基于全排序与混沌多样性的高维目标进化算法

Many-objective Evolutionary Algorithm Based on Full Ranking Computer Engineering and Applications

当前大部分多目标进化算法采用Pareto排序为种群个体指定适应度值;然而随着优化目标个数增加,种群中非支配个体的比例越来越大,造成上述算法的搜索能力迅速下降。针对高维(4个以上)目标优化问题,提出了一种全排序方法;该排序方法与Pareto排序具有一致性,并且能够对非支配解进行比较;因此基于全排序的多目标进化算法不受目标个数增加的影响。为了提高算法的优化效果,设计了一个混沌映射算子,用来周期性地初始化种群,以保证种群的多样性与均匀分布。最后,采用标准测试问题对所提算法与著名的非支配快速排序遗传算法(NSGA2)进行了实验比较。结果表明在高维目标优化问题中,所提算法无论在收敛精度,还是算法运行效率上都高于NSGA2算法。

Most of multiohjective evolutionary algorithm adopt Pareto-ranking and their searching ability decrease rapidly with the increase of objective number. That is because the proportion of nondominated individuals in the population is big. For high-dimensional multi-objective optimization problem, it proposes a full ranking method. The ranking is consistent with Pareto ranking, and the nondominated solution can be compared by the full ranking. In order to improve the efficiency of optimization algorithm, it designs a chaotic model to periodically initialize population. Finally, the proposed algorithm and a well-known nondominated sorting genetic algorithm （NSGA2） are compared using the standard test problems. The experiment results show that the proposed algorithm is better than NSGA2 algorithm both in convergence accuracy and efficiency of the algorithm.