Seite - 265 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1678 forecast accuracyandreliabilityof thedata source. In liveoperational forecast systems, reliability isvaluedhighly,andinputtingdata intoasimplermodelmayworktomakeamorerobustsystem. More features are thusnot always anadvantage, if the improvement in accuracy is insufficient to justify theaddedimplementationandmaintenancecost. Initialexperimentsusinglongshort-termmemorynetworkshavenotshownnotableimprovement over theresultsattainablewith theSVRmodel.However, futureworksshouldexplore this typeof model further,as ithas thepotential tosimplify the featureselectionprocedureandmake iteasier to transfer theseresults toawiderangeofdistrictheatingsystemsaroundtheworld. AuthorContributions:Conceptualization,G.B.A.andM.D.;Methodology,G.B.A.andM.D.;FormalAnalysis, M.D.; Investigation, M.D. and O.S.K.; Data Curation, M.D.; Writing—Original Draft Preparation, M.D.; Writing—Review & Editing, G.B.A. and M.D.; Visualization, M.D.; Supervision, G.B.A. and A.B.; Project Administration,A.B.;FundingAcquisition,G.B.A.andA.B. Funding:This researchhas receivedfunding fromtheEuropeanUnion’sSeventhFrameworkProgrammefor research, technologicaldevelopmentanddemonstrationundergrantagreementnoENER/FP7/609127/READY. Acknowledgments:WewouldliketothanktheDanishMeteorologicalInstituteforprovidingtheweatherforecast data.WealsothankAffaldVarmeAarhusforprovidingdataabout theheat loadandproductionsysteminAarhus. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. Abbreviations Thefollowingabbreviationsareusedin thismanuscript: Pt Heat loadinhour t (MW) Pt−l Heat loadlaggedby lhours (MW) Toutt Outdoor temperature inhour t ( ◦C) vwindt Windspeedinhour t (m/s) Isunt Solar irradiation inhour t (W/m 2) Toutt−l Outdoor temperature laggedby lhours ( ◦C) Isunt−l Solar irradiation laggedby lhours (W/m 2) Pˆt Heat loadforecastedforhour t (MW) α L2regularizationparameterof theMLPmodel C Regularizationparameterof theSVRmodel γ RBFkernelparameterof theSVRmodel RMSE Rootmeansquareerror (MW) MAE Meanabsoluteerror (MW) MAPE Meanabsolutepercentageerror (%) ME Meanerror (MW) OLS Ordinary least squaresregressionmodel MLP Multilayerperceptronmodel SVR Supportvector regressionmodel RBF Radialbasis functionkernel LSTM Longshort-termmemorynetworkmodel References 1. Ma,W.; Fang, S.; Liu,G.; Zhou, R. Modelingofdistrict load forecasting fordistributed energy system. Appl.Energy2017,204, 181–205, doi:10.1016/j.apenergy.2017.07.009. [CrossRef] 2. Frederiksen,S.;Werner,S.DistrictHeatingandCooling; Studentlitteratur: Lund,Sweden,2013. 3. Dotzauer,E. Simplemodel forpredictionof loads indistrict-heatingsystems. Appl. Energy2002,73, 277–284, doi:10.1016/s0306-2619(02)00078-8. [CrossRef] 4. Fang,T.;Lahdelma,R. Evaluationofamultiple linear regressionmodelandSARIMAmodel in forecasting heatdemandfordistrictheatingsystem. Appl. Energy2016,179, 544–552, doi:10.1016/j.apenergy.2016.06.133. [CrossRef] 265