一类基于分治原理的多种群协同进化算法

Multi-population co-evolutionary algorithm based on the 'divide and conquer' principle

根据脑生理学和社会分工的特点和方式,提出一类"分而治之"多种群进化算法。该算法在任务分解机进行任务分配后,子种群独立完成所分配任务,与其它群体几乎不发生联系。在子伤务完成后,各子群中的优秀分子组成新的种群,在整个问题空间完成进化,然后由决定机构根据情况选择相应的可能行动。最后就两个复杂多模态函数优化问题对该算法进行了实验研究,结果表明:合理的"分而治之"方法在效率和效果上明显优于单种群方法。

According to the characteristics and modes of brain physiology and social work division, a multi-population genetic algorithm is presented based on the 'divide and conquer' principle. After assignment of tasks, sub-populations implement their own assigned tasks respectiyely without almost any connection with others. A new population consists of excellent ones of each evolved sub-population and is evolved within the whole search space, and at last, corresponding actions are taken based on the results by the decision unit. Experimental studies are fulfilled on complex multi-mode function's optimization problems. Results demonstrate that the 'divide and conquer' method is much better than the single population counterpart in effectiveness and efficiency.