Towards a Guided Cooperative Search

We present a framework for a guided parallel cooperative search that combines common meta-heuristics to solve combinatorial problem with more robustness and efficiency. Based on the central memory concept, the proposed identification pattern mechanism sends information to individual meta-heuristics about promising and unpromising patterns of the solution space. By fixing or prohibiting specific solution attribute values in particular search methods, we can focus the search to desired regions. This mechanism may thus be applied to enforce a better coordination between the individual methods and control the diversification and intensification of the global search. We apply this mechanism to the Vehicle Routing Problem with Time Windows. Experimental results on an extended set of benchmark problem sets illustrate the benefits of the proposed methodology.