Page - 170 - in Short-Term Load Forecasting by Artificial Intelligent Technologies

Energies2018,11, 2038 Figure3.Actualandforecasting load(kWh) foraweek(9–15May2016). Finally, in thissection,weanalyze theadecuacyof the forecastingmethodXGBoost for thecase studywhen consideringdifferent predictionhorizons (1 h, 2 h, 12h, 24h, and48h). In all cases, weselectedthesameparameters: subsample=0.5,max_depth=6,eta=0.05andnrounds=1700.Accuracy results for the trainingandtestdatasetsaregiveninTable8aswellas themost importantpredictors in eachcase. 170