Research on the calculation method of neutron diffusion spatiotemporal kinetics equation based on the time series convolutional neural network.

The digital twin system of nuclear reactors is the foundation for the intelligent operation and maintenance of reactors and the visualization of operational data. Because of the current problem of low computational accuracy and lack of interpretability of deep learning algorithms for complex physical fields, this paper designs a deep learning and numerical analysis fusion algorithm based on a time series convolutional neural network algorithm, which can be used to solve the neutron diffusion spatiotemporal kinetics equation. The algorithm is based on convolutional neural network architecture, using finite volume method, constructing fixed parameter convolution kernel, and reconstructing loss function to replace convolutional neural network structure. This algorithm retains the convolutional neural network's structure and ensures that all steps have clear physical meaning. Finally, the algorithm logic and GPU acceleration are implemented on the Pytorch platform, and the algorithm's feasibility is verified using the simplified Tsinghua High Flux Reactor(THFR). The results show that the relative errors between the time series convolutional neural network and OpenMC in calculating the neutron flux density and the fission reaction rate are approximately within the range of 1% to 5%. The calculation speed of the time series convolutional neural network for neutron flux density is 1/20 of OpenMC, and the calculation speed of the expansion matrix is 1/8 of the NumPy library. Therefore, the time series convolutional neural network not only ensures the accuracy of calculation and improves the speed of operation, but also ensures the interpretability of the algorithm. In addition to being used to solve the neutron diffusion spatiotemporal kinetics equation, this algorithm is mainly used to develop the virtual physics engine of the digital twin system. Using this algorithm to implement the virtual reality method of the neutron field provides theoretical guidance and algorithm support for the development of the THFR digital twin system. [ABSTRACT FROM AUTHOR]