Neural Networks for Selective Vehicle Routing Heuristics

Abstract

In the vehicle routing problem (VRP), there is a collection of stops with known demands for service, and a fleet of vehicles with known capacities available to serve the stops. The problem is to assign the stops to vehicles, and specify routing sequences for each vehicle so that total distance is minimized. Many heuristic algorithms for the VRP have been developed over the last 25 years. Given a particular vehicle routing problem instance, we address here the issue of selecting which heuristic algorithm to apply to the problem. A modular system of neural networks is the knowledge base that is used to make the selection. Three types of multiple-layer neural networks are involved, and all are trained with the back-propagation learning paradigm. After training, the neural network system was applied to a collection of new problems not used in training. The system was very successful in identifying the optimal algorithm to apply to the new problems. We conclude that neural networks have considerable promise for use in vehicle routing model management systems. INFORMS Journal on Computing, ISSN 1091-9856, was published as ORSA Journal on Computing from 1989 to 1995 under ISSN 0899-1499.