yhighresidentialloadswillhavehighervariationsinshort-termloadduetoweatherconditionsthanregionswithrelativelyhighindustrialloads.Industrialregions,however,willhaveagreatervariationduetoeconomicfactors,suchasholidays.Asanexample,Figure.showstheloadvariationoveroneday,startingatmidnight.Figure.Exampleofloadvariationduringoneday、HybridNeurakNetworksOurshort-termloadforecastingsystemconsistsoftwotypesofnetworks:linearneuralnetworkARMAmodelandfeedforward.Non-linearneuralnetwork.Thenon-linearneuralnetworkisusedtocapturethehighlynon-linearrelationbetweentheloadandvariousinputparameters.WeusethelinearneuralnetworktogenerateanARMAmodelwhichwillbemainlyusedtocapturetheloadvariationoveraveryshorttimeperiod(onehour)..LinearNeutalNetworksThegeneralmultivariatelinearmodeloforderpwithindependentx,istptpititttptpititttuxcxcxcxcxczazazazaz负荷数据训练而产生预测MAPE逆值)。