Modeling and Credibility of Random Ensembles

Abstract

Many simulation applications require the generation of a set of sample functions to characterize the physical system being modeled. Probability modeling using minimal data bases are reviewed and model credibility issues examined. Modeling methods include the Box-Jenkins time series, the fixed- memory and the α-β filters, the generalized expand ing memory filter, the Kalman filter, the Bayes filter, and the square-root filter. Such methods are particularly suitable in situations involving limited data, for instance, where the physical sys tem observations may be limited by cost or by the physical experiment. Theil's Inequality Coefficient (TIC) is presented as a basis for model validation in specific applications. An example illustrates the special utility of TIC for simulation model validation.