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

Area of Science:

• Nuclear Physics
• Computational Science
• Materials Science

Background:

• Nuclear reaction cross section data are vital for nuclear applications.
• Traditional data generation methods are costly and time-consuming.
• Existing machine learning (ML) methods struggle with sparse experimental data.

Purpose of the Study:

• Introduce the Hierarchical Ensemble Generation Model (HEGM), an AI-driven approach for nuclear reaction cross section generation.
• Address data scarcity and improve predictive accuracy in nuclear data generation.
• Provide a more efficient and accurate alternative to traditional methods.

Main Methods:

• Developed the Hierarchical Ensemble Generation Model (HEGM).
• Combined transfer learning, meta-learning, prototype networks, and generative adversarial networks.
• Evaluated HEGM on 16O and 238U isotopes, comparing with conventional ML models and databases.

Main Results:

• HEGM demonstrated superior performance over traditional ML models.
• Achieved a 25% reduction in Mean Squared Error (MSE) and 20% reduction in Mean Absolute Error (MAE).
• Showcased significant accuracy improvements, especially in data-sparse resonance regions.

Conclusions:

• HEGM offers a powerful AI-driven solution for nuclear data generation, particularly in data-limited scenarios.
• The model's hierarchical ensemble structure enhances predictive performance.
• HEGM shows promise for advancing nuclear reactor design and energy research.