View
1
Download
0
Category
Preview:
Citation preview
TheSmartgridsteamispartoftheMontefioreResearchUnitoftheULg,containsaround15researchersandisheadedbyPr.DamienErnst
http://blogs.ulg.ac.be/damien-ernst2
Machinelearningisaboutextracting{patterns,knowledge,information}fromdata
Clusterimages
SIRICortona OK Google
Convertvoicesignalintosentences
Makeon-linerecommandations
Recognizepatternsinimages
Interpretsentences
Google photos
6
Machinelearningstudiesandbuildsalgorithmsthatlearnfromandmakepredictionsondata
SupervisedLearninginanutshell:
Imagineyouhaveasetofdata {(x1, y1), (x2, y2), …, (xn, yn)}
represented by black points on thefigure.
To be able to estimate the value of anoutputy for any inputx, You “train” aMachineLearningalgorithmusingthesedata.Youobtaintheblueline.
Thequalityoftheestimatedependsondataquality/quantity:withmorepoints,e.g. the black circles, you would forinstancegettheredcurve.
7
x
y
Recentadvancesinmachinelearning
Machinelearningalgorithmshaverecentlyshownimpressiveresults,inparticularwhen input data are images: this has led to the identification of a subfield ofMachineLearningcalledDeepLearning.
The term “deep” refers to the fact that those learning architectures, mainlyArtificialNeuralNetworks,aremadeofseverallayers.
Zoom on a neuron
8
Deepneuralnetworkarchitectures
Source: http://www.ais.uni-bonn.de/deep_learning/images/Convolutional_NN.jpg
9
Wait…ANNarenotnew,right?
ANNdatebacktothesixties.TrainingANNwasnotaneasytaskuntilrecently.Recentprogressistwofold:
• Smart(er)trainingapproaches• GPUcalculus
10
Fromsupervisedlearningtoreinforcementlearning
Supervisedlearningtechniques(inparticular(deep)convolutionalnetworks)maybeusedas a block in a more complex structure, inparticular in Dynamic Programming (DP) orModelPredictiveControl(MPC)schemes.
This connects to reinforcement learning, anarea of machine learning originally inspiredby behaviorist psychology, concerned withhowsoftwareagentsoughttotakeactionsinan environment so as to maximize somenotionofcumulativereward.
Deepreinforcementlearningcombinesdeeplearning with reinforcement learning (and,consequently,inDP/MPCschemes).
12
Agent
Environment
ActionReward
PlayingAtariwithdeepreinforcementlearning
13
At Google Deepmind At ULg
Human-levelcontrolthroughdeepreinforcementlearning.Nature,2015.VolodymyrMnih,KorayKavukcuoglu,DavidSilver,AndreiA.Rusu,JoelVeness,MarcG.Bellemare,AlexGraves,MartinRiedmiller,AndreasK.Fidjeland,GeorgOstrovski,StigPetersenCharlesBeattie,AmirSadik,IoannisAntonoglou,HelenKing,DharshanKumaran,DaanWierstra,ShaneLegg&DemisHassabis
Breakingnews
Recentbreakthroughs in the fieldofAI for thegameofGOhavebeendonebyGoogleDeepmind.
These results have been obtained by combining Deep Convolutional NetworkswithMonteCarloTreeSearchtechniques.
Theresultingagent,AlphaGo,achieved99.8%winningrateagainstotherGOAI,anddefeatedtheEuropeanGochampionby5gamesto0.
14
MasteringthegameofGowithdeepneuralnetworksandtreesearch.Nature,2016.DavidSilver,AjaHuang,ChrisJ.Maddison,ArthurGuez,LaurentSifre,GeorgevandenDriessche,JulianSchrittwieser,IoannisAntonoglou,VedaPanneershelvam,MarcLanctot,SanderDieleman,DominikGrewe,JohnNham,NalKalchbrenner,IlyaSutskever,TimothyLillicrap,MadeleineLeach,KorayKavukcuoglu,ThoreGraepel&DemisHassabis
Wanttoknowmore?
