This paper presents a comparative study of optimization strategies for solving the Capacitated Vehicle Routing Problem (CVRP) in real-world beverage distribution logistics. Two solution approaches are investigated: a two-phase method that combines customer clustering with Traveling Salesman Problem (TSP)-based intra-cluster routing, and an adaptive metaheuristic approach that integrates Deep Q-Networks (DQN) with evolutionary optimization. The proposed methods are evaluated using a real-world dataset comprising 58 customers in Jeddah. Experimental results show that the standard Genetic Algorithm (GA) achieved the minimum total travel distance of 865.13 km. The two-phase K-Means–TSP approach produced a competitive solution with a total distance of 970.90 km. In contrast, the DQN-enhanced Adaptive Genetic Algorithm (AGA) reduced the required feet size to 18 vehicles while covering a total distance of 916.66 km. These findings indicate that reinforcement learning–based adaptation can effectively optimize fleet utilization, although additional training and tuning are required to further improve distance minimization performance.
Alzahrani et al. (Thu,) studied this question.