A Reinforcement Learning based Power System Stabilizer for a Grid Connected Wind Energy Conversion System

When connecting renewable sources wind turbines to a power grid, low frequency oscillations caused by wind turbines may threaten the stability of the entire electrical power system. Power system stabilizers (PSSs) are used to damp the low frequency oscillations. However, these PSSs are usually designed based on small-signal models around a fixed wind speed and their performances could be degraded when wind speed varies in a real-time pattern. In this paper, a reinforcement learning (RL) based power system stabilizer is designed for a grid-connected double-fed induction generator (DFIG) based wind system to enable the online optimization of control gains when wind speed varies. In specific, the Q-learning based PSS is designed in the rotor-side controller of the DFIG based wind system. In this method, the active power change is defined as the state, and the control output of the rotor side controller (RSC) is used as the action. A grid-connected DFIG based wind system is simulated and the results show that the Q-learning based PSS can quickly adjust the control parameters online and damp the low frequency oscillation under a time-varying wind speed condition.