Googleislaunchinganewdeeplearningcourse(incollaborationwithUdacity):https://www.udacity.com/course/deep-learning--ud730
YoumayalsobeinterestedinNVidiaDeepLearningcourse:https://developer.nvidia.com/deep-learning-courses
OrevenStanfordMoocaboutMachineLearning:https://www.coursera.org/learn/machine-learning
15
Optimization: decide the values that some variables can take,underasetofcontraints,soastomaximizeanobjective.
A long tradition of numerical solutions and theoretical analysis.Givenassumptionsonmodels,onecaneventuallygetguaranteesaboutsolutions.
Howisoptimizationconnectedtomachinelearning?Learningproblemscanbecastedasoptimizationproblems
Howismachinelearningconnectedtooptimization?Machine learning actually solves some (or part of) optimizationproblems(e.g:RL,ortuningofanalgo,orproxytoanalgo)
Machinelearningistightlycoupledtooptimization
17
Machinelearningistightlycoupledtooptimization
Anillustrationofthesimplexalgorithm. The simplexalgorithmwasinventedbyG.Dantzig. Itdatesback to thesecondworldwar.
This can be used to solvemany practical optimizationproblems.
18
set NUTRIENT ordered; set FOOD ordered;
param cost {FOOD} >= 0; param minNutrient {NUTRIENT} >= 0; param maxNutrient {i in NUTRIENT} >= minNutrient[i]; param amount {NUTRIENT,FOOD} >= 0;
# Variables var Buy {j in FOOD} integer;
# Objective minimize Total_Cost: sum {j in FOOD} cost[j] * Buy[j];
(or minimize nutrient_amount {i in NUTRIENT}: sum {j in FOOD} amount[i,j] * Buy[j];)
# Constraints subject to Diet {i in NUTRIENT}: minNutrient[i] <= sum {j in FOOD} amount[i,j] * Buy[j] <= maxNutrient[i];
Optimizationreliesonananalyticalmodel...
Example: Building the lunch menu, a first application of AI for energy ;)
19
set NUTRIENT ordered; set FOOD ordered;
param cost {FOOD} >= 0; param minNutrient {NUTRIENT} >= 0; param maxNutrient {i in NUTRIENT} >= minNutrient[i]; param amount {NUTRIENT,FOOD} >= 0;
# Variables var Buy {j in FOOD} integer;
# Objective minimize Total_Cost: sum {j in FOOD} cost[j] * Buy[j];
(or minimize nutrient_amount {i in NUTRIENT}: sum {j in FOOD} amount[i,j] * Buy[j];)
# Constraints subject to Diet {i in NUTRIENT}: minNutrient[i] <= sum {j in FOOD} amount[i,j] * Buy[j] <= maxNutrient[i];
Optimizationreliesonananalyticalmodel...
Example: Building the lunch menu, a first application of AI for energy ;)
+ Data
Your lunch menu
20
OptimizationandMachinelearninghavedifferentaims
Intheoptimizationworld,amethodtargetsoneproblemclass,orevenaninstanceofaproblem,andatheoryisobsessedbyoptimality(canIproveitmathematically?)andefficiency(canIcomputeitefficiently?)
Machinelearningisfocusedonstatisticalsignificance(reachinga trade off between overfitting and “misrepresentation”),replicability to other problems with few adaptation, andinterpretabilityofresults
22
Worldenergyconsumptionoutlookaround2010
Fossil fuels Nuclear Renewables
Biomass heatSolar hotwaterGeothermal heatHydropowerEthanolBiodieselBiomass electricityWind powerGeothermal electricitySolar PV powerSolar CSPOcean powerSources: IEA
25
OptimizationhasplentyofapplicationsintheEnergyindustry
Electricalpowersystems:• Productionplanning:unitcommitment• Managinggridconstraints:optimalpowerflow
Oilandgasindustry:• Wheretodig?Inwhichsequence?
Logisticsandtransportation:• VehicleRoutingProblems
Industrialprocesses:• Reductionordisplacementofenergyconsumption
26
Example:Day-aheadelectricitypricesinEuropearedeterminedbyEuphemia
EUPHEMIA is the market coupling algorithm for European Power exchanges, implemented and developed in-house by N-SIDE, a spin-off of UCL and ULg
Used daily by Power Exchanges to fix pan-EU day-ahead electricity prices in 19 EU countries.
