Deep Reinforcement Learning-Based Mitigation of Subsynchronous Oscillation Due to Inverter-Based Resources in Weak Grids

The classification of power system stability has been revised and extended to include new stability issues. Two new categories are introduced in this updated classification. Among them, resonance stability specifically deals with subsynchronous oscillations (SSOs) that occur in various power electronic devices. This paper presents a simulation of the SSO by inverterbased resources (IBRs) under varying power system conditions considering grid reactance and the active power variability of IBRs. In scenarios where the grid reactance is high, the power system exhibits a low short-circuit ratio (SCR) and reduced damping, typical of a weak grid. A mitigation control model for SSO using deep reinforcement learning (DRL) has been developed to address these issues. This model functions as a damping controller and is trained on various SSO cases to effectively control varying frequency oscillation in the power system. The DRL-based mitigation model has been validated against untrained SSO scenarios and has successfully mitigated SSO within seconds.