Implementation of Fast Emulator-based Code Calibration
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States). Risk & Reliability Analysis
Calibration is the process of using experimental data to gain more precise knowledge of simulator inputs. This process commonly involves the use of Markov-chain Monte Carlo, which requires running a simulator thousands of times. If we can create a faster program, called an emulator, that mimics the outputs of the simulator for an input range of interest, then we can speed up the process enough to make it feasible for expensive simulators. To this end, we implement a Gaussian-process emulator capable of reproducing the behavior of various long-running simulators to within acceptable tolerance. This fast emulator can be used in place of a simulator to run Markov-chain Monte Carlo in order to calibrate simulation parameters to experimental data. As a demonstration, this emulator is used to calibrate the inputs of an actual simulator against two sodium-fire experiments.
- Research Organization:
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE Office of Nuclear Energy (NE), Nuclear Reactor Technologies (NE-7); Nuclear Regulatory Commission (NRC) (United States)
- DOE Contract Number:
- AC04-94AL85000
- OSTI ID:
- 1431438
- Report Number(s):
- SAND2016--7862; 646589
- Country of Publication:
- United States
- Language:
- English
Similar Records
Emulator-based Bayesian calibration of a subglacial drainage model
Design of Experiments, Model Calibration and Data Assimilation