Adaptive Tuning of Virtual Synchronous Machine Parameters (J and D) Using Deep Reinforcement Learning

Abstract—Virtual Synchronous Machines (VSMs) emulate the inertial and damping behavior of conventional synchronous generators in inverter-based power systems. Conventional VSM implementations use fixed values of inertia constant (J) and damping coefficient (D), which lead to sub-optimal performance under dynamic grid conditions. This paper proposes an intelligent adaptive tuning framework for J and D using Deep Reinforcement Learning (DRL), employing Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC) algorithms, validated via a MATLAB/Simulink–Python co-simulation environment. Simulation results across five disturbance scenarios demonstrate that the SAC-based controller achieves up to 48.6% reduction in maximum frequency deviation, 41.2% improvement in settling time, and 46.8% reduction in ROCOF compared to fixed-parameter VSM. The complete MATLAB and Python implementation details and simulation graphs are presented, confirming the viability for next-generation grid-forming inverter control. Index Terms—Virtual Synchronous Machine, Deep Reinforcement Learning, PPO, SAC, Adaptive Control, Inertia Emulation, Frequency Regulation, MATLAB/Simulink, Python, Grid-Forming Inverter.