Improved Radiation Estimation at High Energy Facility for Public Health

Long-term exposure to ionizing radiation can lead to the development of cancer and other potentially fatal diseases. Estimating radiation levels at high-energy particle accelerators is crucial for ensuring the safety of personnel, the public, and the environment. This paper explores the application of transformer neural networks, focusing on five different attention mechanisms—Full Attention, Prob Attention, AutoCorrelation, Fourier Attention, and Wavelet Attention—for estimating radiation levels at the Thomas Jefferson National Accelerator Facility (JLab). The goal is to improve the accuracy of radiation estimation by training models using data from key radiation sensors. Four hundred trials were conducted for each attention mechanism, and performance was evaluated using metrics including R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. The results demonstrated that Prob Attention and AutoCorrelation outperformed the other mechanisms, offering substantial improvements over previous models. These findings highlight the importance of selecting and fine-tuning attention mechanisms in transformers for complex time-series analysis tasks in radiation monitoring.