Comparative study on gradient-free optimization methods for inverse source-term estimation of radioactive dispersion...

Area of Science:

• Environmental Science
• Nuclear Engineering
• Computational Science

Background:

• Accurate radionuclide source term estimation is critical for environmental monitoring and emergency response.
• Gradient-free optimization algorithms are increasingly applied to complex inverse problems in environmental modeling.

Purpose of the Study:

• To compare the performance of Ensemble Kalman Inversion (EKI), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA) for radionuclide source term estimation.
• To investigate the impact of source complexity (single vs. multiple radionuclides/sources) and measurement configurations on estimation accuracy.
• To identify strategies for improving source term estimation in environmental radioactivity monitoring.

Main Methods:

• Simulated diverse radionuclide release scenarios, including single and multiple sources with varying compositions.
• Applied and compared three gradient-free optimization algorithms: EKI, PSO, and GA.
• Utilized the Gaussian plume model for atmospheric dispersion under steady-state conditions.

Main Results:

• EKI demonstrated competitive convergence, accuracy, and runtime compared to PSO and GA, especially with GPU parallelization.
• Estimating multiple radionuclides from a single source was more challenging due to limited gamma dose rate information.
• Increasing observation stations did not consistently improve solutions for ill-posed inverse problems.

Conclusions:

• EKI is a promising algorithm for radionuclide source term estimation, offering computational efficiency.
• Relative error significantly impacts multi-radionuclide estimation accuracy from gamma dose measurements.
• The study provides a foundation for integrating dispersion models with real-time data for advanced monitoring systems.