Solving the capacitated clustering problem with variable neighborhood search

Variable neighborhood search (VNS) is a proven heuristic framework for finding good solutions to combinatorial and global optimization problems. In this paper two VNS-based heuristics are proposed for solving the capacitated clustering problem. The first follows a standard VNS approach, and the second a skewed VNS that allows moves to inferior solutions. The performance of the two heuristics is assessed on benchmark instances from the literature. We also compare their performance against a recently published iterated VNS procedure. All VNS procedures outperform the state-of-the-art, but the Skewed VNS is best overall. This would suggest that using acceptance criteria before allowing moves to inferior solutions in Skewed VNS is preferable to the random shaking approach that is used in Iterated VNS to move to new regions of the solution space.