Abstract: Excavation deformation can have numerous adverse effects on construction projects, potentially trigger catastrophic incidents such as soil collapse and cracking of nearby roads or buildings. Therefore, predicting excavation deformation is a crucial aspect of excavation engineering. To enhance the accuracy of these predictions, we propose a variational mode decomposition convolutional neural network-long short-term memory (VMD-CNN-LSTM) prediction model, which takes time series of monitoring data as input. Using onsite monitoring data from Nanjing Jiangbei new district library, the VMD-CNN-LSTM model was applied to forecast the deep horizontal displacement of the continuous underground wall at monitoring point CX07. A comparative analysis of the deformation predictions obtained from the LSTM and CNN-LSTM models demonstrates that the VMD-CNN-LSTM model offers superior accuracy. Further validation of the model's predictive performance was conducted using monitoring data from two other points, confirming the applicability and stability of the VMD-CNN-LSTM model.