Significant Factor Identification Using Discrete Spectral Methods.
Abstract
Discrete event simulations are computer models of complex systems. The accuracy of the model in reflecting the behavior of the real system is determined by, among other things, the values of parameters for distributions used in the model. The modeller could allocate experimental resources more effectively without loss of accuracy in the model if he or she could identify those parameters to which the response of interest has the greatest sensitivity. One method of doing this is to try to model the response as a polynomial function of the model parameters. We are then interested in those terms in the polynomial which have non-zero coefficients. This report attempts to extend the schruben/cogliano methodology to cover a more general class of models which includes discrete-valued parameters, such as policy decisions or capacities of queues. We evaluated the use of discrete-valued functions as a basis for spectral analysis. Several function sets were considered as possibilities, and Walsh functions were selected as the best choice.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1985
- Accession Number
- ADA161933
Entities
People
- Lee W. Schruben
- Paul J. Sanchez
Organizations
- Cornell University