HEGM: Hierarchical Ensemble Generation Model for nuclear reaction cross sections generation

Accurate and reliable nuclear reaction cross section data are crucial for nuclear reactor design, nuclear energy development, and nuclear safety assessments. Traditional methods for generating nuclear data face significant challenges, including high costs, long time frames, and incomplete coverage. Recent advances in machine learning (ML) offer new opportunities for nuclear data generation, but existing methods struggle with the scarcity of experimental data, limiting their ability to generate high-precision and broadly applicable cross section data. This paper introduces the Hierarchical Ensemble Generation Model (HEGM), a novel AI-driven approach to nuclear reaction cross section generation. HEGM combines transfer learning, meta-learning, prototype networks, and generative adversarial networks to address the challenges of sparse data and improve predictive accuracy. We evaluate HEGM’s performance on isotopes ¹⁶O and ²³⁸U and compare it with conventional machine learning models, including K-Nearest Neighbors, Random Forest, and Artificial Neural Networks. The experimental results demonstrate that HEGM significantly outperforms traditional models, achieving a 25% reduction in Mean Squared Error (MSE), a 20% reduction in Mean Absolute Error (MAE), and a 15% improvement in the coefficient of determination ($R^2$) compared to the baseline ENDF/B-VIII.0 database and EXFOR experimental data. Specifically, HEGM achieved a 28% reduction in MAE for ¹⁶O, and a 15% reduction for ²³⁸U. These results highlight the model’s superior accuracy and robustness, particularly in resonance regions where experimental data is sparse. HEGM’s hierarchical ensemble structure allows for enhanced predictive performance, making it an effective tool for nuclear data generation in data-limited regions. The promising results suggest that AI-driven approaches like HEGM can provide a powerful alternative to traditional nuclear data evaluation methods. Future work will explore further model enhancements, including expanding to additional isotopes and nuclear reactions, as well as integrating HEGM with reactor simulation models to improve reactor design and nuclear energy research.