Exploring the Accuracy of a Parallel Cooperative Model for Trajectory-Based Metaheuristics

摘要

Classical cooperative parallel models for metaheuristics have one major issue when the underlying search method is based on the exploration of the neighborhood of one single solution, i.e., a trajectory-based metaheuristic. Whenever a cooperation step takes place by exchanging solutions, either the incoming or the local solution has to be discarded because the subalgorithm does only work with one single solutions. Therefore, important information may be lost. A recent new parallel model for trajectory-based metaheuristics has faced this issue by adding a crossover operator that is aimed at combining valuable information from both the incoming and the local solution. This work is targeted to further evaluate this parallel model by addressing two well-known, hard optimization problems (MAXSAT and RND) using Simulated Annealing as the search method in each subalgorithm. The results have shown that the new model is able to outperform the classical cooperative method under the experimental conditions used.