Seite - 336 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article AHybridBA-ELMModelBasedonFactorAnalysis andSimilar-DayApproachforShort-Term LoadForecasting WeiSunandChongchongZhang* DepartmentofBusinessAdministration,NorthChinaElectricPowerUniversity,Baoding071000,China; bdsunwei@126.com * Correspondence:mr_zhangcc@126.com Received: 9April2018;Accepted: 2May2018;Published: 17May2018 Abstract: Accurate power-load forecasting for the safe and stable operation of a power system is of great significance. However, the randomnon-stationary electric-load time serieswhich is affectedbymanyfactorshinders the improvementofpredictionaccuracy. In lightof this, thispaper innovativelycombinesfactoranalysisandsimilar-daythinkingintoapredictionmodel forshort-term loadforecasting.After factoranalysis, the latent factors thataffect loadessentiallyareextractedfrom anoriginal22 influence factors. Then,consideringthecontributionrateofhistory loaddata,partial autocorrelationfunction(PACF) isemployedtofurtheranalyse the impacteffect. Inaddition,ant colonyclustering(ACC) isadoptedtoexcavate thesimilardays thathavecommonfactorswith the forecastday. Finally,anextremelearningmachine(ELM),whoseinputweightsandbias thresholdare optimizedbyabatalgorithm(BA),hereafter referredasBA-ELM, isestablishedtopredict theelectric load.AsimulationexperienceusingdataderivingfromYangquanCityshowsitseffectivenessand applicability, andtheresultdemonstrates that thehybridmodelcanmeet theneedsof short-term electric loadprediction. Keywords: short-term load forecasting; factor analysis; ant colony clustering; extreme learning machine;batalgorithm 1. Introduction Short-termloadforecastingisanimportantcomponentofsmartgrids,whichnotonlycanachieve thegoalofsavingcostbutalsoensureacontinuousflowofelectricitysupply[1].Moreover,against thebackgroundofenergy-savingandemission-reduction,accurateshort-termloadpredictionplays an important role inavoidingawasteof resources in theprocessofpowerdispatch. Nevertheless, it shouldbenotedthat the inherent irregularityandlinear independenceof the loadingdatapresent anegativeeffectontheexactpower loadprediction. Since the 1950s, short-term load forecasting has been attracting considerable attention from scholars.Generallyspeaking, themethodsfor loadforecastingcanbeclassifiedinto twocategories: traditionalmathematical statisticalmethodsandapproacheswhicharebasedonartificial intelligence. The conventionalmethods like regression analysis [2,3] and time series [4] aremainly based on mathematical statisticmodels suchas thevectorauto-regressionmodel (VAR)andauto-regressive movingaveragemodel (ARMA).With thedevelopmentofscienceandtechnology, theshortcomings of statisticalmodels, such as the effect of regression analysis basedonhistorical data thatwill be weakenedwith theextensionof timeor theresultsof time-seriespredictionthatarenot idealwhen thestochastic factorsare large,arebeginningtoappearandarecriticizedbyresearchers for their low non-linearfittingcapability. Energies2018,11, 1282;doi:10.3390/en11051282 www.mdpi.com/journal/energies336