Research on the automatic arrangement method of residential buildings in residential areas based on deep reinforcement learning

Residential architectural plan design must comply with a number of building requirements, including fire prevention and sunlight exposure. At the same time, there are several design options, such as irregular plot outlines and plots with water bodies. In this study, deep reinforcement learning technology is used to investigate a strategy for automatically arranging residential buildings in residential neighborhoods. Building codes are retrieved and used as restrictions in the reinforcement learning framework, with the usefulness, aesthetics, and so on of the planning layout serving as the reward function. Individual buildings act as agents, directly interacting with the scenarios, accumulating various building layouts and corresponding reward values, and autonomously learning strategies based on this, allowing for the automatic arrangement optimization of residential building layouts under multiple constraints. Finally, the proposed method's effectiveness and feasibility are demonstrated by comparative trials.