Structural Combination of Seasonal Exponential Smoothing Forecasts Applied to Load Forecasting

This article draws from research on ensembles in computational intelligence to propose structural combinations of forecasts, which are point forecast combinations that are based on information from the parameters of the individual models that generated the forecasts. Two types of structural combination are proposed which use seasonal exponential smoothing as base models, and are applied to forecast short-term electricity demand. Although forecasting performance may depend on how ensembles are generated, results show that the proposed combinations can outperform competitive benchmarks. The methods can be used to forecast other seasonal data and be extended to different types of forecasting models.