High and steep slopes are common during the construction of large-scale projects, and their deformation often leads to geological hazards, posing significant threats to life and property. Efficiently collecting displacement data and developing an accurate predictive model are therefore essential. This study proposes a Transformer-CNN hybrid model that integrates convolutional layers and residual structures into the Transformer architecture. The optimized model is applied to displacement data obtained from the Beidou satellite system in a large water conservancy project in Chongqing. Experimental results indicate that the Transformer-CNN model achieves lower MAE, MSE, and RMSE values compared to single-model approaches, demonstrating superior prediction accuracy. These findings suggest that the proposed model offers a practical solution for predicting and analyzing slope deformation in similar engineering projects.