Real-world applications often have multiple objectives which have uncertain factors, noisy fitness function is a significant part of uncertainty. But single objective problem comprise a lager body in current study, while researches on multi-objective evolutionary algorithm for noisy fitness function are relatively rare. There typical multi-objective algorithms are taken in a comparative experiment with interference from different dimension noisy. The effectiveness of four metrics are tested, and the Monte-Carlo integration are used as sample approach, and the feasibility are test by experiment, the scope which is appropriate for monte-carlo integration , and the shortcoming of it is stated. The results of the experiment shows regression of the MOEAs for noisy fitness function, the traditional Monte-Carlo integration is inapplicable when the noisy is large, and new metric for noisy fitness function is inevitable.%多目标优化算法在现实生活中要解决的问题通常是带有不确定性的,适应度存在噪声干扰是不确定性的一个重要方面,所以多目标进化算法求解噪声适应度函数问题具有重要的现实意义,通过实验的方法,研究了3个典型多目标算法在不同规模下噪声干扰下的性能,4个评价方法在噪声环境下的有效性,以及传统的蒙特卡洛积分方法能够适用的范围,实验结果表明,3个典型多目标算法求解噪声适应度函数问题均不理想,传统的蒙特卡洛积分方法随着噪声规模的增加性能下降很快,同时需要更加适用于噪声适应度方程问题的评价方法.
展开▼
