SCALABLE ALGORITHMS FOR UNCERTAINTY QUANTIFICATION OF MULTI-PHYSICS LIGHT WATER REACTOR PROBLEMS WITH FEEDBACK EFFECT

摘要

In conjunction with modern large scale and high fidelity nuclear reactor simulators, it is important to provide the best estimates predictions along with their associated uncertainties. However, quantifying the uncertainty in multi-physics large scale applications with feedback effect often involves a high computational cost. Therefore, scalable algorithms for large scale Uncertainty Quantification (UQ) problems have received considerable attention. This work builds upon previous efforts to develop reduced order modeling based algorithms for large scale UQ problems. Previous efforts explored the application of the subspace methods to both linear and non-linear single-physics UQ problems. In this work, subspace based algorithms for linear and non-linear uncertainty propagation have been extended to large scale multi-physics applications with feedback effect. A previously developed efficient algorithm for Reduced Order Modeling (ROM) is extended to multi-physics UQ problems with feedback, along with the capability to estimate the uncertainty contribution of each of the uncertainty sources. All verifications are performed using CASL progression problem number 6, which is a 3 dimensional PWR assembly depletion problem. Moreover, the proposed methods are used to propagate the uncertainty for the depletions calculations of a core wide problem represented by CASL progression problem number 9. The uncertainty quantification is completed utilizing a new tool kit; the Reduced Order Modeling based Uncertainty/Sensitivity Estimation (ROMUSE).