ABSTRACT The increasing integration of photovoltaic (PV) systems into interconnected power networks significantly impacts transient stability by reducing system inertia and amplifying low‐frequency inter‐area oscillations. This study proposes a reinforcement learning (RL)‐based control strategy to enhance the inter‐area stability in an interconnected power system with PV penetration, focusing on a tailored value‐based approach. The proposed method introduces a novel systematic RL state selection process based on participation factor and residue analysis to ensure observability and controllability of critical oscillatory modes. Hyperparameter optimization is carried out using Taguchi orthogonal arrays which enables efficient identification of optimal learning configurations with substantially fewer trials. Additionally, an action signal smoothing mechanism is introduced to produce continuous‐like control signals suitable for static VAR compensator (SVC) actuation, effectively bridging the discrete nature of the critic‐only agent with the continuous control requirements. The controller's performance is evaluated under severe fault scenarios and varying PV penetration levels. Comparative studies with continuous actor‐critic RL methods, namely deep deterministic policy gradient (DDPG) and soft actor‐critic (SAC), demonstrate that the proposed approach reduces overshoot by 55.34 and 22.58 relative to DDPG and SAC, and improves settling time by 6.79 and 12.6, respectively, underscoring its practical viability for dynamic power system stabilization.
Dewantoro et al. (Thu,) studied this question.