Prediction of 3D nuclear reactor's operational parameters from 2D fuel lattice design information: A data mining approach

Abstract

In this paper the estimation of 3D BWR nuclear reactor parameters starting from 2D data is presented. The 3D parameters are obtained through a steady state simulation of the nuclear reactor's operation, namely; thermal limits and cold shutdown margin. Data mining techniques were applied to build decision trees in order to estimate those 3D reactor parameters. The decision trees were built using local power peaking factor, infinite multiplication factor and relative power values of the fuel lattices calculated at the beginning of its life, the number of fuel pins containing gadolinia, uranium enrichments and gadolinia concentration for pins. Using the CASMO-4/SIMULATE-3 system a total of 18,225 operation cycles were simulated in order to generate the dataset for the construction of decision trees. As a result, it was possible to estimate thermal limits with relative errors lower than 5%. The estimation for cold shutdown margin was lower than 200 pcm. Decision trees use 12, 29, and 36 variables to predict SDM, FLPD and MAPRAT values respectively. Decision trees can estimate those core parameters in 25 s against several hours spent by CMS codes. However, the obtained model is not aimed at replacing core simulators to do fuel reloads licensing. It should be considered instead as a tool for a preliminary and fast assessment in an optimization process. Afterwards, the potential solutions must be reassessed and validated with CMS codes execution.