An experimental procedure for simulation response surface model identification

ABSTRACT

An experimental method for identifying an appropriate model for a simulation response surface is presented. This technique can be used for globally identifying those factors in a simulation that have a significant influence on the output. The experiments are run in the frequency domain. A simulation model is run with input factors that oscillate at different frequencies during a run. The functional form of a response surface model for the simulation is indicated by the frequency spectrum of the output process. The statistical significance of each term in a prospective response surface model can be measured. Conditions are given for which the frequency domain approach is equivalent to ranking terms in a response surface model by their correlation with the output. Frequency domain simulation experiments typically will require many fewer computer runs than conventional run-oriented simulation experiments.

REFERENCES

• 1: Bloomfield. P. Fourier Analysis of Time Series: An Introduction. Wiley, New York, 1976.
• 2: Chatfield, C. The Analysis of Time Series: An Introduction. 3rd. ed. Chapman and Hall, London, 1984.
• 3: Cogliano. V.J. Sensitivity analysis and model identification in simulation studies. Ph.D. dissertation. School of Operations Research and Industrial Engineering, Cornell Univ., Ithaca, N.Y., 1982.
• 4: Cukier, R.I., Fortuin, CM., Shuler, K.E., Petschek, A.G., and Schaibly, J.H. Study of the sensitivity of coupled reaction systems uncertainties in rate coefficients. I: Theory. I, Chem. Phys. 59, 8 (Aug. 1973), 3673-3878.
• 5: Draper, N.R., and Smith, H. Applied Regression Analysis. Wiley, New York, 1966.
• 6: Jacobson, S., Boss, A., and Schruben, L. The frequency selection problem in frequency domain experiments. Tech. Rep. 714, School of Operations Research and Industrial Engineering, Cornell Univ., Ithaca, N.Y., Aug. 1986.
• 7: Jenkins, G.M.. and Watts, D.G. Spectral Analysis and Its Applications. Holden-Day, San Francisco, Calif., 1968.
• 8: Kleijnen, }.P.C. Experimentation with models: Statistical design and analysis techniques. In Progress in Modelling and Simwlation, F.E. Cellier, Ed. Academic Press, New York, 1982, pp. 173-185.
• 9: Kleijnen. J.P.C., and van den Burg, A.J., and van der Ham. R.T. Generalization of simulation results: Practicality of statistical methods. Eur. J. Oper. Res. 3, 1 (1979), 50-64.
• 10: Lennart Ljung, System identification: theory for the user, Prentice-Hall, Inc., Upper Saddle River, NJ, 1986
• 11: Neave. H.R. Observations on "Spectral Analysis of Short Time Series-A Simulation Study" by Granger and Hughes. ,'. R. Stat. Sot. A, 135 (1972), 393-405.
• 12: Rosenthal, R. (1978). Combining results of independent studies. Psychol. Bull. 85, 1 (Jan. 1978), 185-193.
• 13: Sanchez, P.J. Design and analysis of frequency domain experiments. Ph.D. thesis, School of Operations Research and Industrial Engineering, Cornell Univ., Ithaca, N.Y.. 1987.
• 14: Lee Schruben, Simulation optimization using frequency domain methods, Proceedings of the 18th conference on Winter simulation, p.366-369, December 08-10, 1986, Washington, D.C., United States [doi>10.1145/318242.318460]
• 15: 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
• 16: Schruben. L.W., and Margolin, B.H. Pseudo-random number assignment in statistically designed simulation and distribution sampling experiments. 1. Am. Stat. Assoc. 73, 363 (Sep. 1978), 504-520.
• 17: Suri. R. Implementation of sensitivity calculations on a Monte Carlo experiment. 1. Optim. Theory Appt. 40,4 (Aug. 1983), 625-630.

INDEX TERMS

Classification:
  G. Mathematics of Computing
  G.3 PROBABILITY AND STATISTICS
      Subjects: Probabilistic algorithms (including Monte Carlo)

  I. Computing Methodologies
  I.6 SIMULATION AND MODELING

General Terms:
Design, Experimentation