Deep Reinforcement Learning for Stochastic Dynamic Microgrid Energy Management

This paper proposes a deep reinforcement learning (DRL) based algorithm for the stochastic dynamic microgrid energy management. First, we consider AC power flow constraints which makes the problem non-convex and uncertainties in load, renewable generation and real-time electricity price which makes the problem stochastic. Then, the original multi-period optimization problem, which was difficult to deal with, is decomposed into several single-period optimization problems by the proposed deep reinforcement learning based method. Our method can obtain an approximate optimal policy for the original multi-period problem without long-term forecast and optimization or distribution knowledge of the uncertainty by learning from the results of single-period problems. Last, the superiority and effectiveness of the proposed method are demonstrated through several case studies.