Deep Reinforcement Learning for Voltage Stability: DDQN-Based Control with Real-Time Hardware Validation

Abstract

This study proposes an autonomous voltage control framework based on the Double Deep Q-Network (DDQN) algorithm to enhance voltage stability in power distribution systems with high renewable penetration. The proposed controller learns adaptive voltage regulation policies by interacting with dynamic grid environments and observing voltage deviations, power flows, and generator-load dynamics. The method is evaluated on both IEEE 14-bus and IEEE 124-bus test systems and benchmarked against state-of-the-art DRL agents, including DQN, PPO, DDPG, and SAC. The results demonstrate that DDQN provides a favorable balance between control performance and computational efficiency, particularly in large-scale systems. Furthermore, the proposed approach is implemented and validated using an OPAL-RT real-time hardware-in-the-loop platform, confirming its practical applicability for real-time voltage control in next-generation smart grids.