结合近邻密度和信息修正的基本概率赋值生成方法  

A Basic Probability Assignment Generation Method Combining Nearest Neighbor Density and Informa⁃tion Correction

作　　者：白雪婷 陈辉[1] BAI Xueting;CHEN Hui(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan,Anhui 232000,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232000

出　　处：《宜宾学院学报》2024年第6期1-8,14,共9页Journal of Yibin University

基　　金：国家自然科学基金项目(61170060);安徽省重点教学研究项目(2020jyxm0458)。

摘　　要：为了解决D-S证据理论应用中基本概率赋值(BPA)获取困难、生成模型适用度低的问题,提出一种结合近邻密度和信息修正的基本概率赋值生成方法:通过基于KNN算法得出的密度峰值点与样本间的距离为依据生成单焦元BPA函数,通过信念χ~2散度对全子集事件赋值并基于可信度对BPA进行信息修正,用改进的信念熵公式计算各证据的不确定性权重,进行证据的再分配.利用生成的BPA解决少样本和不均衡类样本的实际应用问题,经多个数据集验证诊断精度均达85%以上,优于其他方法.In order to solve the difficulty of obtaining the basic probability assignment(BPA)and the low applicability of genera⁃tion model in D-S evidence theory application,a basic probabilistic assignment generation method combining nearest neighbor density and information correction was proposed.A single focal element BPA function was generated based on the distance be⁃tween the peak density point and the sample data derived from the K-nearest neighbor(KNN)algorithm;the values were as⁃signed to the whole subset of events by beliefχ2 divergence and BPA information was modified based on confidence.The im⁃proved belief entropy formula was utilized to compute the uncertainty weight of each piece of evidence for evidence redistribu⁃tion.The generated BPA is utilized to solve the practical application problems of few samples and imbalanced class samples,and the diagnostic accuracies of the proposed method are all over 85%,which are better than other methods,as verified by multiple datasets.

关 键 词：D-S证据理论 密度 信息修正 信念熵 信念χ~2散度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]