Abstract: Traditional RETE algorithm supports only Boolean first-order logic in which a question is either true or false. However, many of the problems themselves are imprecise. The definition is vague or ambiguous（collectively referred to as “imperfection”）. For instance, some rules themselves are uncertain, in addition, the data itself maybe has an “imperfect” characteristic, because many types of data come from a variety of sensors. However, the traditional RETE algorithm can not effectively deal with “imperfection”. For the issues, an improved RETE algorithm which can be used to handle imperfection effectively is proposed. “Imperfect degree” is introduced to traditional RETE network to measure all types of “imperfect” metric. Constraint operations of constraint nodes are extended at the same time. Moreover, the operator nodes are added to RETE network. At last, the various types of “imperfection” can be effectively supported.

Key words: rule-based system, RETE, imperfection, imperfection reasoning

摘要： 传统的RETE算法仅支持布尔的一阶逻辑推理，在一阶逻辑中，一个问题非真即假，然而现实中许多问题本身是不精确的、定义模糊的或者是有歧义的（统称为“瑕疵”），例如某些规则本身的表示不确定，此外，目前许多数据来自于多种类型的传感器，因而这些数据本身具有“瑕疵”特征。然而传统的RETE算法不能自然有效地处理“瑕疵”，针对此问题，提出可处理瑕疵的RETE算法。通过在传统的RETE中引入“瑕疵度”对多种类型的“瑕疵”度量，对RETE网中约束节点的约束操作进行扩展，并在RETE网中引入操作符节点，从而实现对多种类型的“瑕疵”的有效支持。

关键词: 基于规则的系统, 一种模式匹配算法（RETE）, 瑕疵, 瑕疵推理