A Novel Niche Genetic Algorithm for Multimodal Optimization
Yi Zhang, Hu Zhang · Journal of Convergence Information Technology · 2011
Genetic drift usually takes place when we use genetic algorithm to calculate multimodal functions. It causes the calculations can not obtain multiple globally optimization solutions. For making up this shortage, we propose a new crowding niche genetic algorithm based on the most similar individuals (MSICNGA). This new algorithm adds two novel mechanisms: crowding mechanism based on the most similar individuals and crowding errors repair mechanism. The crowding mechanism aims to maintain population diversity through repeatedly excluding the inferior one of the two most similar individuals in current population; it is able to effectively avoid the setting of any parameters related to the priori knowledge about optimization functions in the operation process. The repair mechanism is used to protect optimum solutions from loss to repair crowding errors through storing the better ones among the individuals crowded out in the operation process; it is capable of making sure algorithm finds out multiple solutions more easily. In order to assess the performance of MSICNGA, we adopt four typical benchmarks to make optimization experiments, and compare the performance with other two commonly known algorithms. Results and analysis show that MSICNGA is able to well maintain population diversity and easily search multiple solutions while calculating multimodal functions, besides, it takes on a better stability and universality for optimizing various kinds of problems.