Abstract: Deformation prediction is of great significance to the safe operation and risk control of concrete dams. In view of the difficulties in long-term accurate prediction and modeling, this study uses a MR model to decompose a deformation sequence into the components of hydrostatic pressure, temperature, aging and remainder, and apply the SARIMA model to mine the information buried in the unstable and irregular sequences of the remainder component. Thus, we have developed a new MR-SARIMA prediction model of long-term deformation of concrete dams. Analysis and application to a dam case show that this new simple model is easy to use, more accurate, more stable, and obviously superior in the prediction of long-term concrete dam deformation featured with periodicity and trends to the previous models reported.

Key words: concrete dam, deformation prediction, sequence decomposition, SARIMA, deep learning