TY - GEN

T1 - Theoretical and Empirical Investigation of Fourier Trajectory Analysis for System Discrimination

AU - Morgan, Lucy

AU - Barton, Russell

PY - 2020

Y1 - 2020

N2 - With few exceptions, simulation output analysis has focused on static characterizations, to determine a property of the steady-state distribution of a performance metric such as a mean, a quantile, or the distribution itself. Analyses often seek to overcome difficulties induced by autocorrelation of the output stream. But sample paths generated by stochastic simulation exhibit dynamic behavior that is characteristic of system structure and associated distributions. In this technical report, we investigate these dynamic characteristics, as captured by the Fourier transform of a dynamic simulation trajectory. We find that Fourier coefficient magnitudes can have greater discriminatory power than the usual test statistics, and with simpler analysis resulting from the statistical independence of coefficient estimates at different frequencies. Theoretical and Empirical results are provided.

AB - With few exceptions, simulation output analysis has focused on static characterizations, to determine a property of the steady-state distribution of a performance metric such as a mean, a quantile, or the distribution itself. Analyses often seek to overcome difficulties induced by autocorrelation of the output stream. But sample paths generated by stochastic simulation exhibit dynamic behavior that is characteristic of system structure and associated distributions. In this technical report, we investigate these dynamic characteristics, as captured by the Fourier transform of a dynamic simulation trajectory. We find that Fourier coefficient magnitudes can have greater discriminatory power than the usual test statistics, and with simpler analysis resulting from the statistical independence of coefficient estimates at different frequencies. Theoretical and Empirical results are provided.

M3 - Other contribution

ER -