Abstract: In conventional dam deformation monitoring models, information mining of dam prototype observation data is limited and forecast precision is not up to standard. Dam deformation prototype data can be regarded as non-stationary time series, and considering the influence factors of dam deformation, it can be decomposed into cyclical factors and random factors. Dam deformation monitoring data are decomposed and reconstructed by multi-scale wavelet analysis method in this paper, BP neural network and Autoregressive Integrated Moving Average Model (ARIMA) are separately used to analyze and forecast the random signal and system signal contained in deformation monitoring data, and the forecast values based on the two models are superimposed, and the multi-scale deformation combination forecast value for concrete dam based on BP-ARIMA is proposed according to the time series principle. Example shows that, compared with the conventional models, active components contained in the monitoring data are effectively excavated, and the forecast precision is improved obviously, meanwhile, the calculation and analysis process is simple in the proposed combination model. A new method of the deformation forecast for high slope and other hydraulic structures is presented.