Research On Vehicle-Cargo Supply And Demand Matching Problem Based On Improved Pointer Network

At present,China’s logistics market has become one of the world’s largest logistics markets.Increased logistics efficiency means better user services and higher corporate benefits.How to quickly and efficiently realize the matching of supply and demand of vehicles and cargos is one of the key points to achieve higher logistics efficiency.In order to cope with the demand for emergencies and to reduce costs,it is necessary to achieve the result of the minimum cost in a short period of time.The issue of vehicle-cargo supply and demand matching mainly includes how the cargo is loaded and how the vehicle is delivered in order.At present,there are two main scenarios in the logistics industry,mainline logistics and end logistics.Mainline logistics has a large single-time transportation volume,fewer destinations and most of them are urban distribution centers.The terminal logistics has a small transportation volume at one time,and the destination is more changeable due to direct connection with users.In this study,a system for vehicle-cargo supply and demand matching that include the terminal and mainline logistics are designed.The design of the system starts with demand analysis and divides the overall system into 4 subsystems.In terms of engineering design,this system uses data layers,business layers and display layers,and proposes the technology stack used by different layers.Among them,the two core subsystems are vehicle-cargo supply and demand matching subsystems in terminal and mainline scenes,and then studied the core algorithm of the supply and demand matching subsystem of the mainline and terminal vehicle-cargo matching,respectively.This study designs a vehicle-cargo matching model for the scenario in which the single transportation volume of vehicles is large and the destination is relatively fixed and few in the mainline logistics scene.The aim is to optimize the distribution of goods in the vehicle to ensure timely delivery to the destination.The model comprehensively considers the vehicle volume,load,travel cost,overdue penalty,and transportation cost of goods,and obtains a profit function.The model takes the city as the destination and uses a greedy algorithm to transform the sorting of vehicles and goods into the solution of vehicle-cargo supply and demand matching problem.The dual-pointernetwork is used to express the decision-making strategy,thereby simulating the Markov decision process,which is used to generate the sorting of vehicles and goods.The policy gradient method is used to train the dual-pointer-network,and the training objective is to maximize the profit function.For the terminal logistics scenario,there are two shortcomings of the traditional Capacitated Vehicle Routing Problem(CVRP)model when applied in this context.Firstly,it is difficult to ensure that the loading capacity of all vehicles is the same in real logistics scenarios.Secondly,if the traditional CVRP model is directly applied to practical scenarios,it may not have feasible solutions.To address the scalability issue of the traditional CVRP model in the real world,this study proposes a more suitable Flexible CVRP model.On the other hand,a Memory Pointer Network(Mem Prt N)method based on reinforcement learning is proposed to solve large-scale CVRP problems that are urgently needed.This method can overcome the exponential growth of complexity to some extent.The model can be trained on small-scale problems and then applied directly to large-scale problems.This generalization ability makes it suitable for solving super-complex Flexible CVRP problems.The experimental simulation section,the optimization ability of the dual-pointer network’s vehicle-cargo matching solution method is comparable to traditional heuristic algorithms for smallscale problems,and surpasses heuristic algorithms for large-scale problems.Moreover,it partly overcomes the problem of time consumption for solving explosion complexity.Results of the solution experiments in the end logistics show that the Mem Ptr N solution method is on par with the OR-Tools solver for small-scale CVRP problems,and slightly worse than recently proposed heuristic algorithms.For large-scale CVRP problems,Mem Ptr N’s solution quality is 19.06% higher than OR-Tools,14.67%better than recently proposed heuristic algorithms,and 59.92% better than the most classical combinatorial optimization algorithms.