定性的シミュレーションに基づく診断システムの構築法

抄録

This paper describes a method for developing diagnostic systems through qualitative simulation. The method can be viewed as a practical approach to knowledge base construction using current qualitative simulation techniques. Unlike the previous experiential approaches, diagnostic rules are derived from a structural model, which is described as qualitative parameters and constraints among them. The rule generation procedure is achieved by finding the difference between the behavior of normal state and that of abnormal state. The system called QR/P is designed and implemented in logic programming languages to test on a medical domain. An experiment on intraocular pressure mechanism generated 60 rules applied to glaucoma diagnostic system directly. The present method has novel features : First, QR/P predicts the behavior of an abnormal state by a slight change of the initial condition representing the normal state. Second, conditional probabilities are given by ambiguities due to qualitativeness. Third, QR/P computes subjective probabilities associated with rules by using ordinaly probability theory. These capabilities contribute to the connection between model-based and experiential approach to developing knowledge based systems.