A new methodology to speed-up fuel lattice design optimization using decision trees and new objective functions.

• In this paper, a methodology to speed up fuel lattices optimization is shown. • A recurrent neural network to explore the solutions space was used. • Two new objective functions were studied. • Decision trees to estimate variables involved in objective functions were constructed. • This methodology was tested obtaining a good NN-DT performance. In this paper a new methodology to speed up the fuel lattice design optimization in a BWR is explored. In previous works, fuel lattice optimization was made using LPPF (Local Power Peaking Factor) at the beginning of the fuel lattice life. However, undesirable LPPF vs. fuel lattice exposure behaviors were observed. Due to this, LPPF vs. fuel lattice exposure was calculated through out fuel lattice life burnup. From a computational point of view, such calculation is very expensive when done using the CASMO-4 code. A new methodology to speed up the optimization was proposed based on two aspects: in one side, using objective functions that take into account LPPF vs. fuel lattice exposure and residual gadolinia; and in other side, using decision trees to estimate some fuel lattice parameters in a fast and reliable way. It could be verified that decision trees estimations had the enough reliability to be used into an optimization process to discard bad fuel lattice configurations and speed up the optimization process. At the end of this process, CASMO-4 code is used to calculate the final fuel lattice parameters. In this way, fuel lattice optimization time was reduced from 6 hours to 15 minutes obtaining good LPPF vs exposure behaviors. [ABSTRACT FROM AUTHOR]