基于三支决策的证据融合策略  

Evidence Fusion Strategy Based on Three-way Decision

作　　者：陈缘 王加阳[1] CHEN Yuan;WANG Jia-yang(School of Computer Science and Engineering,Central South University,Changsha 410083,China)

机构地区：[1]中南大学计算机学院,长沙410083

出　　处：《小型微型计算机系统》2023年第8期1629-1636,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61772031)资助;湖南省自然科学基金项目(2020JJ4753)资助。

摘　　要：针对现有证据合成改进方法过程复杂,计算量较大,不加判别地处理证据冲突等问题,提出一种基于三支决策的证据融合策略.首先将证据关系分为一致、冲突和中立3种,建立基于三支决策的证据关系模型.设计算法筛选出证据支持的主要焦元.利用焦元集合的相似度和向量距离分别计算一致概率和代价函数.依据贝叶斯最小风险决策理论来判定证据关系.接着,在该模型的基础上提出基于三支决策的证据融合策略.根据各证据的主要焦元确定初始证据集的焦元分布.通过把证据划分到不同证据集内部或边界上来实现证据的三支划分.最后对加权修正的证据集融合结果和边界证据进行合成.算例分析表明,本文方法不仅能处理证据悖论,还使融合结果更稳定可靠.In the light of the problem that the existing evidence synthesis improvement methods are complex,computationally inten-sive,and handle evidence conflicts without discrimination,an evidence fusion strategy based on three-way decision is proposed.First-ly,the evidence relationship is divided into three types:uniform,conflicting and neutral.And an evidence relationship model based on three-way decision is established as well.An algorithm is also designed to filter out the main focal elements supported by the evidence.The similarity and vector distance of the set of focal elements are applied to calculate the agreement probability and cost function re-spectively.The Bayesian minimal risk decision theory is used to settle the evidence relationships.Secondly,an evidence fusion strategy based on three-way decision is proposed on the basis of this model.The focal element distribution of the initial evidence set is decided by the main focal elements of each evidence.By allocating the evidence into different evidence sets inside or on the boundary,the three-way division of evidence is achieved.Finally,the fusion results of modified evidence set and the boundary evidence which are weighted are synthesized together.The analysis of the algorithm shows that this method can not only deal with the evidence paradox,but also make the fusion results more stable and reliable.

关 键 词：证据融合 三支决策 代价函数 一致概率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]