Page - 282 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 1605 Table6.Cont. Years Models BPNN LR SVR Bagging RF 1stSMLE 2ndSMLE 3rdSMLE 2022 4471.17 4915.79 4889.67 4956.50 4915.79 4967.25 4501.68 4602.70 2023 4496.30 4980.78 4954.70 5025.85 4980.78 5056.31 4519.38 4633.88 2024 4555.86 5052.61 5023.62 5095.53 5052.61 5029.93 4536.88 4635.57 2025 4578.78 5128.13 5097.75 5163.26 5128.13 4861.17 4553.77 4628.92 2026 4606.92 5205.40 5170.20 5215.23 5205.40 4615.31 4566.33 4659.36 2017–2026 0.74% 1.39% 1.38% 1.42% 1.39% 0.13% 0.38% 0.44% 1965–2016 2.2% 1.3% 1.8% 1.5% 1.6% 2.0% 2.0% 2.1% Figure 8. Illustrated annual CGR for 10-ahead consumption prediction using (a) single models (b) classicensemblemodels (c)SMLEmodels. We summarized all of the above results in Table 7 and Figure 9. In general, combining the forecastersusingSMLEwill significantly improvethefinalprediction.Generally, fromtheanalysisof theexperimentspresented in this study,wecandrawseveral importantconclusionsas follows: Firstly, theSMLE-basedBPNN-SVRmodelwassignificantlysuperior toallmodels in thisstudyregarding similarity, levelaccuracy,anddirectionaccuracy. Throughperformanceenhancement, theSMLE-based BPNN-SVRoutperformed othermodels at the 1.17 statistical significance level, compared to the best benchmarkmodels SVRandbagging, respectively. Secondly, the prediction performance of the SMLE-basedBPNN-SVR, SMLE-based SVRandSMLE-basedBPNNmodelswere better than thesingleandclassic ensemblemethods. These results indicate that thehybrid, basedonstacking method, can efficiently improve thepredictionperformance in the case ofOC. Thirdly, nonlinear models,withseasonaladjustment,weremoresuitableasbase learners for theensemble topredict the timeserieswithannualvolatility thanlinearmethods,duetopropertiesaboveofOC(i.e.,nonlinear andnon-stationary).However, computationally, thenewmethodconsumedmore timebecauseof its wayofsegmenting inputsandtheuseof theensemble. Fourthly, theaverageannual rateof totaloil demanddecreasedfrom1.8%between1965and2016 to0.91%between2017and2026. Finally,onone 282