Parameter Optimization Strategy for Wind Generator Converters Based on Actor-Critic Framework

With the increasing power generation capacity of renewable energy connected to the grid, the issue of power system sub-synchronous oscillation caused by power electronic equipment becomes prominent, posing new challenges to the safe and stable operation of the power system. The commonly used analysis methods based on linearized models encounter the problem of dimensionality disaster when dealing with large-scale target power systems. To address the above issues, in this paper we propose an optimization strategy for the control parameters of wind generator converters based on the principle of reinforcement learning and utilizes the actor-critic learning framework. With the collected operating status data of permanent magnet synchronous generator (PMSG), a reinforcement learning agent (Agent) is trained to evaluate the PMSG's operating status and stability based on the proposed return function and determines an optimal strategy for wind generator converter parameters optimization. The agent obtained by this training method can optimize the parameters of wind generator converters based on time-domain sampling data, effectively suppressing the oscillation phenomenon induced by the converters. Moreover, it is not only able to optimize but also enhance the stability of the power system without the need for a linearized analysis model. The optimization strategy demonstrates superior optimization performance in the presence of noise interference in the sampled data, as confirmed through verification.