoftwareEngineeringChenwei Feng,Mengyu Li,
Yuanzhong Xia
§ Goal:model&predictpoweroutputbasedonspringconstantk,dampercoefficientd
§ 4buoys§ Scikit-learn(Python)§ https://github.cs.adelaide.edu.au/a1668648/HonoursWEC
Onesettingforallbuoys(neuralnetwork,randomforest)
Differentsettingsforeachbuoy(best100settings)
§ Educatedguessasareference:Whatisagoodlayout?Grid?Linear?Hexagonal?
§Wavescomefromtheleft…
PhDstudentMehdiNeshatMSstudentYuanzhong Xia
Andthesamefor4and9buoys…
Similarfor4and9buoys…
Outputof1isolatedbuoy:4.92e5W Thisis1-dimensional… howabout2D?
2buoys– Characterisationofeffectsforoptimisationpurposes§Wavescomefrombottomleft,50msafetydistance§ 1st buoyisat(100,0),shownisthewavefarm’stotalpoweroutputlandscapedependingonthe2nd buoy
Best:notnexttoeachother,butslightlyoffset
4 buoysMappingforeachbuoygiventheotherthreebuoys:
Besteverfound:
16buoys§Weseepatterns…
Besteverfound:
Farm’spower,withomittedbuoy’slocationasdashedline
Maxtetherelongationacrossfarm(arbitrarylimit:3m,here:4.5mreached)
§ Basedonthecharacterisation,wecandesignproblem-specificalgorithms.Whyisthisimportant?
4buoys:maxoutputisverysimilaracrossapproaches(scalenotvisible)
16buoys:+5%fortherightmosttwocustomapproaches
§ Basedonthecharacterisation,wecandesignproblem-specificalgorithms.Whyisthisimportant?
Markus [email protected]
http://cs.adelaide.edu.au/~markus/Theslideswillbemadeavailabletoday.