Deep Reinforcement Learning with Transformers for Voltage Perception and Control in Low Voltage Distribution Networks

With the rapid development of distributed energy resources and electric vehicles, the operational complexity of distribution networks, particularly low-voltage distribution networks (LVDNs), has significantly increased. Due to limited observability, high stochastic load variations, and complex network topologies, achieving real-time voltage situational awareness and dynamic control in LVDNs remains a major challenge. Traditional methods struggle to quickly converge and provide comprehensive and accurate situational awareness under such conditions. This paper proposes a Transformer and SAC-based method for voltage situational awareness and control in LVDNs. The Transformer network dynamically captures the spatiotemporal dependencies of voltage states, enhancing the perception and regulation of critical nodes, while the Soft Actor-Critic (SAC) algorithm ensures adaptive voltage control in stochastic environments, improving real-time responsiveness and accuracy. Simulation results demonstrate that the proposed method exhibits strong robustness and fast convergence across various complex LVDN scenarios, effectively capturing voltage situational changes and significantly reducing voltage fluctuations and violation risks.