An algorithm for the car sequencing problem of the ROADEF 2005 Challenge

The algorithm we present for solving the Car Sequencing Problem as defined in the Roadef Challenge 2005 consists of the following components. First, a constructive heuristic is used to generate an initial sequence. Next, a local search is applied to improve the solution quality. Because a simple local search would stop in a local optimum as soon as no more improvements can be achieved with any single local exchange, we apply metaheuristics to help the underlying local search to escape from local optima and to find better solutions. We implemented and tested about 10 different constructive heuristics. These heuristics have the goal to create a sequence that serves as a good starting point for the local search in a reasonable amount of time. Among the different algorithms tested we finally selected two, each of which performs best on a different type of problem instances. They are described in Section 2. A local search is performed using a specific neighborhood structure, that is a set of neighbors for each sequence. Among the different neighborhood structures we tested, those which proved the most effective and which we use in the final algorithm are: