Route Learning in Vehicle Routing with Pickup and Delivery

Abstract

We consider a multi-cluster vehicle routing problem with pickups and deliveries. A cluster consists of a set of nodes with requests for shipment. A single depot serves all requests in each cluster by means of capacitated trucks which pick up a consignment and deliver the same at nodes with delivery requests. A truck returns to its base depot after its consignment is exhausted or when the requirement of all nodes in a cluster is fully satisfied. Because of truck capacity constraints, multiple trips are required to serve all nodes. A truck can choose from among several alternate routing strategies. Trucks are evaluated for efficiency depending upon deliveries made and length of route covered. The computed efficiency depends on the routing strategy adopted by the truck. We provide details of a Genetic Algorithm for route selection and explain how we can examine whether any routing strategy does significantly better than others. The model and the experiments outlined are representative of a large number of combinational variations possible. The VRPPD1 transportation requirement is common to many companies and the results of these experiments can be used to determine optimal routing strategies for their eet of vehicles. Given the vast expenditure most company's incur in transshipment logistics, this has the potential to significantly reduce costs and increase a company's profitability.