Sponet: solve spatial optimization problem using deep reinforcement learning for urban spatial decision analysis

Urban spatial decision analysis is a critical component of spatial optimization and has profound implications in various fields, such as urban planning, logistics distribution, and emergency management. Existing studies on urban facility location problems are based on heuristic methods. However, few studies have used deep learning to solve this problem. In this study, we introduce a unified framework, SpoNet. It combines the characteristics of location problems with a deep learning model SpoNet can solve spatial optimization problems: p-Median, p-Center, and maximum covering location problem (MCLP). It involves modeling each problem as a Markov Decision Process and using deep reinforcement learning to train the model. To improve the training efficiency and performance, we integrated knowledge SpoNet. The results demonstrated that the proposed method has several advantages. First, it can provide a feasible solution without the need for complex calculations. Second, integrating the knowledge model improved the overall performance of the model. Finally, SpoNet is more accurate than heuristic methods and significantly faster than modern solvers, with a solution time improvement of more than 20 times. Our method has a promising application in urban spatial decision analysis, and further has a positive impact on sustainable cities and communities.