3DVar data assimilation for inversion optimization of main steam system parameters in nuclear power plants using forecast-gradient and residual FCN

The main steam system is crucial for the safety of nuclear power plants, and predicting its operating parameters is essential for monitoring system performance and improving energy efficiency. Since the damping coefficients in steam pipes cannot be directly measured, numerical simulations are typically used. However, model simplifications and uncertainties in modeling parameters lead to discrepancies between the simulations and actual. To address this issue, we propose ResFCN-3DVar, a novel data assimilation method that optimizes modeling parameters using observable quantities. The primary innovation lies in leveraging a residual-based fully connected network (ResFCN) to construct the observation operator, effectively handling the nonlinear and complex system-level model. Additionally, we incorporate a second-order finite difference matrix within the 3DVar framework to improve computational efficiency. Experiments using a dataset simulated by the full-scope simulator showed that the root-mean-square error (RMSE) of the assimilation results remained below 1%, demonstrating the effectiveness of ResFCN-3DVar in optimizing modeling parameters.