Research On Route Optimization Of Multi-Distribution Centers

The prosperity of the national market economy has promoted the rapid development of the logistics industry,and the total revenue of the logistics industry has increased year by year.However,in this development process,the logistics industry has also exposed the disadvantages of high cost,especially high transport costs and low efficiency.It is urgent to control and reduce transport costs,improve the economic efficiency of logistics,and plan vehicle transport routes rationally.Multi-distribution center vehicle routing problem is an extension of VRP.It generally studies that there are multiple depots serving several customers at the same time,and each customer has a certain demand for goods.What is to be solved is to determine which depot lot the customer is serving and to arrange the access path in order to achieve the minimum consumption,the shortest time and other goals.This paper takes the multi-distribution center vehicle routing problem in logistics transportation as the research background,and considers the situation with soft time window to establish a dynamic optimization model.As one of the core problems in transportation organization optimization,VRP is a NP-hard combinatorial optimization problem.Using classical dynamic programming method to solve VRP problem will face the dimension disaster caused by the expansion of the scale of the problem.Therefore,we introduce the approximate dynamic programming algorithm.Approximate dynamic programming algorithm is an area that has developed rapidly in recent years.It has many successful applications in resource allocation,inventory control,supply risk management,urban traffic control,supply chain management,vehicle routing and other practical issues.Firstly,this paper describes the MDVRP problem mathematically,and establish a bi-level programming model considering constraints such as load,mileage,return to the depot and number of visits to customers.The upper model decision maker assigns each customer to each distribution center,and the lower model problem is transformed into a single distribution center vehicle routing optimization problem.The upper model is solved by the improved cluster analysis method,and customers are allocated to distribution centers according to the requirements of distance and time dispersion.The lower problem is modeled as a Markov decision process based on time,action,state,transfer equation and cost function,etc.,and is solved using approximate dynamic programming algorithm.Then,this paper explains in detail the factors affecting the performance of the algorithm,such as initial value,step size and exploration rate,and finally determine the parameter values.The initial value generated by the Rollout algorithm can approach the real value to the greatest extent,so that the algorithm achieves better results.Finally,in order to compare the performance of the algorithm,this paper designs a Simulated Annealing algorithm,one of the classical modern heuristic algorithms,to simulate and analyze the same example.Experiments show that,compared with simulated annealing algorithm,the value function approximation algorithm can explore the path more fully,and get better results in a certain time when the scale is small.However,it is also obvious from the experiment that when the number of customers is large,the state and decision space increase rapidly,the state of the algorithm access and the decision made increase,the calculation time increases,the approximation and update speed of the value function slows down,and the time to obtain the approximate optimal solution is obviously longer.