Deformation monitoring data is an important basis for the quantitative evaluation of structural safety of water conservancy projects. The deformation data of water conservancy projects is a type of typical non-stationary signal, which contains both linear and nonlinear components. According to the characteristics of these components, a combined analysis model based on auto regressive moving average and group method of data dandling (ARMA-GMDH) is constructed to predict the deformation of hydraulic engineering by using the auto regressive moving average model for linear signal and the data grouping method for nonlinear signal. The engineering example shows that the proposed method can effectively predict the linear and nonlinear components of the deformation of water conservancy projects. Compared with the results of several prediction methods, it is found that the proposed combined model has higher prediction accuracy and presents similar deformation trend with the measured data. So, it is applicable to the analysis of both the linear and nonlinear components of the deformation and in turn evaluate the deformation trend and safety status of the hydraulic structure. Therefore, the proposed model has a certain engineering application value.