基于证据推理的确定因子规则库推理方法

Certainty rule base inference method using evidential reasoning approach

不确定性推理方法是人工智能领域的一个主要研究内容,if-then规则是人工智能领域最常见的知识表示方法。针对实际问题通常具有不确定性的特点,提出基于证据推理的确定因子规则库推理方法。首先在Ifthen规则的基础上给出确定因子结构和确定因子规则库知识表示方法,该方法可以有效利用各种类型的不确定性信息,充分考虑了前提、结论以及规则本身的多种不确定性。然后,提出了基于证据推理的确定因子规则库推理方法,通过将已知事实与规则前提进行匹配,推断结论并得到已知事实条件下的前提确定因子。进一步,根据证据推理算法得到结论的确定因子。最后,通过基于证据推理的确定因子规则库推理方法在UCI数据集分类问题的应用算例,说明该方法的可行性和高效性。

Uncertainty inference methods have always been a key research area in artificial intelligence, and if-then rule is the most common representation of knowledge in artificial intellegence area. This paper proposed a certainty rule base inference method using the evidential reasoning approach for uncertainty problems. It gave the certainty factor structure and certainty rule base based on if-then rule to represent the uncertainty of knowledge firstly. It designed such a rule base with certainty de- grees embedded in the consequent terms as well as in the all antecedent terms of each rule, which was shown to be capable of capturing vagueness, incompleteness, uncertainty, and nonlinear causal relationships in an integrated way. Then, it provided the certainty rule base inference methodology using the evidential reasoning approach. The overall representation and inference framework offered a further improvement and great extension of the recently uncertainty inference methods as the evidential rea- soning approach was applied to the rule combination. This method infered the consequence attribute values and calculated the certianty factor of antecedent under the known fact. Subsequently, it gave the certianty factors of the consequence attribute val- ues with the evidential reasoning approach. In the end, it provided an application example of classification problems on UCI machine learning repository to illustrate the proposed rule base representation and inference method as well as demonstrate its feasibility and high efficiency by comparing with some existing approaches.