Integrating the deep learning and multi-objective genetic algorithm to the reloading pattern optimization of HPR1000 reactor core

The deep learning and multi-objective genetic algorithm were employed to optimize the fuel reloading pattern for HPR1000, a state-of-the-art nuclear power reactor designed and operated in China, also known as Hualong-1. In this study, the deep-learning algorithm was applied to establish the rapid evaluator for fuel-reloading patterns, for which the random samples between fuel-reloading patterns and corresponding key core parameters were generated by our home-developed nuclear-design code, named Bamboo-C. The advanced machine-learning platform TensorFlow was utilized for the deep-learning model. Then, the multi-objective genetic algorithm was applied to search the optimal fuel-reloading patterns, combined with the rapid evaluator to evaluate the key core parameters in a very short time. The DAKOTA toolkit was employed for optimization using a multi-objective genetic algorithm, for which the cycle length and power-peak factors were selected as the target parameters to establish the fitness function. For verification and application, the above method has been applied to the fuel-reloading optimization for Cycle 2 of HPR1000 operated in China. The optimization pattern results in an extension of the cycle length by about 21 EFPD (Effective Full Power Day), keeping all the key safety parameters satisfying corresponding safety criteria.