Optimization layer by layer networks for in-core fuel management optimization computations in PWRs

The optimization layer by layer (OLL) learning algorithm is applied to pred iction of assembly-wise power and burnup distribution, the critical soluble boron concentration, and the pin power peaking factor (PPPF) with core bur nup in the pressurized water reactor (PWR) of the Korean Nuclear Unit (KNU) II. It is shown that the OLL trained neural net works can predict core dep letion characteristics as accurately as the high-precision modern nodal met hod codes and that the OLL trained neural networks can compute core depleti on characteristics about 40 times faster than the modern nodal method code. The OLL networks are then utilized for determining the optimum fuel assemb ly (FA) loading; pattern (LP) of the equilibrium cycle KNU 11 PWR core by a simulated annealing (SA) scheme. By demonstrating that the FA LP optimizat ion by the SA scheme can be carried out within 10 to 15 min thanks to the s peedy neutronics evaluation of the OLL networks, it is proposed that the OL L networks can make a satisfactory substitute for core evaluation codes bas ed on modern nodal methods in in-core fuel management optimization computat ions where neutronics analysis for a large number of trial loading patterns has to be carried out. (C) 2001 Elsevier Science Ltd. All rights reserved.