Optimization method of fuel-reloading pattern for PWR based on the improved convolutional neural network and genetic algorithm

Abstract

In this paper, the optimization method of fuel-reloading pattern for PWR has been studied based on the improved convolutional neural network (CNN) and genetic algorithm (GA). It is very important to search out the optimized fuel-reloading pattern to guarantee the safety and economy of the nuclear power plants. During the optimization, large number of fuel-reloading patterns should be evaluated, providing the core parameters (including the cycle length, power-peak factors and so on) to the optimization algorithm to search for the optimized pattern. In our study, the CNN was improved with the advanced Inception-ResNet structure and applied to train the rapid-evaluation model, which can receive the fuel-reloading patterns and feedback corresponding core parameters with sufficient accuracy and very-high efficiency. The GA was applied as the optimization algorithm to search for the optimized fuel-reloading pattern. This proposed optimization method has been applied to the optimization of fuel-reloading pattern for the CNP1000-type PWR reactor operated in China. It can be observed that the CNN can evaluated the core parameters of one-single fuel-reloading pattern in about 0.0005 s and the averaged evaluation errors smaller than 0.6%; the GA can search the optimized fuel-reloading pattern in about 20 min. The study in this paper indicated that the combination of CNN and GA can provide the optimization of fuel-reloading pattern for PWR in very-short time, which can be applied to improve the safety and economy of the nuclear power plants.