基于遗传算法改进BP神经网络的短期风电功率预测研究

02012年“挑战杯”大学生课外学术科技作品竞赛及创业设计大赛作品名称：基于遗传算法改进BP神经网络的短期风电功率预测研究作品类别：社会科学类项目成员：刘知发陈军母桑妮向亚军唐艳利联系电话：[phone]完成时间：2012年3月20日1基于遗传算法改进BP神经网络的短期风电功率预测研究摘要风能发电作为21世纪重要的研究课题之一，是清洁、可再生资源的之首。对降低污染，舒缓能源消耗带来的压力有着至关重要的作用。本文通过时间序列、遗传算法和BP神经网络等方法建立了4个风电功率预测模型。通过Matlab编程，得出了不同方法预测结果，并对其准确性进行比较。本文首先对国内外风电产业发展现状做了分析。在此基础上，第2章确定以移动平均预测法、随机时间序列预测法、BP神经网络预测法对问题进行探讨，通过Excel与Matlab混合编程，得出移动平均预测法、随机时间序列预测法、BP神经网络预测法的准确率分别为82%、70%、84%，合格率分别为85%、65%、92%。得出BP神经网络预测法明显优越于其他两种方法。接着运用BP神经网络预测出的数据做了预测的相对误差分析，从中得出了6组预测值的相对误差（见表3.1），并做了对比误差分析图，通过误差分析图得出“风电机组汇聚会减小风电功率预测误差”的结论。并对造成该结论的原因做了解析，提出了在风力允许范围内，增加风电机组的汇聚度，可进一步减小误差的预期判断。在第2章的基础上，为了进一步提高风电功率实时预测的准确度，建立了遗传算法与BP神经网络相结合来对风电功率进行预测的模型。通过模型对东北某发电厂一周进行预测，并与实测值进行比较，得到其准确率与合格率高达89%与95%。最后，对本次课题得出的结论做了分析，总结风电功率的预测结果和存在的问题，以及提出需要进一步改进的地方。关键词：风电功率预测随机时间序列BP神经网络误差分析遗传算法2ABSTRACTWindpowergeneration,oneofthemostsignificantresearchsubjectsin21stcentury,topsamongthecleanandrenewableresources,whichplaysacriticalroleinreducingpollutionandeasingstressproducedbyenergyconsumption.Thethesis,usingmethodsincludingtimeseries,geneticalgorithmandBPnervesnetwork,establishes4predictionmodelswithwindpower.ThroughMatlabprogramming,theauthorarrivesatvariouspredictionresults,andcheckstheiraccuracy.Firstofall,thethesisanalysesdevelopmentsituationofwindpowergenerationathomeandabroad.Andonthebasisofthefirststep,thesecondchapterdiscussestheissuebymobileaveragepredictionmethod,randomtimeseriespredictionmethodandBPnervesnetworkpredictionmethod.ThroughmixedprogrammingofExcelandMatlab,theaccuracyrateofaveragepredictionmethod,randomtimeseriespredictionmethodandBPnervesnetworkpredictionmethodis82%,70%,and84%respectively,whilequalifiedrateofthemis85%,65%,and92%.Apartfromtheabove-mentioned,theauthoralsodrawsthatBPnervesnetworkpredictionmethodissuperiortotheothertwoones.ThenusingthedataforecastbyBPnervesnetworkpredictionmethod,theauthor,makingpredicativelycomparativeerroranalysis,acquiressixgroupsofcomparativeerrorsofpredictionfigure(seeChart1),anddrawsananalysischartofcomparativeerrors.Asum-upthatgatheringofwindturbineisabletodecreasepredictiveerrorsofwindpowerisreachedbymeansoferroranalysis.Thecausesgivingrisetosuchaconclusionareinterpretedaswell,andanideathatincreasinggatheringdensityofwindturbinecanreduceexpectantjudgmentoferrorsisalsoputforward.BasedontheChapter2,amodel,combingwithgeneticalgorithmmethodandBPnervesnetwork,isestablishedtoforecastwindpower,soastoincreasinglyimprovereal-timemonitoringaccuracyofwindpower.Withthemodel,theauthor,makingaweek-longpredictiontowardsanorthwestpowerplantandcomparingwithactualfigures,comestoaconclusionthattheaccuracyandqualifiedratesareupto89%and95%.Ultimately,theauthoranalyzestheconclusionofthesubject,summarizesthepredictiveresultsandexistingproblems,andproposessomeaspectsthatdemandimproving.KeyWords:predictionofwindpower;randomtimeseries;BPnervesnetwork;erroranalysis;geneticalgorithm3目录摘要...............................................................................................................................................................1ABSTRACT....................................................................................................................................................2目录.............................................................................................................................................................3第1章引言.............................................................................................................................................51.1风电产业发展现状[1,2]......................................................................................................................51.2背景分析与研究意义.......................................................................................................................51.3国内外研究动态...............................................................................................................................61.3.1国外研究现状[3].............................................................................................................................61.3.2国内研究现状................................................................................................................................6第2章风电功率实时预测及误差分析.....................................................................................................82.1移动平均预测法预测风电功率.......................................................................................................82.1.1移动平均法基本概念...................................................................................................................82.1.2移动平均法基本原理...................................................................................................................82.1.3移动平均法的特点.......................................................................................................................82.1.4一次移动平均法...........................................................................................................................82.1.5二次平移预测法....................................................