基于不确定理论的CBRB规则激活权重计算方法

合取型置信规则库系统的推理方法以概率论为基础,通过乘法实现合取关系,然而这种计算方法并不总能准确反映规则的激活程度.鉴于此,引入不确定理论,改进前件属性合取关系的实现方式,以前件属性中最小的个体匹配度作为核心,并保留信息利用的完整性,提出一种新的规则激活权重计算方法.在此基础上,利用扩展置信规则库系统分别在回归和分类问题上对该方法的推理效果进行验证.实验结果表明:新方法有效提高了系统的推理能力,并能稳定地适用于多个数据集.

A new calculation method of rule active weight for conjunction belief rule-base based on uncertainty theory

The reasoning method applied in conjunction belief rule-base is obtained on the basis of probability theory, and therefore the conjunction relation between individual matching degrees are expressed by multiplication. However, the product of individual matching degrees may not exactly reflect the activation degree of rules in all situation. Hence, referring to the method of uncertainty inference, improving the implementation of conjunction relation between antecedent attributes, a new calculation method of rule activation weight is proposed in this paper. The new method regards the minimum individual matching degree of antecedent attributes in each rule as the key, while the other individual matching degrees and the system parameters of conjunction belief rule-base are regarded as auxiliary part to ensure the integrity of information utilization and the stability of reasoning. Then using extended belief rule-base, the reasoning ability of new method is verified on the regression and classification problems in this paper. The experiments illustrate that the new calculation method of rule activation weight applies to most datasets, and generally improve the reasoning ability of conjunction belief rule-base.