“The research team proposed a distributed layer partitioning and grid planning method based on Q-learning. In response to the problems of directed and undirected transformation, grid connectivity, and classification considering multiple factors that arise during the training process, the Reward and Agent were improved according to the different requirements of the two stages of hierarchical partitioning and grid planning, making the model consider elements more comprehensively. Simulation verification was conducted using actual power grid data, providing a new idea and technical means for the planning and design of centralized distributed distribution networks.”