| Abstract: | Simulation is a powerful technique for the analysis of complex systems. The labor intensive nature of simulation studies, however, shows that current software practices are becoming quickly outmoded with the rising demand for simulation systems. This research addresses an alternate perspective of traditional practices. Simulation is visualized as a problem solving method in which the user specifies the model as well as the analysis goals to be attained by executing the model. Expert software selects the appropriate parameter values that meet analysis goals based on a derived understanding of model behavior. The approach, called goal-oriented simulation, is presented in terms of automated theorem proving. Models are described by rules defining component behavior. Transactions required to verify model behavior are represented as goals. Goals are proven using the specification of model behavior as axioms. Statistics are maintained on the amount of simulated time required to prove each goal. If the inference process does not meet performance criteria given by the user, the cause of poor performance is diagnosed and advice generated as to how to modify the model to meet objectives. The goal-oriented technique is set in the framework of an object-oriented rule-based model representation. A restricted prototype of the model experiment and execution components is implemented in Prolog. Research results demonstrate that the approach shows promise as a viable first step toward a knowledge-based simulation environment. |
|---|