Adaptive Residual Compensation Ensemble Models for Improving Solar Energy Generation Forecasting

This letter presents an improved ensemble learning framework for forecasting of solar power generation. A modified ensemble model based on a novel adaptive residual compensation (ARC) algorithm and an evolutionary optimization technique is proposed to improve the forecast accuracy. It is also applied to probabilistic solar power forecasting by using a nonparametric estimator. Extensive experimental results are provided, demonstrating the superiority of the proposed ARC ensemble model over traditional ensemble models, in terms of both deterministic and probabilistic forecasts.