Due to the complex structures associated with neutron resonance cross sections, their accurate evaluation has received considerable attention in the field of nuclear data research. The traditional R-matrix method still faces some difficulties in evaluating the neutron resonance data, especially in briefly reproducing the high-frequency oscillating cross sections. Recently, the applications of machine learning methods in nuclear physics have been expanding. In this paper, a novel Phase Shift Deep Neural Network (PSDNN) method, which not only overcomes the limitations of other machine learning methods in fitting the high-frequency oscillating data, but also is more concise than the R-matrix method, is developed to reproduce the neutron resonance cross sections. The results show that PSDNN method can simultaneously reproduce the low and high-frequency oscillating cross sections for the U() reaction with high accuracy and efficiency. Moreover, from an algorithmic point of view, the PSDNN method lays a solid foundation for further fine-grained processing of experimental data and extraction of critical neutron resonance parameters, opening up new possibilities for practical applications in nuclear data research.

中文翻译：

---

用于研究 235U(n,f) 反应共振截面的相移深度神经网络方法


由于中子共振截面结构复杂，其准确评估在核数据研究领域受到了广泛关注。传统的R矩阵方法在评估中子共振数据时仍然面临一些困难，特别是在简要再现高频振荡截面方面。近年来，机器学习方法在核物理中的应用不断扩大。本文提出了一种新颖的相移深度神经网络（PSDNN）方法，该方法不仅克服了其他机器学习方法在拟合高频振荡数据方面的局限性，而且比R矩阵方法更简洁。重现中子共振截面。结果表明，PSDNN方法可以高精度、高效地同时再现U()反应的低频和高频振荡截面。而且，从算法角度来看，PSDNN方法为进一步细粒度处理实验数据和提取关键中子共振参数奠定了坚实的基础，为核数据研究的实际应用开辟了新的可能性。