Seite - 342 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1282 3.DataAnalysisandPreprocessing 3.1. Selectionof Influenced Indexes Consideringthat thehumanactivitiesarealwaysdisturbedbymanyexternal factorsandthenthe power load isaffected, someeffective featuresareselectedas factors. In thispaper, theselectionof factors ismainlybasedonfouraspects: (1) The historical load. Generally speaking, the historical load impacts on the current load in short-termloadforecasting. Inthispaper, thedailymaximumload,dailyminimumload,average daily load, peak average loadof previousday, valley average loadof previousday, average loadof thedaybefore, average loadof 2 daysbefore, average loadof 3daysbefore, average loadof4daysbefore,average loadof 5daysbeforeandaverage loadof6daysbeforeare taken intoconsideration. (2) The temperature. Aspeople use temperature-adjustingdevices to adapt to the temperature, inapreviousstudy[23–25], temperaturewasconsideredasanessential input featureandthe forecastingresultswereaccurateenough. Inthispaper, themaximumtemperature, theminimum temperatureandtheaverage temperatureareselectedas factors. (3) Theweathercondition.Wemainly take intoaccount theseasonalpatterns,humidity,visibility, weatherpatterns,airpressureandwindspeed. Thefourseasonsarerepresentedas1,2,3and4 respectively. Fordifferentweatherpatterns,wesetdifferentweights: {sunny,cloudy,overcast, rainy}={0,1,2,3}. (4) Thedaytype. In thisaspect, the typeofdayanddateare takenintoconsideration. Thetypeof datemeans thedaysaredivided intoworkdays (Monday–Friday),weekend(Saturday–Sunday), andholidays. Theweightsof three typesofdateare0,1and2respectively. For thedate,weset differentweight: {Monday,Tuesday,Wednesday,Thursday,Friday,Saturday,Sunday}={1,2, 3, 4, 5, 6, 7}. 3.2. FactorAnalysis Originally proposed by British psychologist C.E. Spearman, factor analysis is the study of statistical techniques for extractinghighly interrelatedvariables into onegroup, andeach typeof groupbecomesa factor that reflectsmostof theoriginal informationwith fewer factors.Notonlydoes factoranalysis reduce indicators’dimensionsandimprovethegeneralizationof themodelbutalso thecommonfactors it elicited toportrayandreplaceprimitivevariables cancommendablymirror andexplain thecomplicatedrelationshipbetweenvariables,keepingdatamessageswithessentially no less information. In thispaper, factoranalysis isused toextract factors that canreflect themost informationof theoriginal22 influencingvariables,whoseresult is showninTableTable2. Firstofall,Table1gives theresultofKaiser-Meyer-Olkin (KMO)andtheBarlett testof sphericity that canserveasacriteria to judgewhether thedata is suitable for the factoranalysis. Thestatistic valuemore than0.7can illustrate thecompatibilityandthe0.74obtainedfromthepower loaddata confirmsthecorrectnessof factoranalysis. Table 2 shows six factors that are extracted from 22 original variables. The accumulative contribution rate at 84.434%,more than80%, reflects that thenewsix factors candeliver themost informationof theoriginal indicators. It canbeseenfromTable2 that factor1 thatmainlyrepresents the history load accounts for the largest proportion at 35.128%. In addition, considering that the variables infactor1maynotbesufficientonbehalfof thehistorical load, thepapercarriedoutafurther analysisof thepreviousdatabymeansof thecorrelationanalysiswhichcanbeseen inpart3.2. Factor 2whichmainlyrepresentsmeteorologyelementaccounts for19.646%,andtheremainingfour factors are10.514%,7.746%,6.087%,and5.313%,respectively. 342