Two-Stage Location and Delivery Routing System for Catering Enterprises Using Deep Q-Networks

The location and transportation planning of catering stores are critical decisions that affect operating cost, service quality, and long-term competitiveness. It is therefore important to optimize both site selection and raw material distribution. This study proposes a decision support system designed to help small and medium-sized catering business identify optimal store locations and plan efficient raw material delivery routes. The system consists of two stages. In the location stage, the Weiszfeld algorithm is applied to generate high-quality candidate sites, followed by a Deep Q-Network (DQN) algorithm to select a number of best store locations. In the routing stage, a Q-learning algorithm is applied to determine the optimal delivery route for raw materials. To evaluate the system performance, Starbucks in SIP is used as a case study, where real operational data and population heatmap data are used to generate the testing data. The experimental results show that our model significantly reduces both operational costs and transportation distances. The optimized store layout not only improves customer service coverage but also establishes an efficient logistics network, providing valuable insights for managers to optimize store placement and delivery strategies.