Privacy-Enhanced Safe Reinforcement Learning for the Dispatch of a Local Energy Community

Local Energy Community (LEC) has emerged as a viable community-focus framework to enhance local reliability and energy efficiency by integrating different energy sectors and managing local distributed energy resources (DERs). However, the difficulties associated with handling model complexity, along with privacy concerns arising from interactions between energy operators within the LEC, pose challenges for traditional algorithms in achieving coordinated dispatch. To this end, we develop a novel privacy-enhanced, safe, coordinated dispatch framework that integrates reinforcement learning (RL), the perturbation module, and the safety module. The private states of each energy sector within the LEC are concealed by the independent perturbation module before sharing. A central RL agent is then trained on the concealed state space to learn the optimal policy for coordinated dispatch under the complex and uncertain environment. Furthermore, dispatch actions are evaluated and refined by the safety module before the operators execute them. In this way, we can obtain an optimal policy without disclosing any sector’s private state while ensuring the safe operation of the LEC. Extensive experiments are carried out to validate the superior performance and scalability of the proposed method.