一种基于双精英种群的协同进化算法研究

摘要

A co-evolutionary algorithm based on double elite-populations (DEPEA)is proposed in light of the shortcomings of traditional multi-objective evolutionary algorithm with elitist strategy that its distribution of population convergence is uneven and the capability of global search is insufficient.Learning from the idea of subinterval division and non-dominated sorting,the algorithm divides entire population into two elite populations with different levels and one general population.In combination with the concept of co-evolutionary,two elite populations adopt different evolutionary strategies respectively to achieve the balance on the capability of the algorithm in exploration and ascertainment. The high level elite population and the low level elite population adopts collaborative operation to promote the generation of more excellent individuals;the high level elite population and the general population adopts leading operation to expedite the approaching of general individuals to elite individuals.After the experiment on five standard test functions and comparing them with the results of traditional NSGA-II algorithm and the latest hybrid_MOEA algorithm,it is shown that the new algorithm has better global convergence and can ensure the diversity of the population better.%针对传统带精英策略的多目标进化算法种群收敛分布不够均匀，全局搜索能力不足的缺点，提出一种基于双精英种群的协同进化算法 DEPEA（Double Elite Populations Co-evolutionary Algorithm）。该算法借鉴了子区间划分和非支配排序思想，将整个种群划分成两个不同级别的精英种群和一个普通种群；两个精英种群结合协同进化思想分别采用不同的进化策略实现对算法的探究和探查能力的平衡，高级别的精英种群与低级别的精英种群采用协作操作，促进更优秀的个体产生；高级别的精英种群与普通种群采用引导操作，加快普通个体向精英个体逼近。通过对五个标准的测试函数进行实验，并与传统的 NSGA-II 算法和最新的 hybird＿MOEA 算法结果进行比较与分析，表明该算法不仅具有更好的全局收敛性，且能够更好地保证种群的多样性。