Atracction Basin Estimatign GA: An Adpative and Efficient Technique for Multimodal Optimization


Multimodal optimization aims to discover all or most optima as opposed to only the best optimum. Evolutionary Algorithms provide a natural advantage in this field, because they are population based. However, Standard Evolutionary Algorithms tend to converge only to a single optimum. The radius-based niching evolutionary algorithms aim to solve this problem. However, they are criticized for the difficulty of the proper choice of the radius parameter. Detect-multimodal method does not necessitate using the radius parameter. It separates niches by detecting if two solutions are in same optimum. Although robust, the current detect-multimodal based niching method are computationally expensive. Inspired by the idea of combining radius-based niching method and detect-multimodal based niching method, we propose the Attraction Basin Estimating Genetic Algorithm (ABE) in this paper. It estimates the radius which is called attraction basin in this paper using detect-multimodal method, and use the estimated radius to separate species in the same way as radius-based method. We compare the proposed method with a detect-multimodal based method: Topological Species Conservation Algorithm. The experiments demonstrate that ABE has the similar ability to solve multimodal optimization problems as Topological Species Conservation, but significantly more efficiently.