Very Large Scale Vehicle Routing with Time Windows and Stochastic Demand Using Genetic Algorithms with Parallel Fitness Evaluation

This paper deals with a real-life vehicle routing problem concerning the distribution of products to customers. A non-homogenous fleet of trucks with limited capacity and allowed travel time is available to satisfy the stochastic multiple product demand of a set of different types of customers with earliest and latest time for servicing. The objective is to minimize distribution costs while maximizing customer satisfaction and respecting the constraints concerning the vehicle capacity, the time windows for customer service and the driver working hours per day. A model describing all these requirements has been developed as well as a genetic algorithm to solve the problem. High Performance Computing has been used to allow the pursuit for a near-optimal solution in a sensible amount of time, as the parallel chromosome fitness evaluation counterbalances the increased size and complexity of the problem.