Load Frequency Control of Interconnected Microgrid Enhanced by Deep Reinforcement Learning SAC-PID

In recent years, the incorporation of sustainable energy resources such as wind power has had a significant impact on the stability of microgrids. In this context, our research introduces a proficient method for load frequency regulation utilizing deep reinforcement learning (DRL). Firstly, a two-area interconnected microgrid frequency control model is constructed, including wind power generation, reheat steam turbines, generators and loads. Secondly, the proposed method uses a soft actor critic (SAC) with a double Q-network to tailor the parameters of the proportional integral derivative (PID) controller, to train a DRL agent to obtain optimal controller parameters for this two-area interconnection, and we used the area control error (ACE) as the controller’s input to reduce fluctuations in both grid frequency and interconnection line power flows. Furthermore, we designed a reward function via the square of the ACE to obtain the best-suited parameters for the controller. Finally, the trained controllers are tested under step load disturbance and random load disturbance, verifying that the designed SAC-PID controller provides improved frequency control performance for microgrids compared with the classical PID controller and the PID controller optimized through the particle swarm optimization (PSO) algorithm.