Page - 49 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2019,12, 164 definitions,optimizationof these functions is thusapotential researcharea.However, the integration ofoptimizationtechniquefurthercomplicates theoverallmethodology. (v)Stochasticdistribution-basedmodels:Themodel in [17]predicts thepowerusage timeseriesby usingaprobability-basedapproach. Themodelalsoconfigureshouseholdappliancesbetweenholidays andworkingdays.Amajorassumption in thiswork is thegaussiandistribution-basedon-off cyclesof householdappliances,numberofappliances,andpowerconsumptionpatternofappliances. In this work,notonlyawiderangeofappliances isconsideredbutalsohighflexibilitydegreeofappliances is considered.However,absenceofclosedformsolutionmakes thegaussian-basedforecast strategyvery complex.Moreover, theseassumptionscannotbealwaystrue, thus,accuracyofthepredictedload-time series ishighlyquestionable.Animprovementover [17] ispresented in [18]. This researchworkuses 1 2 regulizer toovercomethecomputationalcomplexityofgaussiandistribution-basedDALFstrategy in[17].Moreover, theproposedDALFstrategycancaptureheteroscedasticityof loadinamoreefficient wayascompared[17]. Simulationsareconductedtoprove that theproposedDALFstrategyperforms better thantheexistingone. Tosumup,weconcludethat [18]hasovercomethecomplexityof [17] to someextent;however, thebasicassumptions (gaussiandistribution-basedon-off cyclesofhousehold appliances,numberofappliances,andpowerconsumptionpatternofappliances) stillhold thebases and thusmake theproposalhighlyquestionable in termsofaccuracy. Asemi-parametric additive forecastmodel ispresented in [19]. Thiswork isbasedonpoint forecastandcalculates theprediction intervalsviaamodifiedbootstrapalgorithm. Similarly,anothersemi-parametricgeneralizedadditive loadforecastmodel ispresented in [20]. In termsof forecasthorizon, thegeneralizedadditive forecast model isbetter thanthenon-generalizedonedueto itsdual forecast capability; short-termandmiddle term.However,both the forecastmodelsarenotsufficient in termsofaccuracywhencomparedto the ANN-basedmodels. Theoverall classificationhierarchyof forecast techniques isshowninFigure2, andtheir summary isgiven inTable1. ! " # $! Figure2.Classificationofexistingforecast techniques. 49