基于神经网络的风电功率预测方法研究

中华人民共和国北京市清华大学附属中学BEIJING/THEPEOPLEREPUBLICOFCHINA/THEHIGHSCHOOLATTACHEDTOTSINGHUAUNIVERSITY基于神经网络的风电功率预测方法研究TheStudyonWindPowerForecastingMethodBasedonNeturalNetwork蒲舒桐ShutongPu指导老师李劲松Supervisor/JingsongLi二〇一七年八月I摘要随着可再生能源的大力发展，大规模风电接入电网，相较于常规能源，风力发电具有间歇性、波动性和随机性，给电网的安全、稳定运行带来了巨大的挑战，而功率预测对风力的合理、安全、有效的利用至关重要，所以能够及时、精确地对风电功率进行预测的意义尤为重大。风电功率短期预测误差主要是内在随机性因素和外在随机性因素造成的。内在随机性因素是指预测系统本身存在缺陷或不完善，外在随机性因素是指系统输入的数据不完善或输入数据存在误差。要解决在风电功率短期预测中随机性因素带来误差的问题，引入性能更为优越的人工神经网络算法来改进以及设计新的预测系统是目前主要研究方向。目前国内外该领域研究逐步深入，国外已有多套成熟的预测系统投入实际使用，而国内在此领域还未到达令人满意的程度。本文综述了国内外风力发电的发展状况，风电功率预测技术的研究现状、基本原理，阐述了不同分类标准下的风电功率预测方法，分析了基于历史数据和数值天气预报的功率预测方法。在此基础上，采用人工神经网络算法进行风电功率短期预测，构建了BP神经网络和卷积神经网络风电功率预测模型，探讨了建模过程中学习精度的确定、隐含层节点数以及训练函数的选取，得到最优的神经网络预测模型。仿真结果显示BP神经网络模型的预测精度和稳定性都不够理想。针对于BP神经网络的易陷入局部极小值、稳定性差的问题，建立卷积神经网络风电功率短期预测模型，仿真结果显示预测系统的预测精度和稳定性有了明显的提高，卷积神经网络算法弥补了BP神经网络算法的不足，验证了该算法的有效性、可行性。关键词：风电场功率预测，短期预测，数值天气预报，BP神经网络，卷积神经网络IIABSTRACTWiththedevelopmentofrenewableenergy,large-scalewindpoweraccesstothegrid,comparedtoconventionalenergy,windpowerisintermittent,volatilityandrandomness,whichhasbroughtgreatchallengestothesafeandstableoperationofthegrid.Itisveryimportanttopredictthereasonable,safeandeffectiveuseofwindpower,soitisverysignificanttopredictthewindpowerintimeandaccurately.Theshort-termpredictionerrorofwindpowerismainlycausedbytheinternalrandomnessfactorandtheexternalrandomnessfactor.Theinherentrandomnessfactormeansthatthepredictionsystemitselfisdefectiveorimperfect,andtheexternalrandomnessfactormeansthatthesysteminputdataisnotperfectortheinputdataisinerror.Tosolvetheproblemofrandomerrorinwindpowershort-termprediction,itisthemainresearchdirectiontointroducetheartificialneuralnetworkalgorithmwithsuperiorperformancetoimproveanddesignnewforecastingsystem.Athomeandabroadinthisfieldresearchgraduallyin-depth,thereareseveralsetsofmatureforeignforecastsystemputintopracticaluse,thedomesticinthisareahasnotyetreachedasatisfactorylevel.Thispapersummarizesthedevelopmentstatusofwindpowergenerationathomeandabroad,theresearchstatusandbasicprinciplesofwindpowerforecastingtechnology,expatiatesonthewindpowerforecastingmethodunderdifferentclassificationstandards,andanalyzesthepowerforecastingmethodbasedonhistoricaldataandnumericalweatherforecast.Basedontheartificialneuralnetwork(ANN)algorithm,theBPneuralnetworkandConvolutedneuralnetworkwindpowerforecastingmodelareconstructed.Themodelingprocessisselectedtodeterminethelearningaccuracy,thenumberofhiddenlayernodesandtheselectionoftrainingfunction,togettheoptimalneuralnetworkpredictionmodel.ThesimulationresultsshowthattheBPneuralnetworkmodelhasgenericpredictionaccuracyandstability.AimingattheproblemthattheBPneuralnetworkiseasytofallintothelocalminimumandpoorstability,theshort-termwindpowermodelofConvolutionneuralnetworkisestablished.Thesimulationresultsshowthatthepredictionaccuracyandstabilityofthepredictionsystemareobviouslyimproved,whicheffectivelysolvestheshortcomingsofBPneuralnetworkalgorithm.ThevalidityandreliabilityoftheConvolutionneuralnetworkalgorithmareverified.KEYWORDS:Windfarmpowerprediction,Short-termforecast,Numericalweatherforecast,BPneuralnetwork，Convolutionalneuralnetwork目录摘要....................................................................................................................................................................IABSTRACT.............................................................................................................................................................II第1章绪论................................................................................................................................................................11.1课题研究背景及其意义..........................................................................................................................11.1.1课题研究背景................................................................................................................................11.1.2课题研究目的与意义....................................................................................................................21.2国内外风电功率预测研究现状..............................................................................................................31.2.1国外风电功率预测研究现状.......................................................................................................31.2.2国内风电功率预测研究现状.......................................................................................................51.3本文的主要工作.......................................................................................................................................6第2章风电功率短期预测方法..............................................................................................................................72.1引言.............................................................................................................................................................72.2风电功率预测方法分类..........................................................................................................................72.2.