Multimodal estimation of distribution algorithm based on cooperative clustering strategy
Shanshan Huang, Houming Jiang · 2018
Multimodal optimization is becoming more and more important. Among all the optimization algorithms, estimation of distribution algorithm (EDA) attracts more attention because of its simple principle and special mechanism. Unfortunately, it loses feasibility and effectiveness in multimodal problems because of premature convergence and the missing of multimodality-specific mechanism. So niching is adopted to help EDA to locate diverse promising solution regions in multimodal optimization. However, many of the niching techniques are either sensitive to parameters or cost a number of fitness evaluations. In this paper a new proposed fast cooperative clustering strategy is adopted to offer improved assistance for locating multiple optima based on both decision and target space information. Taking the advantage of EDA in preserving high diversity, this paper proposes a cooperative clustering based multimodal EDA (CMEDA). Integrated with new proposed cooperative clustering strategy, multimodal problems are divided into certain promising regions automatically. Then each cluster independently runs a separate optimizer in parallel to search promising regions carefully, which can avoid premature convergence greatly. Based on the results of clustering, the value of key parameter is determined by statistic information but not artificial setting. Then a balance between exploration and exploitation is achieved. Experimental results indicate that the proposed technique is an effective and efficient algorithm which can not only explores and exploits the promising regions in the search space effectively but also obtain the global optima superior to the typical multimodal EDA (MEDA).