Abstract: The accuracy of numerical model prediction for concrete dam displacement is severely affected by the time-frequency nonlinearity of the displacement monitoring data. In view of this, the wavelet technology is employed to analyze the effect of multiple cross environment driving in the prototype data, and then the nonlinear autoregressive model with exogenous input (NARX) and autoregressive integrated moving average model (ARIMA) are integrated to established an autoregressive model system under multiscale combination mechanism to overcome the difficulty of information mining for monitoring sequences with complex chaotic characteristics. Engineering examples manifest that the present model has better fitting accuracy and predictive ability, as well as noise resistance and robustness than traditional models. In addition, being optimized and extended, the model can also be applied to the prediction of other hydraulic structures.

Key words: concrete dam, displacement forecast, multiscale combination mechanism, autoregressive model system, NARX, ARIMA