Enhanced DQN Based Parameter Optimization for VSG Control Under Pulsed Power Loads

This paper presents a novel approach to virtual synchronous generator (VSG) control through the implementation of reinforcement learning, specifically utilizing a Deep Q-Network (DQN) algorithm. The proposed methodology addresses the inherent limitations of conventional VSG control strategies, particularly in managing dynamic conditions of pulsed power loads (PPLs). A enhanced DQN framework is developed to dynamically optimize critical VSG parameters—virtual inertia and damping coefficients. Simulation results show that the DQN-based approach effectively improves system stability and response speed compared to traditional fixed-parameter VSG controllers. The findings highlight the potential of reinforcement learning in advancing VSG control strategies for modern power systems with stringent stability requirements.