Background Neutron multiplicity measurement technology, as a core method in the field of non-destructive testing, plays a critical role in determining the mass of fissionable material (²³⁵U). However, it faces technical bottlenecks such as prolonged measurement cycles and measurement deviations under non-ideal conditions.

Purpose This paper aims to explore feasible pathways for integrating neutron multiplicity measurement methods with neural network technology. The goal is to provide new research perspectives for advancing neutron multiplicity measurement technology toward greater efficiency and intelligence.

Methods Leveraging Geant4 and MATLAB software, an Active Well Coincidence Counter (AWCC) simulation model was constructed to achieve high-precision simulation of the entire active neutron multiplicity measurement process. Building upon this, three neural networks—Backpropagation Neural Network (BPNN), Convolutional Neural Network (CNN), and Long Short-Term Memory network (LSTM)—were developed using the PyTorch framework to analyze and investigate neutron multiplicity distribution data.

Results Compared with traditional calculation methods based on the active neutron multiplicity equation, neural network models represented by CNN and LSTM demonstrate significant advantages in measurement accuracy and efficiency. Specifically, in terms of relative error metrics, neural network models can reduce errors to lower levels; in the time dimension of measurement, they substantially shorten data processing cycles, effectively overcoming the timeliness constraints inherent to traditional approaches.

Conclusions This achievement fully validates the theoretical feasibility and technical superiority of the neural network-based neutron multiplicity measurement approach, providing a novel solution for advancing neutron multiplicity detection toward greater efficiency and intelligence. Subsequent work will enhance the adaptability and noise resistance of neural network models for complex data by increasing simulation scenario complexity and introducing diversified factors such as noise interference and geometric variations. Meanwhile, building upon simulation studies, physical experimental validation will be conducted using AWCC instrumentation to drive the transition of neural network-based neutron multiplicity measurement technology from simulation to engineering application.