Surrogate-based multi-experiment calibration of the BISON fission gas behavior model

Abstract

The physics-based fission gas behavior model recently implemented in BISON has demonstrated significant improvements over previous models but there remains considerable uncertainty in the predicted behaviors due to the intrinsic uncertainties of several parameters used in the model. One of the methods to reduce the uncertainty in fission gas behavior prediction is to utilize experimental data from multiple integral experiments to adjust the individual model parameters. The research performed here was in support of that effort and developed a comprehensive framework which utilized surrogate models for BISON in order to reduce the computational intensity of the resulting optimization problem. Both single-valued and high dimensional outputs were investigated using data from several experiments of the Risø-3 database. The Derringer’s desirability function method was used to combine multiple outputs from separate surrogate models into a single objective function and a weighting factor was introduced to provide analysts a tool to properly weight the importance of each output from each experiment in the optimization process. The method was applied for the multi-experiment calibration problem of the BISON fission gas behavior model. An optimized set of fission gas behavior model parameters was obtained in a demonstration of the multi-experiment calibration method, which showed significant accuracy improvement of the BISON fission gas behavior predictions. By using the surrogate-based approach, the computational cost was reduced by more than two orders of magnitude.