Under the dual influence of internal factors and the external environment, dam deformation shows a high degree of nonlinear characteristics. In the traditional dam deformation prediction model, an abnormal value of the residual sequence reflects an abnormal situation or potential fault in the dam deformation prediction. Therefore, it is particularly important to fully mine the effective information in the residual sequence. Considering the nonlinear, time-varying and chaotic characteristics of dam time series, a hybrid monitoring model of concrete arch dams based on residual sequence correction is proposed in this paper. Firstly, combined with the finite element method, the statistical model was used to establish the initial mixed prediction model based on the monitoring data. In view of the specific characteristics and periodicity of the prediction residuals of the hybrid model, this paper uses symplectic geometric mode decomposition (SGMD) to decompose the residual sequence. Then, the convolutional neural network (CNN) is optimized by the improved sparrow search algorithm (ISSA), and the decomposed modal components are predicted and reconstructed by the bidirectional gated recurrent unit (BiGRU) of the attention mechanism. Subsequently, the differential autoregressive moving average model (ARIMA) is used to correct the reconstructed residual sequence. Finally, the modified residual sequence is combined with the initial hybrid model to construct a hybrid monitoring model of a concrete arch dam based on residual sequence correction. Through the refinement of the residuals, the accuracy and reliability of the monitoring data are further improved. This provides a new and effective method for safety monitoring and early warning of concrete arch dams.