Method Research on Inversion Optimization of Core Physical Model and Industrial Application with Measured Data Fusion

Abstract

During the operation of pressurized water reactor (PWR), the fuel assemblies would inevitably occur the phenomenon of bowing, due to the factors such as axial irradiation growth, high speed impact of coolant during operation and so on. This phenomenon is intuitively manifested as the power quadrant tilt, which would influence the safe operation. In order to quantify the impact of fuel-assembly bowing on the core key parameters and the safe operation, it is necessary to accurately establish the core physical model and perform simulation for the fuel-assembly bowing. However, the fuel-assembly bowing cannot be measured in real-time during core operation, which leads to discrepancies between the simulated core and the actual core, manifesting as the errors of power distribution between simulation and the measurements. To improve simulation accuracy, an optimal method of core physical model-inversion with the measured data fusion was proposed in this paper. Several steps were involved in this method. Firstly, the dataset would be generated, in which the inputs were different core physical models and the outputs were the samples of model-simulated value that corresponding to the measured data. Secondly, the artificial neural network (ANN) algorithm would be applied to train the dataset, and the functional relationship between the physical model and the model-simulated value which corresponding the measured data would be established. Thirdly, the three-dimensional variational (3DVAR) assimilation algorithm would be adopted to characterize the differences between the actual core physical model and the simulated model by the cost function, and the measured data would be applied to inverse a more realistic core physical model based on the 3DVAR. Finally, the optimized model would be obtained when the minimum value of cost function was solved. To verify the proposed method, the measured data of power distribution from a commercial PWR nuclear power plant (NPP) in China were applied for the inversion optimization of fuel assembly bowing model, in which the fuel assembly bowing was measured at the end of cycle (EOC) in the spent fuel pool. The optimized model was taken as the bowing amplitude of each fuel-assembly, since the assumption was introduced that the bowing shapes of fuel assemblies were consistent between the pool and the core. The numerical results indicate that the relative error of the power distribution under the inversion optimization model is significantly reduced, with the maximum value decreasing from 13.4% to 7.7%. Furthermore, the relative errors of power distribution under the optimized model are all meet the industrial criteria (±10%). Therefore, the proposed method in this study has the capability to improve the accuracy of numerical simulations notably, and possesses good industrial applications in digital twins of nuclear core.

Keywords

• measured data fusion
• model inversion optimization
• three-dimensional variational algorithm
• artificial neural network algorithm