Seite - 131 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 3283 Table6showstheMAPEofrandomforest foreachclusterunderdifferentmTryandthepredicted resultswith thebest accuracy aremarked inbold. Since the input variable is 9, sqrt and log2 are recognizedas3andtheresultsare thesame.Wechoose thesqrt that is commonlyused[16,43]. Table6.MAPEresultsof randomforest. Cluster# NumberofFeatures Auto sqrt log2 ClusterA 3.983 3.945 3.945 ClusterB 4.900 4.684 4.684 ClusterC 3.579 3.266 3.266 Figure6a–cshowtheuseof forestsof trees toevaluate the importanceof features inanartificial classificationtask. Thebluebarsdenotethefeature importanceof theforest,alongwiththeir inter-trees variability. In thefigure,LSTM,whichrefers to theLSTM-RNNthat reflects the trendofdayof the week,has thehighest impactonthemodelconfigurationforall clusters.Other featureshavedifferent impacts,dependingonthecluster type. (a) Cluster A (b) Cluster B Figure6.Cont. 131