Reference: Vina, A. & Hayes-Roth, B. Knowledge-Based Real Time Control: The Use of Abstraction to Satisfy Deadlines. Knowledge Systems Laboratory, September, 1991.
Abstract: In this paper we address the use of model-based knowledge representation and reasoning techniques under a blackboard framework in real time computer-based control. Representing knowledge using models is one of the most powerful ways to codify the controller's knowledge about the system being controlled. By using a model and the current values for the monitored parameters, the controller can base all their process automation tasks on the outcomes of a precise simulation of the system behavior. But models have a serious drawback: they have difficulty accommodating normal quantities of components found in current systems. To overcome this problem when working under real time constraints, we propose a techniques which allows a control system to achieve real time performance by means of the following computational steps: (1) starting with the basic user-defined domain model, the engineer precomputes a hierarchy of abstract models, at increasing levels of abstraction, using the aggregation of causal links to guide the abstraction process; (2) continuing off-line, the engineer estimates the worst-case processing time for every model in the hierarchy; (3) on-line, the controller uses the time constraints of the situation and the worst-case processing times of the models to choose one model with the level of abstraction that will guarantee deadline satisfaction; and (4) the controller performs system simulation using the selected model. We have used the BB1 blackboard architecture [Hayes-Roth, 1985; 1988; 1990] to implement our work. The results, however, can easily be transferred to another computational framework.