Volt/Var Control for Three-Phase Unbalanced Distribution Network Based on Trust Region Safe Reinforcement Learning

With the widespread integration of renewable energy, power flow in the system has become extremely complex and variable. This not only exacerbates the operational safety issues of distribution networks but also intensifies the three-phase unbalance situation. The traditional volt/var control (VVC) model is facing significant challenges such as high-dimensional nonlinearity and low efficiency. To address these problems, this paper proposes a volt/var control for three-phase unbalanced distribution network based on trust region safe reinforcement learning. Firstly, a model is constructed based on the three-phase linear power flow matrix. Then it is transformed into a Markov Decision Process (MDP) to overcome the computational burden. Secondly, a trust region construction method based on the Clip mechanism is introduced to ensure stable policy gradient updates and computational efficiency. Further, the Lagrange multiplier is introduced in the trust region optimization to convert the node voltage safety boundary into a cost function, establishing a distribution network safety reinforcement learning (SDRL) model, which limits the output of dangerous action. Finally, through case studies, it is verified that the proposed method can effectively mitigate three-phase unbalance, enhance online decision-making efficiency, and strictly guarantee the safe operation of distribution networks, demonstrating significant feasibility and superiority.