Abstract: Concrete face slab is the main seepage prevention structure of concrete face rockfill dams, and the deformations of joints are very important to the dam's seepage prevention, safety and stability. The joint system of concrete slabs is affected by multiple factors, such as water levels, deformation of rockfill dams, and temperature. Therefore, the deformation and evolution of the joint system is a dynamic and nonlinear complex process. We use multiple time series of the slab joint deformation and its influencing factors and Long Short-Term Memory (LSTM) to establish a prediction model for the slab joint deformation. First, we analyze the correlation between the slab joint deformation and its influencing factors, in which the time series of the influencing factors that have a greater impact on the slab joint deformation will be selected. The selected time series data is constructed through sliding time windows and is input into the LSTM network. Taking the joint deformation as the network output, the prediction model of slab joint deformation based on multivariate time series and LSTM network is trained. The predicted results are compared with results of Vector Autoregression (VAR), univariate LSTM model and Holt-Winters model. The comparison results show that the multivariable LSTM model has higher prediction accuracy, and has certain engineering practical value for real-time dynamic prediction of slab joint deformation.