Deep Reinforcement Learning Integrated PID for Hybrid Adaptive Control Approach in Automatic Voltage Regulation Systems

Automatic voltage regulation (AVR) systems play an important role in maintaining voltage stability, ensuring efficiency, and enhancing the reliability of synchronous generators. Although conventional PID (proportional-integral-derivative) controllers are widely adopted for AVR systems due to their simplicity and robustness, their performance is still limited under dynamic operating conditions. In this paper, this problem is addressed by developing an intelligent controller using a deep reinforcement learning (DRL)-based PID controller, which integrates PID with a reinforcement learning agent to create an adaptive intelligent controller for AVR systems. A comprehensive evaluation of AVR system performance under four control configurations is presented: (1) a conventional PID controller optimised using three recent hybrid optimisation algorithms, (2) a fractional-order proportional-integral-derivative (FOPID) controller tuned with the same hybrid algorithms, (3) a proposed DRL-based FOPID controller, and (4) a proposed DRL-based PID controller. The DRL-based PID controller parameters are adapted by using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, which allows the improvement of generalisation and adaptive learning. The simulation results demonstrate that the proposed DRL-FOPID controller significantly improves performance compared to both the conventional PID and conventional FOPID controllers that were tuned using a hybrid optimisation algorithm. The results emphasise the DRL-based controller in the development of intelligent controllers for AVR systems.

Hali tarjima qilinmagan