The article proposes a method for assessing the neutron energy spectrum and personnel effective dose rate based on the readings of a Bonner spectrometer (BSS) for high-energy neutron fields. Neutron flux density can be obtained from BSS measurements by solving the system of Fredholm integral equations of the first kind. The paper unfolds the spectra using supervised “random forest” machine learning algorithm with optimisation of the model hyperparameters.

The model was trained and tested on a database of 251 spectra for various power facilities (80% of data was used for training the model, while 20% was used for testing it). The input features of the model were the spectrometer readings for BSS moderator spheres and the categorical feature “spectrum type” describing the facility and conditions under which the spectrum was obtained. The output parameters of the model were the spectrum description in the form of a histogram for 60 energy values and the dose rate calculated from the spectrum for the corresponding conversion factor.

Since the dataset of real spectra is small, database of 104 synthetic data generated using the Frascati Unfolding Interactive Tool method was developed. The second model for this synthetic dataset was trainted and compared with the first one. The effect of the error in the initial data on the spectrum and the dose rate obtained from it was estimated by the Monte Carlo method using random samples. The test dataset showed that the unfolded spectra are close in nature to the original ones and have a high correlation with them.

The paper proposes a method for selecting the optimal number of moderator spheres based on the explainable artificial intelligence method “Shapley additive explanation” (SHAP). The SHAP method was used to demonstrate the degree of influence of measurements with moderator spheres of different diameters on the spectrum prediction. It was shown that resulting spectrum is most influenced by measurements with moderator sphere of 10″. Optimisation of the choice of spheres leads to a decrease in the personnel doses during measurements. The model was trained and calculations were performed on the JINR Multifunctional Information and Computing Complex.