Abstract: To examine the interaction mechanism between flood discharge and a sluice chamber, a novel method is developed coupling fluid-structure interaction (FSI) Finite Element analysis with a Back Propagation Neural Network, based on stress-strain characteristics, and applied to the Datengxia water control hub project. This method facilitates the development of a digital twin based on numerical simulation data. We construct a finite element model of COMSOL for flood discharge and sluice chamber structure, and simulate five flood discharge scenarios of 23400 m3/s, 30600 m3/s, 39000 m3/s, 42300 m3/s, and 66200 m3/s. Then, we examine the FSI process of the sluice chamber and its corresponding load patterns. A total of 1250 monitoring points are arranged throughout the sluice chamber and the flow domain. The time-sequence data for four hydraulic parameters are extracted at a 15-second interval－flow velocity ( ), pressure ( ), turbulence intensity ( ), and vorticity ( ). And, stress and displacement data are simultaneously collected from the sluice chamber, so that training datasets for the BP Neural Network (BPNN) can be constructed. Finally, we develop a BPNN model for predictions of the sluice chamber’s stress and displacement, using spatial coordinates and hydraulic parameters as inputs, and train and validate it. Results show a high predictive accuracy of this FSI collaborative BPNN method－the coefficient of determination (R2) reaches up to 0.975 for stress and 0.987 for displacement. Specifically, 96.0% of the stress predictions have an error below 10% with the maximum absolute error of 0.097 MPa; 99.1% of the predicted displacements have an error below 10% with the maximum absolute error of 0.395 mm, or significantly below the allowable deformation threshold of 0.45 mm for chamber joints. This study verifies the feasibility of our new method, the reliability of BPNN in predicting stress and displacement in the sluice chamber, and the advantage of methodology.

Key words: sluice chambers, fluid-structure interaction finite element analysis, BP neural network, simulation-based prediction, flow field reconstruction