Computing market prices & volumes by: • coupling national markets • maximizing total economical welfare • optimizing network capacity utilization • modeling complex economical constraints
Extension to whole Europe in progress
http://energy.n-side.com/day-ahead/
27
EvolutionoftheenergysystemGlobal Grid(s) versus Microgrids
Prof. Damien Ernst - University of LiègeELIA Stakeholders͛ days
28
WhyarewenowtalkingaboutAI,andnotjustaboutoptimization?
Wearenowtryingtooptimizemoreandmorelocally,becauserenewable energy sources are distributed, data is ubiquitousandcomputationpoweraswell.
However,theratio“gain/(timetospendforgatheringthedataand solving the problem)” is way smaller than for largecentralizedprojects.
AI offers the possibility to automate the data gathering,modelingandoptimizationstages.Forinstance,learnfromthehabitsofusersofahouse,proposesomecarpoolingoptions,correlateallthiswithcalendarevents.
32
Rethinkingtheoperationofdistributionsystems
Activenetworkmanagement.Smart modulation of generation sources, loads and storages so as to operatesafelytheelectricalnetworkwithouthavingtorelyonsignificant investments ininfrastructure.
GREDORproject.Redesigning in an integrated way the whole decision chain used for managingdistributionnetworksinordertoperformactivenetworkmanagementoptimally(i.e.,maximisationofsocialwelfare).
www.gredor.be
33
Empoweringconsumersanddistributedgeneration
Microgrids aremodern, localized, small-scalegrids, contrary to the traditional, centralizedelectricitygrid(macrogrid).
Some microgrids can operate disconnectedfrom the centralized grid and operateautonomously, strengthen grid resilience andhelpmitigategriddisturbances.
Optimizing the sizing and the operation of amicrogrid requires both optimization and AItechniques.
34
ReferencesHowtodiscountdeepreinforcementlearning:towardsnewdynamicstrategies.V.François-Lavet,R.Fonteneau,D.Ernst.DeepReinforcementLearningWorkshop,NIPS2015.http://arxiv.org/abs/1512.02011
Benchmarkingforbayesianreinforcementlearning.M.Castronovo,D.Ernst,A.Couëtoux,R.Fonteneau,http://arxiv.org/pdf/1509.04064.pdf
ImitativeLearningforOnlinePlanninginMicrogrids.S.Aittahar,V.François-Lavet,S.Lodeweyckx,D.Ernst,R.Fonteneau.DataAnalyticsforRenewableEnergyIntegration,Volume9518oftheseriesLectureNotesinComputerSciencepp1-15,2015.
Theglobalgrid.S.Chatzivasileiadis,D.Ernst,G.Andersson.RenewableEnergy,Volume57,September2013,Pages372–383.
GlobalGrid(s)versusMicrogrids.Ernst,D.http://hdl.handle.net/2268/188217
TheGREDORproject.Redesigningthedecisionchainformanagingdistributionnetworks.Ernst,D.http://hdl.handle.net/2268/188487
Activenetworkmanagementforelectricaldistributionsystems:problemformulationandbenchmark.Gemine,Q.,Ernst,D.,&Cornélusse,B.(2014).arXivpreprintarXiv:1405.2806.
DSIMA:Atestbedforthequantitativeanalysisofinteractionmodelswithindistributionnetworks.Mathieu,S.,Louveaux,Q.,Ernst,D.,&Cornélusse,B.(2016).SustainableEnergy,GridsandNetworks,5,78-93.
ActiveManagementofLow-VoltageNetworksforMitigatingOvervoltagesDuetoPhotovoltaicUnits.Olivier,F.,Aristidou,P.,Ernst,D.,VanCutsem,T.IEEETransactionsonSmartGrid,2016.
Supervisedlearningofintra-dailyrecoursestrategiesforgenerationmanagementunderuncertainties.Cornélusse,B.,Vignal,G.,Defourny,B.,&Wehenkel,L.(2009,June).InPowerTech,2009IEEEBucharest(pp.1-8).IEEE.
38
Recommended