Multi-Objective evolutionary algorithms for vehicle routing problems

The Vehicle Routing Problem, which main objective is to find the lowest-cost set of routes to deliver goods to customers, has many applications in transportation services. In the past, costs have been mainly associated to the number of routes and the travel distance, however, in real-world problems there exist additional objectives. Since there is no known exact method to efficiently solve the problem in polynomial time, many heuristic techniques have been considered, among which, evolutionary methods have proved to be suitable for solving the problem. Despite this method being able to provide a set of solutions that represent the trade-offs between multiple objectives, very few studies have concentrated on the optimisation of more than one objective, and even fewer have explicitly considered the diversity of solutions, which is crucial for the good performance of any evolutionary computation technique. This thesis proposes a novel Multi-Objective Evolutionary Algorithm to solve two variants of the Vehicle Routing Problem, regarding the optimisation of at least two objectives. This approach incorporates a method for measuring the similarity of solutions, which is used to enhance population diversity, and operators that effectively explore and exploit the search space. The algorithm is applied to typical benchmark problems and empirical analyses indicate that it efficiently solves the variants being studied. Moreover, the proposed method has proved to be competitive with recent approaches and outperforms the successful multi-objective optimiser NSGA-II.