A Reduced Order Model Based on ANN-POD Algorithm for Steady-State Neutronics and Thermal-Hydraulics Coupling Problem

Abstract

The neutronics and thermal-hydraulics (N/TH) coupling behavior analysis is a key issue for nuclear power plant design and safety analysis. Due to the high-dimensional partial differential equations (PDEs) derived from the N/TH system, it is usually time consuming to solve such a large-scale nonlinear equation by the traditional numerical solution method of PDEs. To solve this problem, this work develops a reduced order model based on the proper orthogonal decomposition (POD) and artificial neural networks (ANNs) to simulate the N/TH coupling system. In detail, the POD method is used to extract the POD modes and corresponding coefficients from a set of full-order model results under different boundary conditions. Then, the backpropagation neural network (BPNN) is utilized to map the relationship between the boundary conditions and POD coefficients. Therefore, the physical fields under the new boundary conditions could be calculated by the predicated POD coefficients from ANN and POD modes from snapshot. In order to assess the performance of an ANN-POD-based reduced order method, a simplified pressurized water reactor model under different inlet coolant temperatures and inlet coolant velocities is utilized. The results show that the new reduced order model can accurately predict the distribution of the physical fields, as well as the effective multiplication factor in the N/TH coupling nuclear system, whose relative errors are within 1%.