A reinforcement learning-based online learning strategy for real-time short-term load forecasting

Real-time Short-Term Load Forecasting (STLF) is crucial for energy management and power system operations. Conventional Machine Learning (ML) methodologies for STLF are often challenged by the inherent variability in energy demand. To tackle the challenge associated with inherent variability, this paper presents a novel Reinforcement Learning (RL)-enhanced STLF method. Different from conventional methods, our method dynamically improves the STLF model by selecting optimal training data to capture recent power usage trends and possible variations in demand patterns. By doing so, our method can significantly reduce the impact of unforeseen fluctuations in real-time forecasting. In addition to the novel RL-enhanced STLF method, we propose a comprehensive evaluation framework, encompassing three key dimensions: accuracy, runtime efficiency, and robustness. Tested on three distinct real-world energy datasets, our RL-enhanced method demonstrates superior forecasting performance across three evaluation metrics by achieving accurate and robust predictions in real-time under varying scenarios. Furthermore, our approach provides uncertainty bounds for practical predictions applications. These results underscore the significant advancements made by our RL-based method in forecasting precision, efficiency, and robustness. We have made our algorithm openly accessible online to promote continued development and advancement of STLF methods.