Seepage monitoring is one of the important contents of seepage safety evaluation of earth-rock dams. The seepage pore water pressure is affected by multiple external factors, so the time series of seepage pore water pressure at measuring points is often characterized by nonstationarity and local abrupt changes. Regarding to this, the EEMD-LSTM-ARIMA model for seepage pore water pressure prediction of earth-rock dams is constructed according to the concept of decomposition-reconstruction-combination. Firstly, the time series features are decomposed by the ensemble empirical mode decomposition (EEMD), and the extracted feature components are predicted by the long short-term memory (LSTM) neural network. At the same time, the residual error is corrected by the autoregressive integrated moving average (ARIMA), and the improved prediction model is reconstructed by combining the prediction results of LSTM and ARIMA. Taking an earth-rock dam on a deep overburden as an example, the measured seepage pore water pressure series of two typical measuring points behind the cutoff wall of the main riverbed dam are selected as the research objects for application verification. The results show that, compared with the single LSTM model and ARIMA model, the mean absolute error, the mean square error and the root mean square error of the proposed prediction model are the smallest, and the prediction accuracy of the proposed model is obviously superior to the other two models. Therefore, the proposed model can provide a new approach for accurate prediction and analysis of seepage pore water pressure of earth-rock dams.