Application of system identification techniques to simulation model abstraction

This paper describes preliminary research into the applicability of system identification techniques to simulation model abstraction. Model abstraction enables the construction of a valid, low-resolution surrogate to a more detailed, high-resolution simulation model. When rapid, approximate results will suffice, we can also apply system identification directly to actual system data, bypassing the simulation stage. Four non-traditional system identification techniques are discussed in relation to their ability to produce linear, time-invariant, state-space formulations of multivariable random systems. A simple example is provided in which one of the techniques, Hidden Markov Models, is used to identify the transition probabilities within a simulated Markov Chain. The example is used to illustrate the challenges in general simulation model abstraction caused by model transformation procedures, problem size, uncertainty, and computational complexity. At this stage, we can say that the application of systems identification to simulation model abstraction is promising, yet challenging.