Studies on fast neutron multiplicity measurement based on neural network

In the measurement of fast neutron multiplicity, the multiplicity counting rates of neutrons, including singles, doubles, and triplets, are often substituted into the measurement equation to solve quality problems. To simplify the solution process and directly obtain the sample quality through S, D, and T, a neural network and multivariate nonlinear fitting are used for analysis. First, multiple sets of data are measured through a detection system built with Geant4. After the training of the back propagation neural network, the corresponding relationship between S, D, T, and m is established. It is verified that there are different degrees of discrepancy between the predicted values of the neural network and the simulated and theoretical values. To improve the accuracy of predictions, genetic algorithm optimization and M coefficient correction are introduced. To analyze the stability of the neural network model, a 10% error perturbation is introduced for S, D, and T. The double rate has the greatest influence on the deviation of the predicted value, indicating that the double rate is the key parameter in the analysis of neutron multiplicity. On this basis, a functional relationship is obtained through multivariate nonlinear fitting, and the validation of the fitting equation is verified by simplifying the fast neutron multiplicity measurement technology equation.