The present work considers the use of metaheuristics and parallel computing to solve a real problem of vehicle routing involving a heterogeneous fleet, time windows and split deliveries, in which customer demand can exceed vehicle capacity. The problem consists of determining a set of economical routes that meet each customer's needs while still being subject to all the constraints. The strategy adopted to solve the problem consists of an adaptation of the constructive heuristics proposed by Clarke & Wright (1964) as the initial solution. More sophisticated algorithms are then applied to achieve improvements, such as parallel genetic algorithms supported by a cluster of computers. The results indicate that the basic constructive heuristic provides satisfactory results for the problem, but that it can be improved through the use of more sophisticated techniques. The use of the parallel genetic algorithm with multiple populations and an initial solution, which presented the best results, reduced the total operational costs by about 10% compared with the constructive heuristic, and by 13% when compared with the company's original solutions.
vehicle routing problem; time windows; split deliveries; metaheuristics
