By Pinto, T.; Alves, C.; Val\'erio de Carvalho, J.
Springer Proceedings in Mathematics and Statistics
In this thesis, we propose a set of algorithms regarding real combinatorial optimization problems in the context of transportation of goods. These problems consist in the combination of the vehicle routing problem with the two-dimensional bin-packing problem, which is also known as the vehicle routing problem with two-dimensional loading constraints. We also analyzed two related problems, namely the elementary shortest path and the vehicle routing problem with mixed linehauls and backhauls. In both problems, two-dimensional loading constraints are explicitly considered. Two column generation based approaches are proposed for the vehicle routing problem with two-dimensional constraints. The rst one relies on a branch-and-price algorithm with di erent branching schemes. A family of dual valid inequalities is also de ned, aiming to accelerate the convergence of the algorithm. The second approach is based on a set of di erent heuristics strategies, which are applied to the reformulated model. The elementary shortest path problem with two-dimensional constraints is addressed due to its importance in solving the subproblem of the column generation algorithms. To the best of our knowledge, we contribute with the rst approach for this problem, through di erent constructive strategies to achieve feasible solutions, and a variable neighborhood search algorithm in order to search for improved solutions. In what concerns the vehicle routing problem with mixed linehaul and backhauls and two-dimensional loading constraints, di erent variable neighborhood search algorithms are proposed. These algorithms explored various neighborhood structures, being some of those developed based on the features of the problem. All the proposed methods were implemented and experimentally tested. An exhaustive set of computational tests was conducted, using, for this purpose, a large group of benchmark instances. In some cases, a large set of benchmark instances was adapted in order asses the quality of the proposed models. All the obtained results are presented and discussed.