Predictive Models Informed by Simulation, Measurement, and Surrogates
In this chapter techniques for combining simulation and experiments through surrogate models are presented. We begin with the classic problem of using experimental data to fix parameters in a simulation. To do this properly, we require Markov Chain Monte Carlo (MCMC) sampling, and this method is discussed in Sect. 11.2. Section 11.3 using MCMC to estimate calibration parameters. The formalism of Kennedy and O’Hagan is then used to introduce a discrepancy function that contains the difference between simulation and experiment, and MCMC is used to estimate both the discrepancy and surrogate simultaneously. Finally, Sect. 11.5 shows how a hierarchy of fidelities can be used to make predictions.
- Goh J (2014) Prediction and calibration using outputs from multiple computer simulators. PhD thesis, Simon Fraser UniversityGoogle Scholar