Research on Extended CVRP Problem Based on Memory Pointer Network

The traditional CVRP model is often too simple compared to the problems in actual production and life. CVRP problems with strict feasible solutions are difficult to appear in actual production life. Therefore, it is more practical to study a broader, a little slack, super-large-scale CVRP problem. First, this paper proposes an extended CVRP model for the above problems. The model takes minimizing the driving distance of all vehicles and maximizing the number of users served as the optimization goal, and the vehicle carrying capacity is the constraint condition. The above CVRP problem is generalized to make it closer to actual production and life. Then this paper proposes a Memory Pointer Network (MemPtrN) for how to solve the super-large-scale CVRP problem. The network expresses the matching strategy between the vehicle and the user, so as to solve the CVRP problem. The solution method has good versatility. The Memory Pointer Network can be trained on small-scale problems and then used directly on large-scale problems, and good solution results can still be obtained. Compared with OR-Tools, Google's operations research solver, the quality of solutions on small-scale problems is the same, and the quality of solutions on large-scale problems is better than this solver.