Research on Reactive Power Optimization of Active Distribution Network Based on Deep Reinforcement Learning

The installed proportion of renewable energy is constantly increasing. Due to the intermittency and volatility of renewable energy output, its connection to the power grid can easily affect voltage, leading to voltage violations and safety issues in grid operation. At the same time, considering fast regulation characteristics of Static Var Generator (SVG), a method utilizing deep reinforcement learning is introduced for controlling active distribution network reactive power and voltage. This method converts reactive power and voltage control problem of the power grid under renewable energy access conditions into a Markov Decision Process (MDP), and then constructs a reactive power and voltage control model under renewable energy access. Based on the actual operating environment requirements of the power grid, constraint conditions are proposed, and the model is solved using Deep Deterministic Policy Gradient (DDPG) algorithm. Finally, simulations were performed for improved 33 node instances. The effectiveness of the reactive power and voltage optimization control strategy given in this paper is verified by comparing the reactive power and voltage optimization control effects under different levels of renewable energy access conditions.