Reactive Power Optimization Strategy for Power Grid with High Proportion of Renewable Energy Based on Reinforcement Learning Algorithm

As the high proportion of renewable energy grid connection leads to intensified grid voltage fluctuations, traditional reactive power optimization methods have obvious deficiencies in dynamic regulation capability and real-time performance, making it difficult to cope with the voltage over-limit and network loss problems caused by the randomness of wind and solar output. To this end, this paper proposes a reactive power optimization strategy based on the deep deterministic policy gradient (DDPG) algorithm, constructs a hierarchical reinforcement learning framework that includes 12-dimensional state spaces such as node voltage, renewable energy output, and load rate, and continuous/discrete action spaces such as SVG and capacitor banks, designs a composite reward function that integrates voltage deviation, network loss cost, and equipment action penalty, and implements minute-level online training through the interactive interface between the power grid simulation platform PSS/E and Python. Tests in the improved IEEE 33-node system show that this method reduces the voltage deviation rate to 0.41 % and the average network loss to 123.2 kW, verifying that the strategy can effectively coordinate the collaborative control of new energy stations and traditional reactive equipment, providing a new way to solve the problem of dynamic reactive power optimization in high-penetration power grids.