Simulation metamodels for modeling output distribution parameters

  • Authors:
  • Isabel R. Santos;Pedro R. Santos

  • Affiliations:
  • Technical University of Lisbon (IST), Lisboa, Portugal;Technical University of Lisbon (IST), Lisboa, Portugal

  • Venue:
  • Proceedings of the 39th conference on Winter simulation: 40 years! The best is yet to come
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Metamodels are functions with calibrated parameters, used as abstractions and simplifications of the simulation model. A metamodel exposes the system's input-output relationship and can be used as an analysis tool for solving optimization problems or as a surrogate for building blocks in larger scale simulations. Our approach is to analyze statistically the response by modeling the normal distribution mean and variance functions, in order to better depict the problem and improve the knowledge about the system. The metamodel is checked using the confidence intervals of the estimated distribution parameters, and new design points are employed for predictive validation. An example is used to illustrate the development of analysis and surrogate metamodels.