数据缺失的扩展置信规则库推理方法  

Extended Belief Rule Base Reasoning Approach with Missing Data

作　　者：刘永裕 巩晓婷[2] 方炜杰 傅仰耿[1] Liu Yongyu;Gong Xiaoting;Fang Weijie;Fu Yanggeng(College of Mathematics and Computer Science,Fuzhou University,Fuzhou 350116;Institute of Decision Sciences,Fuzhou University,Fuzhou 350116)

机构地区：[1]福州大学数学与计算机科学学院,福州350116 [2]福州大学决策科学研究所,福州350116

出　　处：《计算机研究与发展》2022年第3期661-673,共13页Journal of Computer Research and Development

基　　金：国家自然科学基金项目(61773123);福建省自然科学基金项目(2019J01647);工信部工业互联网创新发展工程项目(TC19083WB)。

摘　　要：数据驱动的扩展置信规则库专家系统能够处理含有定量数据或定性知识的不确定性问题.该方法已被广泛地研究和应用,但仍缺乏在不完整数据问题上的研究.鉴于此,针对不完整数据集上的问题,提出一种新的扩展置信规则库专家系统推理方法.首先提出基于析取范式的扩展规则结构,并通过实验讨论了在新的规则结构下,置信规则前提属性参考值个数对推理方法的影响;然后提出通过不完整数据生成具有不完整置信规则,并构成析取范式置信规则库的方法,同时引入衰减因子修正不完整规则权重,使不完整规则可以更合理地参与信息融合过程;最后,选取若干个公共数据集对所提方法进行验证.与其他方法的实验对比显示,新方法在完整数据集上有良好表现的同时,对具有不同缺失程度和缺失模式的数据集表现出更好更稳定的推理效果.The data-driven constructed extended belief rule-based system can deal with uncertainty problems with both quantitative data and qualitative knowledge.It has been widely researched and applied in recent years,but infrequently been involved in the field of incomplete data.This study conducts research focusing on the performance of the extended belief rule-based system applied to incomplete datasets and proposes a novel reasoning approach for the case of data missing.First,a disjunctive extended rule base is constructed and the optimal number of antecedent attribute referential values is discussed through validation experiments.Then a method for generating a disjunctive belief rule base from incomplete data and consisting of disjunctive belief rule base is proposed,and an attenuation factor is introduced to modify the weight of incomplete rules to make the aggregation of information more reasonable.Finally,this paper conducts experiments on several commonly used datasets selected from UCI to validate the improvement of the proposed method.The experiments are designed with various degrees and patterns of data missing,and the performance of the improved system is analyzed and compared with some conventional mechanisms.Experimental comparison with other methods shows that while the new method performs well on complete datasets,it also shows better and more stable inference effects on datasets with different degrees of missing and patterns.

关 键 词：置信规则库 证据推理 数据驱动 数据缺失 不完整数据集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]