基于局部神经网络的电力需求预测研究摘要电力需求预测是实现电力系统安全、经济运行的基础，对一个电力系统而言,提高电网运行的安全性和经济性，改善电能质量，都依赖于准确的电力需求预测。中长期电力预测可以为新发电机组的安装以及电网的规划、增容和改建等提供决策支持，是电力规划部门的重要工作之一。本文提出基于局部神经网络的预测模型进行电力需求预测。首先，采用模式预处理的方法对原始时间序列归一化处理等；其次，用时间窗将时间序列分割成一系列矢量形式的样本，并在此基础上用主成分分析方法对样本进行特征提取，以降低样本的维数；再次，运用KNN搜索获得待预测样本的K个近邻；最后，用近邻样本训练一个RBF神经网络，并利用该局部神经网络进行待预测样本的预测。在局部RBF神经网络模型的选择上，本文采用基于网格搜索与交叉验证法相结合的方法进行神经网络模型参数的选择，避免了参数选择的盲目性、随意性，提高了预测精度。本文通过仿真实验对比了局部神经网络预测模型与全局RBF网络预测模型的性能和执行效率，结果表明局部神经网络预测模型无论在预测性能上还是在执行效率上都好于全局RBF网络预测模型。关键词：局部神经网络；主成份分析；KNN搜索；网格搜索；交叉验证LocalNeuralNetworkbasedPowerDemandForecastingAbstractPowerdemandforecastingisthebasicissuetoensurethestableandeconomicaloperationofpowersystem.Forapowersystem,boththeimprovementofelectricityqualityandtheincreasingofoperationstabilityandeconomydependonthepreciseforecastingofpowerdemand.Mediumorlongtermpowerdemandforecastingcanoffersomedecisionsupportsfortheinstallationofnewgeneratorsandtheplanning,capacityincreaseandrebuildingoftheelectricnetwork,whichisoneofthemostimportantworkofthepowerplanningdepartments.Alocalneuralnetworkmodelisproposedforpowerdemandforecasting.Firstly,preprocessingoforiginaltimeseriessuchasnormalizationisdonewithsomepatternpreprocessingmethods.Secondly,thetimeseriesaresegmentedintoaseriesofsampleswithvectorformbytimewindow,andthenfeatureextractionwithprincipalcomponentanalysis(PCA)isdoneonthesamplesetinordertoreducedimensionsofsamples.Thirdly,knearestneighborsofthesampletobepredictedisobtainedbyknearestneighborssearching(KNNS).Finally,theknearestneighborsamplesareusedtotrainaRBFneuralnetwork,andthenthesampletobepredictedispredictedwiththeRBFneuralnetworkwelltrained.Inthispaper,gridsearchingandcross-validationmethodareusedtosearchtheoptimalparameteroftheRBFneuralnetwork,whichcanavoidtheblindnessandcasualnessoftheparameterselectionandimprovetheforecastingaccuracy.Comparisonofthepredictionperformanceandimplementationefficiencybetweenthelocalneuralnetworkmodelandtheglobalartificialneuralnetwork(ANN)isdonebysimulationexperiments.Experimentalresultsshowedthatthepredictionperformanceandexecutionefficiencyoflocalneuralnetworkmodelarebetterthanthoseofglobalartificialneuralnetworkmodel.Keywords：Localneuralnetwork;Principalcomponentanalysis;k-nearestneighborssearching;Gridsearching;Cross-validation目录1.绪论···········································································································11.1短期电量需求的意义及任务··········································································11.1.1电量预测的意义················································································...
VIP
