基于广义罚函数可行性准则的DE算法对不确定数据的处理

Application of improved DE algorithm based on generalized penalty function feasibility criteria in uncertain data processing

针对不确定数据集成效率低的问题,构造基于区域分割的广义罚函数可行性准则,分析了分割搜索区域的迭代点特征和可行性准则的性质与优势,据此提出一种基于广义罚函数可行性准则改进的DE算法(DE-GPFFC算法).机器学习数据集UCI中不确定数据集的数值结果显示:不确定数据集中最优可行点趋向概率0.5分布,其他数据点趋向概率0,1分布,其中趋向于概率0.5分布的数据点位于可行域int(D),其他数据点位于非可行域out(D).DE-GPFFC算法使得不确定数据集在可行域边界Round(D)进行跨区域搜索,有效提高了不确定数据分类集成效率.

Aiming at the problem of low integration efficiency of uncertain data,this paper constructs a generalized penalty function feasibility criteria(GPFFC)based on region segmentation,analyzes the characteristics of the iterative point feature and the nature and advantages of the screening criterion in the segmented search region,and proposes a improved DE algorithm based on GPFFC(DE-GPFFC).The numerical results of the UCI uncertain data set show that the optimal feasible point tends to have a probability distribution of 0.5 in the uncertain data set,and other data points tend to have a probability of 0,1 distribution,wherein the data points tending to the probability of 0.5 are located in the feasible domain int(D),other data points are located in the non-feasible domain out(D).The DE-GPFFC algorithm makes the indeterminate data set cross-region search at the feasible domain boundary Round(D),which effectively improves the efficiency of uncertainty data classification integration.