Abstract: As a high potential energy accumulation body, the graphite tailings dam exhibits a bimodal particle size distribution. Due to the mineral processing technology, resulting in complex internal stress conditions and failure mechanisms, with high and unpredictable landslide riskst.Therefore, this study proposes an intelligent monitoring and prediction model for graphite tailings dam deformation based on Small Baseline Subset Interferometric Synthetic Aperture Radar（SBAS-InSAR）technology and convolutional long short-term memory network Convolutional Neural Network-Long Short-Term Memory（CNN-LSTM）is proposed.Firstly, SBAS-InSAR technology is used to process 24 SAR images from December 2019 to December 2021 to obtain the cumulative deformation and annual average deformation rate of the study area. By comparing the data of GNSS homonymous monitoring points on site and combining with the analysis of error evaluation index, the accuracy of InSAR monitoring is verified.Then, the correlation characteristics between rainfall and settlement are analyzed, and the periodic variation characteristics of settlement and rainfall are obtained,which reveals the internal mechanism of tailings dam deformation.Finally,the CNN-LSTM model is constructed,and the Long Short-term Memory（LSTM）and bidirectional recurrent neural network model ( BiGRU ) are introduced.The training and prediction results are evaluated using error metrics and loss functions.The results show that:（1）The top of the deformation of the tailings dam exhibits settlement, and the settlement of the slope vertex b is 189.74 mm. Settlement decreases outward from the top until it transitions to uplift at the slope toe, where points a and f experience uplifts of 13.8 mm and 26.8 mm, respectively;（2）The maximum absolute error between SBAS-InSAR technology and GNSS monitoring results is 4.67 mm, and the error distribution is uniform.SBAS-InSAR technology meets the accuracy requirements for graphite dam deformation monitoring.（3）Rainfall is the main influencing factor of tailings dam deformation. With the change of water content in graphite dam,the deformation exhibits periodic fluctuations.（4）A comparative analysis of three prediction models shows that the CNN–LSTM model achieves high curve fitting between the training and testing sets of the loss function, demonstrating effective training and superior performance in predicting graphite tailings deformation. Prediction error metrics across six feature points show a maximum root mean square error below 2.06 mm,a mean absolute error below 1.60 mm, and a maximum coefficient of determination is 0.89.Thus, this study can provide technical support for monitoring and early warning of graphite tailings disasters in northern regions.