An advanced real-time dispatching strategy for a distributed energy system based on the reinforcement learning algorithm

A desirable dispatching strategy is essentially important for securely and economically operating of wind-thermal hybrid distribution systems. Existing dispatch strategies usually assume that wind power has priority of injection. For real-time control, such strategies are simple and easy to realize, but they lack flexibility and incur higher operation and maintenance (O&M) costs. This study analyzed the power dispatching process as a dynamic sequential control problem and established a Markov decision process model to explore the optimal coordinated dispatch strategy for coping with wind and demand disturbance. As a salient feature, the improved dispatch strategy minimizes the long-run expected operation and maintenance costs. To evaluate the model efficiently, a Monte Carlo method and the Q-learning algorithm were employed to the growing computational cost over the state space. Through a specified numerical case, we demonstrated the properties of the coordinated dispatch strategy and used it to address a 24-h real-time dispatching problem. The proposed algorithm shows high efficiency in solving real-time dispatching problems.