The application of digital twin in the field of nuclear energy engineering has attracted increasing attention. To meet the speed and accuracy requirements of reactor digital twin online calculation, rapid prediction of reactor physical field is required. This study focuses on Integral PWR-200, the thermal hydraulic of the primary circuit and core physical coupling model is established by using RELAP5 and REMARK programs. To reconstruct the physical field of reactor power distribution, neutron flux density distribution, and coolant temperature distribution, a non-intrusive reduced-order model combining Proper Orthogonal Decomposition (POD) and Back Propagation Neural Network based on Differential Evolution optimization (DE-BPNN) is proposed. Firstly, POD is used to extract the low-dimensional reduced-order base modes of the physical field, then a multi-layer neural network model is constructed to map the operating parameters to mode coefficients using BPNN. Addressing the slow convergence rate and the tendency to fall into local optima of BPNN model, the Differential Evolution (DE) algorithm is applied to globally optimize the initial weights and thresholds of BPNN. The results show that compared with conventional BPNN model, DE-BPNN model has faster convergence speed and higher prediction accuracy. The prediction method proposed in this paper provides a new technical approach for physical field prediction in nuclear reactors.

Proper Orthogonal Decomposition; Model order reduction; Differential Evolution algorithm; BP neural network; Neutronics and thermal-hydraulics coupling