Comparative Analysis of Machine Learning Models for Short-Term Renewable Energy Forecasting: A Comprehensive Study

Abstract

The inherent intermittency of renewable energy sources (RES) poses significant challenges to their increasing integration into contemporary power grids.  Ensuring grid stability, streamlining dispatch processes, and promoting market integration all depend on precise energy generation forecasting.  Six different statistical modelling and machine learning techniques for short-term renewable energy forecasting are thoroughly compared in this paper.  Among the models assessed are Random Forest, a custom hybrid Ensemble model, the Prophet time-series model, Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM) networks, and a conceptual framework for Reinforcement Learning (RL).  Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R-squared (R²) score are used to systematically evaluate these models using a standardised dataset of renewable power generation. The experimental results show that the hybrid Ensemble model offers a better balance of high accuracy (MAE: 1.909) and enhanced robustness by reducing the risks of individual model failure, even though XGBoost achieves the highest raw accuracy with the lowest MAE of 1.814.  As academic researchers and industry practitioners navigate the shift to a sustainable energy future, the findings highlight the importance of integrating various modelling philosophies to produce forecasting systems that are dependable and resilient.