Seite - 288 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article ANovelHybridIntervalPredictionApproachBased onModifiedLowerUpperBoundEstimationin CombinationwithMulti-ObjectiveSalpSwarm AlgorithmforShort-TermLoadForecasting JiyangWang1,YuyangGao2,* ID andXuejunChen3 1 Facultyof InformationTechnology,MacauUniversityofScienceandTechnology,Macau999078,China; 17098531i011001@student.must.edu.mo 2 SchoolofStatistics,DongbeiUniversityofFinanceandEconomics,Dalian116025,China 3 GansuMeteorologicalServiceCentre,Lanzhou730020,China;xuejunchen1971@163.com * Correspondence: gaoyuyang@hotmail.com;Tel.:+86-18340831947 Received: 25May2018;Accepted: 10 June2018;Published: 14 June2018 Abstract:Effectiveandreliable loadforecasting isan importantbasis forpowersystemplanningand operationdecisions. Its forecastingaccuracydirectlyaffects thesafetyandeconomyof theoperation of thepowersystem.However,attaining thedesiredpoint forecastingaccuracyhasbeenregardedas achallengebecauseof the intrinsic complexityandinstabilityof thepower load. Considering the difficultiesofaccuratepoint forecasting, intervalprediction isable to tolerate increaseduncertainty andprovidemore informationforpracticaloperationdecisions. In this study,anovelhybridsystem for short-term load forecasting (STLF) is proposed by integrating a data preprocessingmodule, amulti-objectiveoptimizationmodule,andanintervalpredictionmodule. Inthissystem,thetraining process isperformedbymaximizing the coverageprobability andbyminimizing the forecasting intervalwidthat thesametime. Toverifytheperformanceoftheproposedhybridsystem,half-hourly loaddataare set as illustrativecasesand twoexperimentsare carriedout in four stateswith four quarters inAustralia. Thesimulationresultsverifiedthesuperiorityof theproposedtechniqueand theeffectsof thesubmoduleswereanalyzedbycomparingtheoutcomeswith thoseofbenchmark models. Furthermore, it isprovedthat theproposedhybridsystemisvaluable in improvingpower gridmanagement. Keywords: short-termloadforecasting; intervalprediction; lowerupperboundestimation;artificial intelligence;multi-objectiveoptimizationalgorithm;datapreprocessing 1. Introduction Loadforecasting isofupmostsignificanceandaffects theconstructionandoperationofpower systems. In thepreparationof thepowersystemplanningstage, if the loadforecastingresult is lower thantherealdemand, the installedanddistributioncapacitiesof theplannedpowersystemwillbe insufficient. Thepowergeneratedwill notbeable tomeet electricitydemandof the community it serves, and theentire systemwillnotbeable tooperate inastablemanner. Conversely, if the load forecast is toohigh, itwill result inpowergeneration, transmission,anddistribution,ata largerscale, that cannotbe fullyused in the realpower system. The investmentefficiencyand theefficiencyof theresourceutilizationwillbereducedin thissituation. Therefore,effectiveandreliablepower load forecastingcanpromoteabalanceddevelopmentof thepowersystemwhile improvingtheutilization of energy. There are variouspower load forecastingmethods and, commonly, load forecasting is classified intoshort-term,medium-term,andlong-term,basedontheapplicationfieldandforecasting Energies2018,11, 1561;doi:10.3390/en11061561 www.mdpi.com/journal/energies288