Real-Time Optimal Dispatch of Microgrid Based on Deep Deterministic Policy Gradient Algorithm

Microgrid is a typical form of smart grid, which integrates distributed renewable energy sources (DRESs) to eliminate on-grid impact and achieve efficient applications. However, the real-time dispatch of microgrids is faced with challenges resulted by uncertainties of DRESs and nonlinear constraints. To deal with this problem, a real-time optimal dispatch strategy based on deep reinforcement learning (DRL) is proposed in this paper. Specifically, a detailed battery model is formulated, which considers the power loss during the discharge and charge process. Next, we formulate the real-time optimal dispatch problem as a Markov decision process, which consists of state, action, state transition, immediate reward and the action-value function. A Deep Deterministic Policy Gradient (DDPG) algorithm framework is designed to obtain the optimal dispatch strategy through off-line training and on-line decision making. Finally, the case study is performed based on real historical data of microgrid operation, the simulation results demonstrate the effectiveness and feasibility of the proposed method.