Adaptive Control for Virtual Synchronous Generator Parameters Based on Soft Actor Critic.

Area of Science:

• Electrical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Traditional virtual synchronous generators (VSGs) face challenges with active power and frequency oscillations.
• Existing control methods may lead to prolonged system transients or suboptimal performance.

Purpose of the Study:

• To introduce a model-free reinforcement learning (RL) optimization method for VSG control.
• To address and mitigate active power and frequency oscillations in VSGs.
• To improve transient response and reduce setting time.

Main Methods:

• Utilized a reinforcement learning agent with active power and frequency response as state inputs.
• Adjusted virtual inertia and damping coefficients for optimal VSG performance.
• Incorporated a setting-time term into the reward function alongside power and frequency deviations.
• Employed the Soft Actor Critic (SAC) algorithm for policy optimization.

Main Results:

• The proposed RL-based method effectively suppresses active power and frequency oscillations.
• Significantly reduced system setting time compared to other approaches.
• Demonstrated superior convergence and avoided policy overestimation bias through the SAC algorithm.

Conclusions:

• The model-free RL approach offers an effective solution for VSG stability enhancement.
• The inclusion of a setting-time term in the reward function improves transient performance.
• The SAC algorithm provides robust and efficient optimization for VSG control.