Monthly Renewable Energy Forecasting Based on Multi-Model Fusion Techniques

This paper introduces a novel approach to fore-casting renewable energy outputs by employing a model fusion methodology that integrates four predictive algorithms: ARIMA, XGBoost, GRU, and conventional statistical algorithms. This method not only leverages the strengths of each model to minimize generalization errors but also boosts prediction accuracy significantly. Distinctively, it integrates economic considerations into the renewable integration forecasts by incorporating a specific term and its coefficient in a linear regression model, with the weight of this term determined using a Generalized Additive Model. The methodology involves a critical step where both meteorological and energy output data are aligned to a monthly dimension to ensure consistency in temporal analysis. Moreover, the study differentiates between scenarios with and without distributed solar systems, which allows for targeted and accurate model application. Numerical tests on both wind and solar forecasts have shown a prediction accuracy over 82%.