交互式遗传算法的噪声及降噪策略  

Noises in interactive genetic algorithms and strategies for its reduction

作　　者：周勇[1] 巩敦卫[2] 

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221008 [2]中国矿业大学信息与电气工程学院,江苏徐州221008

出　　处：《控制理论与应用》2008年第2期223-227,共5页Control Theory & Applications

基　　金：国家自然科学基金资助项目(60575046);中国矿业大学青年科研基金资助项目(OD4549)

摘　　要：在交互式遗传算法中,人对进化个体的评价含有噪声.如何降低噪声对评价的不利影响,至今没有有效的方法.这严重制约了交互式遗传算法在复杂优化问题中的广泛应用.本文首先针对交互式遗传算法中人评价个体适应值的不确定性和漂移性,分析交互式遗传算法噪声的来源,定义认知评价度和疲劳评价度,给出交互式遗传算法的3阶段噪声模型;然后,给出基于个体海明距离的认知评价度和疲劳评价度刻画以及基于适应值可信度的降噪策略;最后,通过在服装设计中的应用实例验证噪声模型的正确性和降噪策略的有效性.Noises exist in human＇s subjective evaluation on individuals in interactive genetic algorithms. There is no effective way to alleviate the influence of noises on evaluation, which restricts broad applications of interactive genetic algorithms to complicated optimization problems. In this paper, sources of noises in interactive genetic algorithms are analyzed in relation with uncertainties and fluctuation in human＇s subjective evaluation on individuals; the degree of evaluating cognition and the degree of evaluating fatigue are defined; and a model for noises with 3 phases in interactive genetic algorithms is proposed. Then, descriptions of the degree of evaluating cognition and the degree of evaluating fatigue based on individuals＇ Hamming distance, and a strategy for reducing noises based on fitness reliabilities are presented. Finally, the efficiencies of the model for noises and the strategy of reducing noises are validated through the application examples in fashion design.

关 键 词：遗传算法 交互 噪声 降噪策略 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]