?Abouttheauthor:AMatlabsimulationdeveloperwholovesscientificresearch.Hecultivateshismindandimproveshistechnologysimultaneously.ForcooperationonMATLABprojects,pleasesendaprivatemessage.
Personalhomepage:MatlabResearchStudio
Personalcredo:Investigatethingstogainknowledge.
Withtherapiddevelopmentofrenewableenergy,windpowergeneration,asanimportantpartofit,hasreceivedwidespreadattention.However,duetotheuncertaintyandvolatilityofwindpowergeneration,accuratepredictionofwindpowergenerationhasbecomeanimportantissue.Inordertosolvethisproblem,manyresearchershavebeguntoexploretheuseofmachinelearningalgorithmsforwindpowerdataprediction.
Amongmachinelearningalgorithms,longshort-termmemorynetwork(LSTM)isacommonlyusedrecurrentneuralnetwork(RNN)modelthatperformswellinprocessingsequencedata.LSTMnetworksareabletocapturelong-termdependenciesinsequencedata,therebyimprovingpredictionaccuracy.However,thetraditionalLSTMmodeloftenhassomeproblemswhenprocessingwindpowerdata,suchasslowmodeltrainingspeedandlowpredictionaccuracy.
Inordertosolvetheseproblems,thispaperproposesalongshort-termmemorySMA-biLSTMmodeloptimizedbasedontheSlimeMoldAlgorithm(SMA).Theslimemoldalgorithmisaheuristicalgorithmthatsimulatesthebehaviorofslimemoldsintheprocessofsearchingforfood.Ithasthecharacteristicsofglobalsearchandadaptability.ByapplyingtheslimemoldalgorithmtotheoptimizationprocessoftheLSTMmodel,thetrainingspeedandpredictionaccuracyofthemodelcanbeeffectivelyimproved.
Intheexperiment,weusedasetofwindpowerdatasetstotrainandtestthemodel.First,wepreprocessedtheoriginalwindpowerdata,includingdatacleaning,featureextraction,etc.Then,weusedtheslimemoldalgorithmtooptimizetheparametersoftheSMA-biLSTMmodel.Finally,wecomparedtheresultsofwindpowerdatapredictionusingthetraditionalLSTMmodelandtheoptimizedSMA-biLSTMmodel.
ExperimentalresultsshowthattheoptimizedSMA-biLSTMmodelshowsobviousadvantagesinwindpowerdataprediction.ComparedwiththetraditionalLSTMmodel,theSMA-biLSTMmodelhashigherpredictionaccuracyandfastertrainingspeed.ThisshowsthattheperformanceoftheLSTMmodelinwindpowerdatapredictioncanbeeffectivelyimprovedbyintroducingtheslimemoldalgorithmforoptimization.
Insummary,thispaperproposesalongshort-termmemorySMA-biLSTMmodeloptimizedbasedontheslimemoldalgorithmforwindpowerdataprediction.Experimentalresultsshowthatthemodelhasgoodperformanceinwindpowerdataprediction.Futureresearchcanfurtherexploretheapplicationofotheroptimizationalgorithmstofurtherimprovetheaccuracyandefficiencyofwindpowerdataprediction.
functionhuatu(fitness,process,type) figure plot(fitness) gridon title([type,'Fitnesscurve']) xlabel('Numberofiterations/times') ylabel('Fitnessvalue/MSE') figure subplot(2,2,1) plot(process(:,1)) gridon xlabel('Numberofiterations/times') ylabel('L1/piece') subplot(2,2,2) plot(process(:,2)) gridon xlabel('Numberofiterations/times') ylabel('L2/piece') subplot(2,2,3) plot(process(:,3)) gridon xlabel('Numberofiterations/times') ylabel('K/times') subplot(2,2,4) plot(process(:,4)) gridon xlabel('Numberofiterations/times') ylabel('lr') subtitle([type,'Hyperparameterschangewiththenumberofiterations'])
[1]LongZhongxiu.Researchandimplementationoflandslidepredictionbasedonsoilslopedataclassificationmodel[D].SouthwestJiaotongUniversity,2020.
[2]WangYongsheng.Researchonshort-termwindpoweroutputpowerpredictionbasedondeeplearning[J].[2023-09-08].
[3]WangHaiyue.Researchontheapplicationoffuzzyinferencesystembasedongranularcomputingintimeseriesdataprediction[D].ShandongNormalUniversity,2019.