On fidelity and model selection for discrete event simulation

Abstract

In simulation, perhaps the most common use of the term ‘fidelity’ refers to the faithfulness with which model behavior reflects modeled system behavior. While there have been studies of fidelity seeking absolute and quantitative measures, there is not yet a consensus on a workable fidelity metric. We propose a formal modeling framework for comparing discrete event system simulation models in terms of fidelity, using a relative fidelity indicator. Based on the framework, we consider the possibility that the higher fidelity simulation models can also be more productive, even though they are more expensive to develop and use, since they can be used to achieve multiple objectives. First, we propose a formal simulation modeling framework within which the fidelity of simulation models can be discussed. With this framework and a simple example, we then define a relative fidelity indicator that provides a systematic way of comparing the fidelity of two simulation models. The relative fidelity indicator focuses on the most important characteristics in simulation studies: the input and output interfaces and the variables used for specifying a real-world system and simulation models. It does not require any special modeling formalism for model comparison. Based on the relative fidelity indicator and simulation modeling framework, we state the optimum simulation model selection problem to achieve given simulation objectives. Under a practical assumption, we analyze the simulation model selection problem and derive properties related to simulation modeling and the fidelity of simulation models.