Research On Large-scale Multi-depot Vehicle Scheduling Method With Fluctuant Demand

Logistics distribution is an important part of the modern enterprise logistics, anddistribution network optimization is more and more important in the logistics networkintegration. With the expansion of companies scale, companies have severalproduction bases or distribution center. Meanwhile, customers are in different regional.Thus, there is a large scale distribution network including multi-producers andmulti-customers. In the multi-to-multi large-scale distribution network, the demandcharacteristics of small customers are low quantities and high frequencies, whichusually bring the waste of transportation resources, and higher LTL (less than a truckload) transportation cost. So, how to respond to fluctuant customer orders quickly,integrate the logistics resources, and improve the service level are major challengesthat companies are facing. Based on the background, the paper analyzes thesingle-period and multi-period multi-depot vehicle scheduling problem.To solve the single-period vehicle routing problem, the two-stage fuzzyclustering algorithm is proposed. In the first stage, k-means method is used to clustercustomers into static groups according to customers’ geographical locations. Then,based on multiple attributes of customer orders, the fuzzy cluster technique is appliedto generate dynamic clusters. Further, considering fluctuant customer orders, themulti-period vehicle routing problem is developed. On the basis of two-stage fuzzyclustering method, the three-phase solution methodology framework is proposed.Firstly, static clustering is conducted to generate static groups, and then the servicepriorities of each producer serving the static customer groups are ranked according tothe distance performance. Secondly, with the continuous arrival of customer orders,the order processing determines which orders are served in the current period, andcustomer orders of the period T are determined. Further, the fuzzy clusteringtechnique is applied to conduct dynamic clustering based on the customer orderattributes of the period T. Similarly, the service priorities of generated dynamiccustomer groups will be ranked according to the time attributes of orders. Thirdly, theimproved GA is designed to solve the VRP by changing the selecting operator and thecrossover operator. The stochastic simulation experiment and the real case show theproposed algorithm is efficient.