A benchmark problem for model abstraction techniques

Modeling of real systems relies on the arduous task of describing the physical phenomena in terms of mathematical models, which often require excessive amounts of computation time in their execution. In the last few years there has been a growing acceptance of model abstraction whose emphasis rests on the development of more manageable models. Abstraction refers to the intelligent capture of the essence of the behavior of a model, without all the details. In the past, metamodels have been generated from complex models, such as the Tactical Electronic Reconnaissance Simulation Model (TERSM). The scope of this paper is to explore the ability of previously developed TERSM metamodels to accurately simulate the benchmark model using both Limited subsets of the original data, and data subsets whose values are interpolated or extrapolated from the original data set used to generate and fit the model. This paper establishes a baseline from which additional metamodels can be compared and analyzed.