抄録
Many knowledge-based expert systems have been developed. And their reliability is becoming a major bottle-neck for their practical use. This paper proposes a dynamic verification method for knowledge-based systems to detect their errors automatically. This is such a method that makes knowledge-based system to be verified infer actually and keep a good watch over their reasoning process.
To detect systems errors, the verifier “Correctness Checker” traces the reasoning process by watching their working memory, which is a storehouse of reasoning data and therefore can be considered to reflect the systems inference exactly. And we adopt the system specification in the form of constraints as evaluation criteria. We consider that there should be any error in the system if its working memory is changed to violate their constraints.
In this paper, we describe (1) static logical verification method for rule-based systems and its limitations, (2) new framework of dynamic verification method and evaluation criteria, and (3) execution example of dynamic verification for an operation decision-making system for electric power supply systems.
This dynamic verification method has the following advances: (1) Verification for a whole inference system is possible. (2) We can certify what items verified systems satisfy explicitly. (3) The Correctness Checker is a module completely independent of knowledge-based systems to be verified.
