A global optimization algorithm based on combined evolutionary programming/cluster analysis

In this paper, a robust optimization algorithm based on combined evolutionary programming (EP)/cluster analysis is presented. In order to alleviate the problem of premature convergence when the objective function is highly multimodal, an efficient algorithm is developed that combines the conventional EP with clustering method. In this approach, the cluster analysis provides an accurate estimate on the size and location of clusters of subpopulations, which are formed during the optimization process. This information is then used to put pressure on the members of each individual cluster for sharing the fitness value, similar to the approach used in the well-known niching technique. The developed algorithm has been successfully applied to many mathematical test problems as well as several RF/microwave engineering problems. The results for a multimodal function optimizations and an example on the synthesis of a microwave multiple-coupled resonator for wireless application using optimization approach are presented.