Vehicle Routing Optimization For Milk-run Logistics And Implementation Of Vehicle Prediction System

Milk-run is a logistics strategy widely applied in the logistics of auto parts.It ensures that vehicles complete pickup and delivery of small batches and high-frequency on time.With the development of Lean Production,both logistic companies,auto and other manufacturing companies have large practice demands in milk-run.The optimization of logistics distribution routing with small batches and high frequency is still in the stage of exploration and development.Many scholars' researches on this problem ignore some practical constraints,making milk-run has not achieved its expected results.The most of milk-run vehicle routing algorithms still cannot solve the problems of low vehicle loading rate and high logistics cost.At present,most studies prefer vehicle routing problem(VRP)based on milk-run,while the prediction of vehicle demands is rarely studied.The prediction of vehicle demands is still relying on artificial experience to make forecast,which did not match the high-information level of VRP.Milk-run VRP of this dissertation can be reduced to an NP-hard problem.The research on it can not only provide theoretical foundation for Combinatorial Optimization,but also apply the theoretical research results to actual life.Simultaneously,the design and implementation of the vehicles prediction system can make up for shortcomings of few number of milk-run scheduling system.On the fact that the current logistics has the problems of low vehicle loading rate and high logistics cost,this thesis introduces the multiple orders constraints that actually exist in logistics,and studies the milk-run vehicles routing problem based on miltiple orders.Then this dissertation designs a two-layer search algorithm based on neighborhood search for the vehicle routing problem.The algorithm introduces a two-layer search structure,in which the inner layer performs large neighborhood search on the current solution by running three large neighborhood search operators in parallel to optimize the routes.Then this structure transmits the inner solution to the outer layer for small neighborhood search.Finally,the structure through the continuous iterative search of inner and outer layers to obtain approximate optimal solutions or even optimal solutions.The dissertation compares the proposed algorithm with 1D_LNS algorithm on 56 pdp instances and modified mopdp instances with 100 customers.The first group of results shows that the proposed algorithm is suitable for multiple types of location distribution and has strong ability to get optimal solution.The second group of results shows that the algorithm is better than 1D_LNSon mopdp instances,which indicates that the proposed algorithm has great advantages in solving small batch and high frequency delivery situation.On the same time,it further reflects the superiority of the milk-run.Aiming at the situation that the vehicle scheduling problem of milk-run is rarely studied,this thesis explores the problem of milk-run vehicles prediction.The dissertation designs a vehicle demand prediction system based on LSTM algorithm.The system implements data manage,prediction model and corresponding interface display operation functions by Database,Python and Web technologies.The system mainly provides users with vehicle demand forecast services and forecast accuracy monitoring services.