第41卷第3期陕西科技大学学报Vol.41No.32023年6月JournalofShaanxiUniversityofScience&TechnologyJun.2023*文章编号:2096-398X(2023)03-0200-08考虑特征提取和优化LSSVM的短期光伏功率预测岳有军,刘金林,赵辉,王红君(天津理工大学电气工程与自动化学院天津市复杂系统控制理论及应用重点实验室,天津300384)摘要:针对传统光伏功率预测因特征提取不足导致预测精度不高的问题,提出一种基于卷积神经网络(ConvolutionalNeuralNetwork,CNN)和长短期记忆网络(LongShort-TermMem-ory,LSTM)进行特征提取以及改进麻雀算法(ImproveSparrowSearchAlgorithm,ISSA)优化最小二乘支持向量机(LeastSquaresSupportVectorMachine,LSSVM)的短期光伏功率预测模型.该模型首先结合CNN、LSTM的优点构成CNN-LSTM特征提取模型,用于提取光伏发电功率数据中的隐藏特征和长期依赖性特征,然后将提取出的特征向量输入到经ISSA优化的LSSVM模型中进行预测,得到最终的预测结果.实验结果表明,所提出的CNN-LSTM-ISSA-LSSVM模型能取得很好的预测精度,且明显高于其它模型,验证了其有效性.关键词:光伏功率预测;卷积神经网络;长短期记忆网络;麻雀搜索算法;最小二乘支持向量机中图分类号:TM615文献标志码:AShort-termPVpowerpredictionconsideringfeatureextractionandoptimisedLSSVMYUEYou-jun,LIUJin-lin,ZHAOHui,WANGHong-jun(CollegeofElectricalEngineeringandAutomation,TianjinKeyLaboratoryofControlTheoryandApplicationforComplexSystems,TianjinUniversityofTechnology,Tianjin300384,China)Abstract:Tosolvetheproblemthatthepredictionaccuracyoftraditionalphotovoltaicpowerpredictionisnothighduetoinsufficientfeatureextraction,aconvolutionalneuralnetwork(CNN)andlongshort-termmemory(LSTM)forfeatureextractionandimprovedsparrowsearchalgorithm(ISSA)areproposed.Optimizetheshort-termphotovoltaicpowerpredic-tionmodeloftheleastsquaressupportvectormachine(LSSVM).ThemodelfirstcombinestheadvantagesofCNNandLSTMtoformaCNN-LSTMfeatureextractionmodel,whichisusedtoextracthiddenfeaturesandlong-termdependencefeaturesinphotovoltaicpowergen-erationdata,andtheninputtheextractedfeaturevectorsintotheLSSVMmodeloptimizedbyISSAforprediction,andobtainthefinalpredictionresults.ExperimentalresultsshowthattheproposedCNN-LSTM-ISSA-LSSVMmodelcana...
