Method research on equilibrium cycle reloading pattern optimization for the HPR1000 reactor core using deep learning-enhanced multi-objective genetic algorithm

A novel methodology for designing and optimizing a PWR’s equilibrium cycle reloading pattern using deep learning and a multi-objective genetic algorithm (MOGA) has been developed. The deep-learning model efficiently and accurately predicted reactor physics parameters, particularly fuel assembly burnups at the end of the cycle (EOC), and formed a fitness function. The fitness function takes the absolute difference between the conformable fuel assemblies’ burnups at the beginning of the cycle (BOC) and the EOC, which narrows down the potential equilibrium reloading patterns. The deep-learning model was coupled with MOGA, which simultaneously optimized multiple objectives for the design and optimization of an equilibrium cycle. Applied to the HPR1000 reactor, the method achieved the first equilibrium cycle length of 473.1 Effective Full Power Days (EFPDs) and an average of 471.1 EFPDs over ten cycles, meeting all the parameters of reactor safety design criteria.