用态势模型预测基因表达式编程的进化难度

Gene Expression Programming Evolution Difficulty Prediction Based on Posture Model

在基因表达式编程(gene expression programming,简称GEP)中,由于不同问题得到的适应度-距离相关系数(fitness-distance correlation,简称FDC)值很相近,所以难以用FDC预测GEP求解不同问题的进化难度.为了解决该问题,提出了态势模型及其区间密度指标来预测GEP的进化难度.主要工作包括:(1)提出了GEP染色体之间的距离和态势模型的新概念;(2)提出了态势模型中的区间密度指标;(3)从动力学角度证明了态势模型是对GEP原搜索空间的一种映射,并且该映射保持了种群在原搜索空间中移动的动力学性质;(4)分析了用态势模型区间密度预测GEP进化难度的合理性;(5)用实验验证了区间密度能够准确预测GEP求解问题的进化难度.

Fitness Distance Correlation （FDC） can hardly predict the evolution difficulty of Gene Expression Programming （GEP） because problems with different hardness would result in very similar FDC values in GEP. To solve the problem, the authors propose a posture model and region density to predict GEP＇s evolution difficulty. This study made the following contributions： （1） It introduces the concepts of the chromosomes＇ distance and posture model in GEP; （2） It proposes region density of a posture model; （3） It proves that the posture model is a mapping from the original searching space, and the mapping preserves the population＇s dynamic migration property in the original searching space; （4） It demonstrates the validity of using posture model and region density to predict GEP＇s evolution difficulty; （5） It conducts extensive experiments to show that the new model can precisely predict the evolution difficulty of GEP.