Simulation optimization using frequency domain methods

ABSTRACT

Simulation programs can be quite complex, sometimes involving a great number of input factors and parameters. Conventional experiments where each setting of the input values requires a separate simulation run can be quite time consuming and expensive; such experiments will be referred to as “run oriented” simulation experiments. An alternative approach is to allow the input variables in a simulation to vary according to specific patterns during a single run. Various output spectra can then be analyzed to gain information about the simulation; such experiments will be referred to as “frequency-domain” simulation experiments. This technique was initially developed primarily to aid in factor screening and to perform a global sensitivity analysis of the input parameters in a simulation. It has since been developed into a method for identifying a meta-model for the simulation response surface. In this paper an overview of frequency domain simulation experiments is first presented. A new technique for including discrete factors in frequency domain experiments will also be discussed. Recently frequency domain optimality criteria have been discovered that can be used to identify local optima in the simulation response. The focus of this paper is on these optimality criteria. A brief discussion of how optimization algorithms might be designed using frequency domain simulation experiments is presented. This last topic is the subject of current research in frequency domain simulation experiments.

REFERENCES

• 1: Chatfield, C. (1984). The Analysis of Time Series: An Introduction (3rd.Ed.). Chapman and Hall.
• 2: Cogliano, V.J. (1982). "Sensitivity Analysis and Model Identification in Simulation Studies", Ph.D. dissertation, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY.
• 3: Jacobson, S., and Lee Schruben, (1986) "An Algorithm for Selecting Driving Frequencies for Frequency Domain Simulation Experiments", Technical Report, S.O.R.I.E., Cornell University, Ithaca, N~.
• 4: Kleijnen, J.P.C., A.J. van den BURG, and R.Th. van tier ttmm (1979). "Generalization of Simulation Results: Practicality of Statistical Methods", ~uropean Journal of Overations Research Vol. 3, No. 1. pp. 50-64.
• 5: K}eijnen, J.P.C. (1980). "Experimentation with Models: Statistical Design and Analysis Techniques", proceedings for "Simulation '80", interlaken, Switzerland, June 1980.
• 6: Rosenthal, R. (1978). "Combining Results of Independent Studies", Psychological Bullet in, Vol. 85, pp. 185-193~
• 7: Sanchez, P.J., and L. W. Schruben (1985) "Significant Factor Identification using Discrete Spectral Methods", Tech. Rpt. 654, SOEIE, Cornell University, NY.
• 8: Lee W. Schruben , Vincent James Cogliano, Simulation sensitivity analysis: A frequency domain approach, Proceedings of the 13th conference on Winter simulation, p.455-459, December 09-11, 1981, Atlanta, Georgia, United States
• 9: Schruben, L. and V. J. Cogliano, (1985)o "An Experimental Procedure for Simulation Response Surface Model Identification", Tech. Rpt. 669, SORIE, Cornell University, Ithaca, N.Y. 14853.
• 10: Smith, D. E., (1976), "Optimization of a Computer Simulation Response", Technica} Report 106-3, Desmatics Inc. State College, PA. 16801.

INDEX TERMS

Primary Classification:
  I. Computing Methodologies
  I.6 SIMULATION AND MODELING

Additional Classification:
  G. Mathematics of Computing
  G.1 NUMERICAL ANALYSIS

  I. Computing Methodologies
  I.6 SIMULATION AND MODELING

General Terms:
Design, Experimentation, Measurement, Performance, Theory