Ensembling methods for countrywide short-term forecasting of gas demand

Gas demand is made of three components: Residential, Industrial, and Thermoelectric Demand. Herein, the one-day-ahead prediction each component studied, using Italian data as a case study. Statistical properties relationships with temperature are discussed, preliminary step for an effective feature selection. Nine "base forecasters" implemented compared: Ridge Regression, Gaussian Processes, Nearest Neighbours, Artificial Neural Networks, Torus Model, LASSO, Elastic Net, Random Forest, Support Vector Regression (SVR). Based on them, four ensemble predictors crafted: simple average, weighted subset SVR aggregation. We found that perform consistently better than base ones. Moreover, our models outperformed Transmission System Operator (TSO) predictions in two-year out-of-sample validation. Such results suggest combining may lead to significant performance improvements gas forecasting.