Expert system reasoning about dynamic systems by semi-quantitative simulation

Current symbolic knowledge representation techniques are inadequate for describing complex dynamic systems, which are time-varying and contain feedback loops. These systems are particularly difficult to represent when quantitative knowledge about the relationships within the system is incomplete. This paper describes a symbolic extension of the system dynamics method which can answer 'what if' questions about system dynamics models semi-quantitatively. The method consists of (1) definition of generic model 'building blocks' which map directly onto difference equations; (2) development of a symbolic causal model of a system in terms of the generic functional 'building blocks', any available quantitative or semi-quantitative parameters, and a set of generic default values; (3) automatic translation of the model into a system of first-order difference equations; and (4) numerical integration of the equations by standard methods. A complex model with semi-quantitative parameters, representing the human cardiovascular system, is used to illustrate the method. This method couples the descriptive abilities of mathematics with the symbolic power of causal inference methods by providing the same knowledge base for both computational levels. It also eases truth maintenance, knowledge acquisition, and explanation. It may prove useful for expert system development in a variety of application domains whose processes are time-varying and homeostatic.