Seite - 15 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2226 Table 6. Parameters combination of LS-SVR determined by CQFOA and other algorithms for GEFCom2014(July). OptimizationAlgorithms Parameters MAPEofValidation(%) ComputationTimes(s) γ σ LS-SVR-CQPSO[36] 375 92 0.96 139 LS-SVR-CQTS[37] 543 59 1.04 107 LS-SVR-CQGA[38] 684 62 0.98 159 LS-SVR-CQBA[39] 498 90 0.95 239 LS-SVR-FOA 413 48 1.51 79 LS-SVR-QFOA 384 83 1.07 212 LS-SVR-CQFOA, 482 79 0.79 147 Basedonthesametrainingsettings,anotherrepresentativemodel, theback-propagationneural network (BPNN) is comparedwith the proposedmodel. The forecasting results of thesemodels mentionedaboveandtheactualvaluesfor IDAS2014,GEFCom2014(Jan.) andGEFCom2014(July)are given inFigures2–4, respectively. This indicates that theproposedLS-SVR-CQFOAmodelachievesa betterperformance thantheotheralternativemodels, i.e., closer to theactual loadvalues. Figure2.ForecastingvaluesofLS-SVR-CQFOAandothermodels for IDAS2014. 15