Prediction of spent nuclear fuel isotopic composition for the VVER-1000 reactor utilizing regression tree

Abstract

In this paper, a predictive model using regression trees is represented, which can predict the depleted isotopic composition (IC) of 21 isotopes of the VVER-1000 reactor based on effective days, initial enrichments, percentage of gadolinium absorber, and zones of fuel elements within the fuel assembly (FA). First, multiple regression trees (RT) are generated for a particular isotopic, and predictions of the isotopic composition of that isotope have been made by averaging the results from all trees. Multiple regression trees (RT) models have been generated by applying optimal cross-validation (CV) fold number, which is identified by investigating four parameters of the model. The model’s performance has been evaluated by the root mean squared error (RMSE), R2 value, and p-value for the heteroscedasticity test. It has been found that the percentage of RMSE for the training set and test set is less than 0.15%, the R2 value is close to unity, and the p-value for the heteroscedasticity test is higher than 5% with a 95% confidence interval. After completing the training process takes a fraction of a second to compute isotopic composition.