To achieve high-precision and high-efficiency flood evolution prediction, a rapid flood evolution prediction algorithm that combined the dual advantages of data driving and physical modelling was proposed by integrating the residual neural network (ResNet18) and the hydrodynamic model Telemac2D. High precision flood inundation data generated by Telemac2D were used as training samples to construct a deep neural network model based on the one dimensional ResNet 18. This model was subsequently employed to conduct real time dynamic predictions and comparative verification of flood inundation depth and evolution pathways in the Shangjia River section of the lower Huangbai River Basin. The results demonstrated that for the 240 test sets, the developed ResNet 18 model achieved a mean absolute error of 0.033 2 m and a root mean square error of 0.089 8 m, respectively. The spatial distribution of the predicted inundation extent exhibited a high degree of consistency with simulation results from Telemac2D, with a correlation coefficient of 0.998 1. Furthermore, the model provided more accurate water depth predictions at measurement points compared to the convolutional neural network model and enhanced computational efficiency by over 300 times relative to the traditional hydrodynamic model.