Dimensionality reduction combined with particle swarm optimization algorithm for solving high-dimensional nuclear data target accuracy evaluation problem
Kaiwen Qin, Qintuo Zhang, Yibao Liu, Bo Yang, Xiaobin Tang, Nailiang Zhuang
Abstract
Nuclear data is a crucial source of uncertainty in calculating reactor physics. After the uncertainty of core physical parameters caused by nuclear data is quantified, it is necessary to evaluate the key nuclear data that require improvement to achieve the target accuracy of core physical parameters. Nuclear data comprises nuclides, reaction cross sections , and energy groups, which makes the assessment of nuclear data targets a high-dimensional nonlinear optimization problem with constraints. In high-dimensional space, data sparsity and computational complexity will increase exponentially, leading to algorithm performance degradation , as well as explosive growth in computing scale and time. To address the problem of high-dimensional data, this paper successfully transforms the high-dimensional nuclear data target accuracy evaluation problem into a low-dimensional problem by proposing two methods of dimensionality reduction: contribution factor screening and perturbation screening. The paper has achieved remarkable results, and these methods provide new ideas for solving the problem of target accuracy evaluation of high-dimensional nuclear data and offer useful references for research and practice in related fields.