Page - 388 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1138 6DPSOLQJ 3RLQWV 6DPSOLQJ 3HULRG PLQXWHV $FWXDO 3URSRVHG PHWKRG /670 511V 6$(V %311V Figure17.The loadcurvesofCustomer53990001basedontheproposedmethodandtheothercurrentmethods. 4.Conclusions Toincreasethestabilityandeconomyofpowergrids,amethodforshort-termloadforecastingwith multi-sourcedatausingGRUneuralnetworks isproposedinthispaper,whichfocusesonindividual customers. Theproposedstructureof thewholenetworkisshowninFigure5.Thereal-worldloaddata of individual customers is recordedfromDongguanPowerSupplyBureauofChinaSouthernPower Grid inGuangdongProvince,China. Before training, thecustomerswith loaddataareclustered into threecategoriesbyK-meansclusteringalgorithmtoreducethe interferenceofdifferentelectricityuse characteristics. Then, the environment factors arequantifiedandput into the inputof theproposed networksformoreinformation.TheGRUunitsare introducedintothenetworkfor itssimplerstructure andfasterconvergencecomparedtoLSTMblocks. Theresults inFigures12and13showthatclustering andmulti-source inputcanhelpto improvetheperformanceof loadforecasting.TheaverageMAPEcan belowas10.98%fortheproposedmethod,whichoutperformstheothercurrentmethodssuchasBPNNs, SAEs,RNNsandLSTM.The improvement isnotable (Figures15–17). Ingeneral, theavailabilityand superiorityof theproposedmethodareverifiedinthispaper. Inthefuture,combiningwiththetechnique ofpeakprediction couldbe a subjectworth studying for load forecasting. Moreover, since the load forecastingfor thecustomers inallpowergridareas isa large-scale task, transfer learningandcontinuous learningwillbeconsideredbasedontheproposedframeworkforhigh-efficiencyloadforecasting. AuthorContributions:Y.W.andM.L.conceivedthemain idea;Z.B.andS.Z. collectedthedataanddesignedthe experiments;Y.W.,M.L.andZ.B.performedtheexperiments;andY.W.wrote thepaper. Acknowledgments: Thiswork is supportedby theZhejiangProvincialNatural ScienceFoundationofChina underGrantLZ15F030001. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. References 1. Espinoza,M.;Suykens, J.A.;Belmans,R.;DeMoor,B.Electric loadforecasting. IEEEContr. Syst.Mag. 2007, 27, 43–57. [CrossRef] 2. Hong,T.Energyforecasting: Past,present,andfuture.Foresight Int. J.Appl. Forecast. 2014,32, 43–48. 3. Gross,G.;Galiana,F.D.Short-termloadforecasting.Proc. IEEE1987,75, 1558–1573. [CrossRef] 388