Study of Assembly Nuclide Density Prediction Based on Data Mining Technology

Abstract

The DRAGON program was used to calculate 9600 samples, and the nuclide densities of 235U, 238U, 239Pu, 241Pu, 137Cs, 244Cm and 154Nd nuclides were used as the prediction parameters. Linear regression model, regression tree model constructed based on decision tree, multilayer perception (MLP) model and random forest model were selected to carry out model training. Pearson correlation coefficient (PCC), mean absolute error (MAE), relative absolute error (RAE) and relative root mean square error (RRSE) were chosen to evaluate the fitting effect of the models; the trained models were used in the test set for the target. The trained models were used to predict the target nuclides in the test set, and their prediction accuracy was evaluated by relative errors. The results show that the training time of the data models is less than 3 s. After the evaluation of the selected parameters, the MLP model has the best training effect among the four models for all the predicted kernels, and its correlation is above 0.999. The average deviation of the MLP model for all the predicted kernels is less than 1%. This paper initially verifies the feasibility of data mining techniques in predicting the density of assembly nuclei.