Model Decomposition and Simulation

Qualitative reasoning uses incomplete kmiowlcdge to coni— pute a description of the possible behaviors for dynamic systems. For complex systems containing a large number of variables and constraints, t.he simulation frequently is inti-actable or results in a large, incomprehensible behavioral description. Abstraction and aggm-egation techniques are reqiied during the simulation to eliminate irrelevant details and highlight the imnportant characteristics of the behavior. The total temporal ordering of unrelated events provided by a traditional state-based qualitative representatiomi is one such irrelevant distinction. Model decomposition and simulation addresses this probleni. Model decomposi ion uses a causal analysis of the model to partition the variables into tightly connected coat— ponents. The components are simulated separately in the order dictated by the causal analysis beginning with causally upstream connpomtents. Iniforniation from the simulation of causally upstream coniponents is used to constrain the behavior of downstreammm components. If a feedback loop exists between components or a set of components are acausally related, then a concurrent simulation is performed for these components. A truth mainte— namicc system is used to record said retract assumuptions nuade during this concurrent simulation. Model decomposition provides a general architecture which separates the method of siniulation from the model decomposition algorithm. This architecture can he used to introduce alternative abstraction techniques to eli miii— nate other irrelevant distinctions.