XGBOOST ALGORITHM FOR ORECASTING ELECTRICITY CONSUMPTION OF GERMANY

Stability requires energy demand prediction. We train and test 24-hour German load forecasting models. ENTSO-E Transparency Platform data covered European generation, transmission, consumption. It uses instead of PJM for the eastern US, adds holidays lag features to XGB model, benchmarks with a linear model random forest. Grid search CV refines final model. National RMSE is 1740MW, which suitable gradient boosting H-24 H-48 most important this job. Weekends help, but less. Regional holidays, average temperatures, characteristics could improve (beyond H-48).