一种自适应小生境分布性保持策略

An Adaptive Niche for Keeping the Diversity of Solutions in Multi-Objective Evolutionary Algorithm

小生境技术被广泛应用在多目标进化的分布性保持方面.但是,小生境半径不易控制等限制了其在分布性保持等方面的发展.本文提出了一种自适应小生境分布性保持策略(Adaptive Niche,AN).AN通过对Pareto解集生成最小生成树来自适应调整小生境半径,同时扩大搜索小生境半径并改变计算方法使之能够识别小生境边沿的个体,便于对其修剪与评价.通过与NSGA-II,SPEA2在不同形状测试函数上进行对比实验,结果表明,AN能够对Pareto最优面进行高效地分布性保持.

Niche is an effective and widely used diversity preservation technique in multi-objective evolutionary algorithms（MOEAs）.However,it suffers from two feedbacks：the determination of niche radius is far from trivial and the fitness value evaluated by niche technique is too coarse to be reliable in some scenarios.This paper proposes an Adaptive Niche（AN） technique,in which the parameter value of niche radius can be automatically tuned according to the current population.Furthermore,the individuals locate on the niche boundary and inside are tackled differently,when evaluating the fitness value.Comprehensive experiments demonstrate the superiority of proposed AN,compared to several state-of-the-art MOEAs.