Seite - 18 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2226 Table10.ResultsofWilcoxonsigned-ranktest for IDAS2014. ComparedModels WilcoxonSigned-RankTest T0.025 =81 T0.05 =91 p-Value LS-SVR-CQFOAvs. BPNN 0T 0T 0.000** LS-SVR-CQFOAvs. LS-SVR-CQPSO 72T 72T 0.022** LS-SVR-CQFOAvs. LS-SVR-CQTS 64T 64T 0.017** LS-SVR-CQFOAvs. LS-SVR-CQGA 67T 67T 0.018** LS-SVR-CQFOAvs. LS-SVR-CQBA 60T 60T 0.012** LS-SVR-CQFOAvs. LS-SVR-FOA 50T 50T 0.009** LS-SVR-CQFOAvs. LS-SVR-QFOA 68T 68T 0.019** TDenotes that theLS-SVR-CQGAmodelsignificantlyoutperformstheothermodels. ** implies thep-value is lower thanα=0.025; * implies thep-value is lower thanα=0.05. Table11.ResultsofWilcoxonsigned-ranktest forGEFCom2014(Jan.). ComparedModels WilcoxonSigned-RankTest T0.025 =81 T0.05 =91 p-Value LS-SVR-CQFOAvs. BPNN 0T 0T 0.000** LS-SVR-CQFOAvs. LS-SVR-CQPSO 74T 74T 0.023** LS-SVR-CQFOAvs. LS-SVR-CQTS 75T 75T 0.024** LS-SVR-CQFOAvs. LS-SVR-CQGA 78T 78T 0.026** LS-SVR-CQFOAvs. LS-SVR-CQBA 80T 80T 0.027** LS-SVR-CQFOAvs. LS-SVR-FOA 65T 65T 0.018** LS-SVR-CQFOAvs. LS-SVR-QFOA 72T 72T 0.022** TDenotes that theLS-SVR-CQGAmodelsignificantlyoutperformstheothermodels. ** implies thep-value is lower thanα=0.025; * implies thep-value is lower thanα=0.05. Table12.ResultsofWilcoxonsigned-ranktest forGEFCom2014(July). ComparedModels WilcoxonSigned-RankTest T0.025 =81 T0.05 =91 p-Value LS-SVR-CQFOAvs. BPNN 0T 0T 0.000** LS-SVR-CQFOAvs. LS-SVR-CQPSO 73T 73T 0.023** LS-SVR-CQFOAvs. LS-SVR-CQTS 76T 76T 0.024** LS-SVR-CQFOAvs. LS-SVR-CQGA 77T 77T 0.026** LS-SVR-CQFOAvs. LS-SVR-CQBA 79T 79T 0.027** LS-SVR-CQFOAvs. LS-SVR-FOA 65T 65T 0.018** LS-SVR-CQFOAvs. LS-SVR-QFOA 71T 71T 0.022** TDenotes that theLS-SVR-CQGAmodelsignificantlyoutperformstheothermodels. ** implies thep-value is lower thanα=0.025; * implies thep-value is lower thanα=0.05. 4.Discussion Takingthe IDAS2014datasetasanexample,firstly, the forecastingresultsof theseLS-SVR-based modelsareall closer to theactual loadvalues thantheBPNNmodel. ThisshowsthatLS-SVR-based models can simulatenonlinear systemsofmicrogrid loadmore accurately than theBPNNmodel, dueto itsadvantages indealingwithnonlinearproblems. Secondly, inTable 4, the selectedFOAandQFOAalgorithmscouldachieve thebest solution, (γ,σ)= (581,109) and (γ, σ) = (638, 124), with forecasting error, (RMSE = 15.93, MAPE= 2.48%, MAE=15.63)and(RMSE=14.87,MAPE=2.32%,MAE=14.61), respectively.However, thesolution canbefurther improvedbytheproposedCQFOAalgorithmto(γ,σ)= (734,104)withmoreaccurate forecastingperformance, (RMSE=14.10,MAPE=2.21%,MAE=13.88). Similar results couldalso be learnedintheGEFCom2014(Jan.) andtheGEFCom2014(July) fromTables5and6, respectively. This illustrates that theproposedapproach is feasible, i.e.,hybridizingtheFOAwithQCMandchaotic 18