Secondary Frequency Control of Islanded Microgrid Based on Deep Reinforcement Learning

With the large-scale integration of distributed generation into microgrids, the volatility of renewable energy generation and system random disturbances pose significant threats to the frequency stability and operational control of islanded microgrids. To address this, a secondary frequency control method based on deep reinforcement learning is proposed. The droop control characteristics of islanded microgrids are analyzed, and a secondary frequency controller structure based on deep Q-Networks is presented. The frequency deviation is used as the state input variable, and the design of the state space, action space, reward function, neural network, and hyperparameters in the deep Q-Networks algorithm is carried out. The reward function balances the goals of frequency recovery and power allocation among distributed energy resources , ensuring consistency in action selection among the intelligent agents. An offline learning process is used to train the deep reinforcement learning-based secondary frequency controller. A simulation model of the islanded microgrid is developed in Matlab/Simulink, and multiple disturbance scenarios are tested to validate the controller's performance. The results show that, compared to traditional PID control and Q-Learning-based controllers, the proposed method achieves more stable secondary frequency control and adapts to coordinate the power allocation of distributed generation units according to their capacities, ensuring the stable operation of the system.