Study on a General Reinforcement Learning Simulation Platform for Regional Power Grid Control Requirements

Due to the powerful state awareness and optimization decision-making capabilities of Deep Reinforcement Learning (DRL), which avoid the traditional processes of physical modeling and formula solving, DRL has become one of the important means to solve new power system dispatch problems. However, the lack of real-world power grid simulation environments has hindered the application of DRL models in actual regional power grids. This paper proposes a deep reinforcement learning simulation platform suitable for regional power grids, which takes into account constraints such as unit operation, grid safety, and power balance. The platform establishes an action space consistent with the operational adjustments of equipment in a real power grid, enabling efficient interaction and accurate feedback between intelligent agents and the grid environment. The validation results show that the proposed simulation platform can effectively interface with actual power grids and DRL agents, achieving active power optimization in regional grids. The results provide foundational support for the practical application of deep reinforcement learning in power grids.