Parallel Variable Neighbourhood Search for the Car Sequencing Problem

Variable Neighborhood Search (VNS) is a relatively new metaheuristic for solving hard combinatorial optimisation problems. One such optimisation problem is the Car Sequencing Problem (CarSP), where a sequence of cars along the assembly line with minimum production costs has to be found. Although VNS is a successful metaheuristic, it takes a long time until a suitable solution is found for real-world instances of CarSP. Two approaches should be investigated in more detail: Firstly, the efficiency of neighborhoods, i.e. the relation of computation time and the solution improvement, should be used for identifying efficient neighborhood orderings on the fly. Secondly, the high potential of parallelisation techniques should be exploited. Within this thesis both approaches are combined. Computational tests showed that a substantial reduction of the computation time is possible. Further, the tests revealed that no “perfect” neighborhood ordering can be identified which implies that such a parallel self-adaptive approach is valuable and necessary for obtaining good solution qualities.