Seite - 44 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

energies Article Short-TermLoadForecastinginSmartGrids: AnIntelligentModularApproach AshfaqAhmad1,*,NadeemJavaid2 ,AbdulMateen2 ,MuhammadAwais2 and ZahoorAliKhan3 1 SchoolofElectricalEngineeringandComputing,TheUniversityofNewcastle,Callaghan2308,Australia 2 DepartmentofComputerScience,COMSATSUniversity Islamabad, Islamabad44000,Pakistan; nadeemjavaid@comsats.edu.pk(N.J.); ammateen49@gmail.com(A.M.); amawais@hotmail.com(M.A.) 3 Computer InformationScience,HigherCollegesofTechnology,Fujairah4114,UAE;zkhan1@hct.ac.ae * Correspondence: ashfaqahmad@ieee.org;Tel.: +61-416-618-613 Received: 11November2018;Accepted: 1 January2019;Published: 4 January2019 Abstract:Dailyoperationsandplanning inasmartgridrequireaday-aheadloadforecastingof its customers. Theaccuracyofday-ahead load-forecastingmodelshasa significant impactonmany decisionssuchasschedulingof fuelpurchases, systemsecurityassessment, economicschedulingof generatingcapacity,andplanningforenergytransactions.However,day-aheadloadforecasting isa challenging taskdue to itsdependenceonexternal factors suchasmeteorological andexogenous variables. Furthermore, theexistingday-aheadload-forecastingmodelsenhanceforecastaccuracy bypaying thecostof increasedexecution time. Aimingat improving the forecast accuracywhile notpayingthe increasedexecutions timecost,ahybridartificialneuralnetwork-basedday-ahead load-forecastingmodel forsmartgrids isproposedinthispaper. Theproposedforecastingmodel comprises threemodules: (i)apre-processingmodule; (ii)a forecastmodule;and(iii)anoptimization module. In thefirstmodule, correlated laggedloaddataalongwith influentialmeteorologicaland exogenous variables are fed as inputs to a feature selection techniquewhich removes irrelevant and/orredundantsamples fromthe inputs. In thesecondmodule,asigmoidfunction(activation) and amultivariate auto regressive algorithm (training) in the artificial neural network are used. Thethirdmoduleusesaheuristics-basedoptimizationtechniquetominimize the forecasterror. In the thirdmodule, ourmodifiedversion of an enhanceddifferential evolution algorithm is used. Theproposedmethodisvalidatedviasimulationswhere it is testedonthedatasetsofDAYTOWN (Ohio,USA)andEKPC(Kentucky,USA). Incomparisonto twoexistingday-aheadload-forecasting models, results showimprovedperformanceof theproposedmodel in termsofaccuracy,execution time,andscalability. Keywords: artificialneuralnetwork; loadprediction;smartgrid;heuristicoptimization;energytrade; accuracy 1. Introduction Anexisting/traditionalgridsystemneedsrenovationtobridge theever-increasinggapbetween demandandsupply andalso tomeet essential challenges suchasgrid reliability, grid robustness, customer electricity cost minimization, etc. [1]. In this regard, recent integration of advanced communication technologies and infrastructures into traditional grids have led to the formation ofsocalledsmartgrids(SGs)[2]. Thenationalnational instituteofstandardsandtechnology(NIST)[3] conceptualdiagramofsmartgrid (SG) is showninFigure1. Thisconceptualdiagramcanbeusedasa referencemodelforstandardizationworksinsevenSGdomains: generation, transmission,distribution, endusers,markets,operations,andserviceproviders. Eachdomain involvesoneormoreSGactors (e.g.,devices,systems,programs,etc.) tomakedecisionsforrealizinganapplicationbasedonexchange Energies2019,12, 164;doi:10.3390/en12010164 www.mdpi.com/journal/energies44