Page - 51 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2019,12, 164 3.TheProposedForecastStrategy ANNsarewidely used as forecasters because these networks canpredict the non-linearities ofSGs’ loadwith lowconvergence time. However, sometimes theachievedpredictionaccuracy is notupto themark. Thus, leading to theadoptionofoptimization techniques that cansignificantly enhancethepredictionaccuracyofANNs.However, thecostpaidtoachievehighaccuracyis increased convergence time. Therefore,weaimtowards thedevelopmentof anewDALFstrategyusing the conceptofhybrid integrationsubject to: (i) improvementofpredictionaccuracy;and(ii) reductionof convergence time. OurproposedDALFstrategyisimplementedinthreeinterconnectedmodules: (i)apre-processing module; (ii) a forecast module; and (iii) an optimization module. Given the input data, the pre-processingmoduleremovesredundantandirrelevantsamples fromthe inputdata.Usingsigmoid activationfunctionandMARA,thehybridANN-basedforecastmodulepredicts theDALofanSG. Finally, theoptimizationmoduleminimizespredictionerrors to improveaccuracyof theoverallDALF strategy. Blockdiagramof theproposedmodel is shown inFigure3. Detaileddescriptionof each module isas follows. Figure3.Blockdiagramof theproposedmodularapproachforanhour. 3.1. Pre-ProcessingModule SincetheANN-basedforecasterpredicts loadofthenextday, theinputdatamustbepre-processed subject toremovalof redundantandirrelevantsamplesdueto tworeasons: (i) redundant featuresdo notprovidemore informationandthusunnecessarily increase theexecutiontimeduringthe training process (will be laterdiscussed in the forecastmodule); and (ii) irrelevant featuresdonotprovide useful informationandactasoutliers.Detaileddescriptionof thepre-processormodule isas follows. Asmentionedearlier, thedatapreparationmodulereceives the input load-timeseries (historical). Suppose, following is the input loaddata: P= ⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣ p(h1,d1) p(h2,d1) p(h3,d1) . . . p(hm,d1) p(h1,d2) p(h2,d2) p(h3,d2) . . . p(hm,d2) p(h1,d3) p(h2,d3) p(h3,d3) . . . p(hm,d3) p(h1,d4) p(h2,d4) p(h3,d4) . . . p(hm,d4) p(h1,d5) p(h2,d5) p(h3,d5) . . . p(hm,d5) ... ... ... ... ... p(h1,dn) p(h2,dn) p(h3,dn) . . . p(hm,dn) ⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦ (1) where,dn is thenthday,hm is themthhourof theday,and p(hm,dn) ispowerusagevalueof theof the nthdayat themthhour. Similarly,wehave inputdewpoint temperaturedata inamatrixTDP, input drybulb temperaturedata inamatrixTDB, andthe input typeofday(workingdayorholiday)data inamatrixDT. Choosingn is totallydependentonthechoiceofdesigner.Greatervalueofnmeans 51