Page - 380 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1138 for individualcustomers. Thestructurecanbeextendedif there ismore information in thepractical situation. Thebasic theory isalsoacceptable formedium-termloadforecastingandlong-termload forecasting,butdifferent influence factors shouldbeconsideredandthemodel shouldbechanged withdifferent input,output,andinnerstructure forgoodperformance. 3. ExperimentsandResults In thissection, theexperimentsaredescribedindetailandtheresultsareshowninfiguresand tables. The specific discussion for results is elaborated after the results andprove the improved performancecomparedtoothermethods. Thedata forexperimentsarerecordedinSection2.2. 3.1. ClusteringAnalysis forLoadCurveof IndividualCustomers Before theshort-termloadforecastingusingGRUneuralnetworks, the loadcurvesof individual customers are clustered to different categories for samples with K-means clustering algorithm. The parameterK is selectedas3byElbowmethod. Thereare746customers in theWanjiangarea in Dongguancity. The loadmeasurementdatashouldbeprocessedwith0–1standardization to thesame scale forclustering toreduce the impactofdifferentmagnitudesanddimensions. Theclustering is donefor10 timeswith loadcurves in10days for the individualcustomers. Theclusteringresultsare obtainedwith theaverageresults in10daysandthenumberofeachclusteringcategory isshownin Table4. Thestandardizedcurves for30selectedcustomers in threecategoriesonaweekdayareshown inFigures6–8. Table4.Numberofeachclusteringcategory. Categories Category1 Category2 Category3 NumberofCustomers 221 308 217 6DPSOLQJ 3RLQWV 6DPSOLQJ 3HULRG PLQXWHV Figure6.Loadcurvesof30customers inCategory1. Ascanbeseen inFigures6–8,differentcustomershavedifferentcharacteristicsofelectricityuse. AccordingtoFigure6, thereare twoelectricpeaks inaday. Theeveningpeak ishigher thanthenoon peak. Theclassic representationof this characteristic inFigure6 is residential customers. Different fromFigure 6, Figure 7maintains thepeak from9a.m. to late at night exceptnoon. Theyare the general loadcurvesof industryandbusinesscustomers. InFigure8, thereare twoelectricpeaks in themorningandafternoon. It shouldbelong to thegovernmentand institutional customers. Even thoughafewcustomershavedifferenceswiththeoverall curve, this is thebestclusteringfor themand itdoesnot influence theoverallperformancegreatly.With theclusteringof individualcustomers, the 380