High-risk nuclide screening and parameter sensitivity analysis based on numerical simulation and machine learning.

Area of Science:

• Environmental Science
• Nuclear Safety
• Computational Science

Background:

• Nuclear accidents release hazardous radionuclides, necessitating rapid identification of high-risk isotopes.
• Traditional screening methods have limitations, requiring more effective approaches for nuclear emergency response.

Purpose of the Study:

• To develop and compare machine learning models for screening high-risk radionuclides.
• To predict the timing and extent of groundwater contamination by these nuclides.
• To analyze the sensitivity of model predictions to key environmental parameters.

Main Methods:

• Comparison of Support Vector Machine (SVM), Random Forest (RF), and Back Propagation Neural Network (BPNN) algorithms.
• Application of machine learning for classification (screening) and regression (prediction).
• Sensitivity analysis of initial leakage concentration ratio (C0/Cp), distribution coefficient (Kd), and decay coefficient (λ).

Main Results:

• Random Forest (RF) achieved the highest accuracy in screening high-risk nuclides, with Kd > λ > C0/Cp influence.
• Back Propagation Neural Network (BPNN) demonstrated superior performance in predicting groundwater pollution timing.
• BPNN predictions showed positive correlation with Kd and λ, and negative correlation with C0/Cp, with Kd > C0/Cp > λ influence.

Conclusions:

• Machine learning, particularly RF and BPNN, offers a powerful tool for nuclear accident consequence assessment.
• Parameter influence varies between screening and prediction tasks, highlighting the complexity of radionuclide transport.
• Understanding parameter interactions is crucial for accurate environmental risk assessment following nuclear events.