With the acceleration of urbanization, residents’ demands for personalized and functional residential building layouts are growing. Traditional design methods face challenges such as low efficiency and difficulty balancing multiple objectives when addressing multi-objective optimization. This paper focuses on the generation and multi-objective optimization of residential building layouts using reinforcement learning. First, a basic framework consisting of a layout feature dataset and a multi-dimensional evaluation system is constructed. Next, an improved reinforcement learning model is used to intelligently generate layout elements. A multi-objective optimization algorithm is integrated to form a closed "generation-optimization-screening" loop. Experimental verification demonstrates that this approach significantly improves design efficiency and effectively balances conflicting objectives such as functionality, space, environment, and cost. Experimental survey results show that the design time for a 60㎡ two-bedroom apartment is only 19.3% (28/145) of that required by traditional manual design and 62.2% (28/45) of that required by a single algorithm. The design time for a 90㎡ three-bedroom apartment is 19.4% (35/180) of that required by traditional manual design and 60.3% (35/58) of that required by a single algorithm, highlighting the design efficiency advantage of this method. The proposed method outperforms traditional methods in both multi-objective satisfaction and design efficiency, providing a new approach for intelligent residential layout design.
Dongdong He (Thu,) studied this question.