This paper develops a deformation monitoring model that combines gated recurrent unit (GRU) network, self-attention (SA) mechanism, and temporal convolutional network (TCN). The GRU extracts temporal features from environmental factors, while the TCN captures long-term dependencies and generates feature representations through dilated causal convolution. Coupling a self-attention mechanism with the GRU-TCN framework enhances the focus on critical feature information. In addition, a factor selection method based on maximal information coefficient and kernel principal component analysis is proposed. The performance of the proposed GRU-SA-TCN model is tested on a super high arch dam. Comparative studies indicate the superiority of the proposed model in predicting dam deformation, demonstrating high prediction accuracy and reliability. This study provides a robust framework for monitoring the deformation of high arch dams.