Reactor physics fast calculation method based on model order reduction and machine learning

The rapid development of artificial intelligence technology has provided new ideas and methods for reactor physics calculations. Based on AI technology, a fast calculation method for reactor physics has been established, which combines model order reduction and machine learning to address the challenges of excessive parameter quantities in machine learning-based parameter prediction. During the training process, the full-order model is established using the two-step core nuclear design software package TORCH, and the model order reduction theory is applied, which are then trained using the random forest machine learning method. In the prediction process, the basis weight coefficients are rapidly calculated for specific input parameters, and the core distribution results are reconstructed. A reactor physics fast calculation program has been developed and verified using a M310-type pressurized water reactor nuclear power plant with 9522 samples. All results show that the fast calculation method based on model order reduction and machine learning has good computational efficiency and accuracy. The calculation time can be reduced to 0.1 s and the proportion of samples with less than 1 % deviation in various core physics parameters is higher than 90 %.