Model abstraction results using state-space system identifications

In this paper we report on state-space system identification approaches to dynamic behavioral abstraction of military simulation models. Two stochastic simulation models were identified under a variety of scenarios. The `Attrition Simulation' is a model of two opposing forces with multiple weapon system types. The `Mission Simulation' is a model of a squadron of aircraft performing battlefield air interdiction. Four system identification techniques: Maximum Entropy, Compartmental Models, Canonical State-Space Models, and Hidden Markov Models (HMM), were applied to these simulation models. The system identification techniques were evaluated on how well their resulting abstractions replicated the distributions of the simulation states as well as the decision outputs. Encouraging results were achieved by the HMM technique applied to the Attrition Simulation--and by the Maximum Entropy technique applied to the Mission Simulation.