Q-Learning-Based Approach for Mitigating Peak Shaving Impact on Total Demand Load Forecasting

Effective management of peak energy demand is crucial for ensuring grid stability and minimizing operational costs in modern utility systems. However, peak shaving strategies, commonly employed by utilities to mitigate peak demand, can inadvertently introduce distortions in load patterns, thereby compromising the accuracy of load forecasting models. In this paper, we address this challenge by proposing a novel approach leveraging reinforcement learning techniques, specifically Q-learning, to mitigate the adverse effects of peak shaving on load forecasting accuracy. Our approach involves integrating Qlearning at a higher level within the forecasting framework to dynamically adjust forecasting values and optimize the trade-off between peak reduction and forecasting accuracy. We evaluate the effectiveness of our proposed approach through the publicity Australian dataset. Results demonstrate that using our approach, the Mean average percentage error can significantly be reduced from 4.84% to 4.62%. Our work contributes to advancing the state-of-the-art in load forecasting and provides valuable insights for utilities seeking to optimize peak demand management strategies.