A BP neural network-based model for the neutron ambient dose equivalent estimation along flight routes in China

This study developed a BP neural network model to approximate calculations of route-integrated neutron ambient dose equivalent, along commercial flight routes in China. The CARI-7A was used to calculate the neutron ambient dose equivalent (H∗(10)), for 696 domestic flight routes at cruising altitudes ranging from 8000 to 12 000 m. Based on these CARI-7A-derived results, a Back Propagation (BP) neural network was trained using input variables including cruising altitude, cruising time, and geographical information. Trained, validated, and tested using 14 616 data points to ensure predictive accuracy, the BP neural network demonstrated excellent performance with an adjusted R² of 0.9998 and a minimal mean square error (MSE) of 0.0002; when compared with measured neutron H∗(10) data from 30 domestic flight routes, the model predictions showed an average relative deviation of 14.00%, with 80.0% of the routes falling within ±20% of the measured values. The developed BP neural network model provides a rapid approximation of neutron H∗(10) for Chinese air routes. Rather than replacing direct radiation measurements, the model may serve as a supplementary tool for preliminary route-level assessment of neutron H∗(10), particularly for routes not conveniently covered by direct measurements or CARI-7A airport settings. Further validation with a larger number of measured flights is still needed before broader operational application.