Experiments and fuel reloads optimization for the Jules Horowitz Reactor (JHR) using a genetic algorithm combined with a convolutional neural network

This paper presents the experiments and fuel reloads optimization of the Jules Horowitz research Reactor (JHR) using a genetic algorithm coupled with a convolutional neural network. At the end of a cycle, the most burned fuel elements will be removed and replaced by fresh fuel elements. Since the purpose of this reactor is to conduct experiments, it is crucial to reach experimental performances. With this aim, the goal was to find the best location for each fuel element for the next cycle, i.e. the optimal core configuration that satisfies not only operations and safety constraints but also experimental goals. The analysis was based on an algorithm using two Python libraries: PyGAD for the genetic algorithm and TensorFlow for the convolutional neural network, thus providing a simple implementation of the method. It has been shown the convolutional neural network predicts the core parameters in 12 ms with a low error. Moreover, the genetic algorithm converges toward an optimal core configuration in a very short computation time (∼7.5 minutes) as well. This paper demonstrates that the combination of a convolutional neural network and a genetic algorithm allows obtaining the optimal core configuration for a research reactor in particular to maximize its experimental performances.