基于抽样方法的核数据敏感性与不确定性分析

Background]: The input parameters for nuclear reactor physics calculations primarily include nuclear data, geometric dimensions, and fuel enrichment, among others. Among these, nuclear data is widely recognized as one of the most significant sources of uncertainty impacting the accuracy of reactor analysis. Consequently, quantitative research on the propagation of nuclear data uncertainty enhances the precision of nuclear reactor physics calculations, thereby ensuring the safety of nuclear reactor systems and improving their economic efficiency. [Purpose]: This study aims to introduce a highly adaptable method for analyzing nuclear data uncertainty. This method enables the perturbation of nuclear databases and generates perturbation files with broad applicability, which can be utilized to conduct uncertainty analysis across various computational tools and models. [Methods]: This study first employed the direct numerical perturbation method to conduct a sensitivity analysis on typical nuclide cross-sections, thereby validating the feasibility of the nuclear data sampling tool SANDY for perturbing nuclide cross-sections. Subsequently, within the range permitted by covariance, SANDY was utilized to perform multiple sampling perturbations on nuclide cross-sections, constructing multiple sets of perturbed nuclear databases for uncertainty analysis. To validate the effectiveness of the method, the study selected the BEAVRS benchmark, a commercial pressurized water reactor with complex geometric features, and calculated the effective multiplication factor (keff) for each perturbed database based on its hot zero-power core physics model using the Monte Carlo transport code OpenMC. To enhance computational efficiency, the sampling process was optimized: the parallelizable application perturbation module in SANDY was isolated and ProcessPoolExecutor was used to accelerate the generation of perturbation files; simultaneously, the OpenMC calculation process was parallelized using MPI technology on a supercomputing cloud platform. Finally, by fitting the distribution curve of keff , the impact of nuclide cross-section uncertainties in the nuclear database on keff was quantified. [Results]: Under the hot zero power condition of the BEAVRS benchmark, the sensitivity coefficients, ranked from highest to lowest, are as follows: 235U fission cross-section, 238U capture cross-section, 1H elastic scattering cross-section, 235U capture cross-section, and 238U fission cross-section. Among these, the 235U fission cross-section exhibits the highest sensitivity, with a sensitivity coefficient of 0.45. The uncertainties, ranked from largest to smallest, are as follows: 235U fission cross-section, 238U capture cross-section, 235U capture cross-section, and 238U inelastic scattering cross-section. Notably, the uncertainty of the 238U inelastic scattering cross-section has the most significant impact on keff, with a standard deviation of up to 91*10-5. [Conclusions]: This study proposes a nuclear data cross-section uncertainty analysis method based on SANDY. Through validation with the BEAVRS benchmark, the feasibility of this method under complex geometric conditions has been demonstrated. Furthermore, the perturbation files generated by SANDY support both ace and hdf5 formats, rendering them compatible with various Monte Carlo programs such as OpenMC and MCNP. This underscores the methods robust cross-tool and multi-model application capabilities, highlighting its strong adaptability. The research indicates that the sampling method based on SANDY not only effectively perturbs nuclear databases but is also compatible with multiple Monte Carlo computational tools, rendering it suitable for uncertainty analysis across models and under various operational conditions.