Sequential Designs for Simulation Experiments: Nonlinear Regression Metamodeling

M.I. Reis dos Santos and P.M. Reis dos Santos (Portugal)


metamodels; nonlinear regression; simulation; design of experiments.


In this paper we propose a method to select an experimen tal design for estimating nonlinear simulation metamodels. Through a careful selection of design points, the method provides better ļ¬tting results than equally spaced point selection, with the same simulation effort. This method accounts for the input/output function of the simulation model, possibly a mathematical function nonlinear in the parameters. In spite of the fact that the paper concentrates on nonlinear regression metamodels, the method may be applied to other type of metamodels. The procedure is easy to construct (so, it is attractive to be used in practice) and focus on simulations scenarios in sub-regions where the in put/output behavior has more interest. This procedure is il lustrated with an application to a automobile parts factory. Finally, we draw some conclusions.area.

Important Links:

Go Back