Objectives The traditional method uses a single monitoring model to conduct the early warning of dam displacement behavior. However, the single monitoring model may not reflect the real behavior of the dam due to its low accuracy, thus causing false alarms.

Methods In order to improve the reliability of early warn‍ing results, a new multi-model early warning method is proposed. The hydraulic-season-time (HST) model and autoregressive moving average (ARMA) model, which are the most common models in engineering, are taken as the basic model. First, the advantages and disadvantages of the two models in dealing with dam displacement monitoring data are analyzed. Then the one-dimensional residual distribution of the two models is discussed by kernel density estimation. On this basis, the frequency analysis of the joint residuals of HST-ARMA is carried out and it shows the joint residuals is weakly correlated in non-tail, but strongly correlated in tail. Finally, the Copula function is used to fit the joint residuals of HST-ARMA, and the joint distribution function is obtained, which realizes the joint early warning of dam displacement behavior with multiple models.

Results It is verified by several measuring points on the wire alignment system of concrete dam. The case study shows that if a single HST model or ARMA model is used for early warning, the false alarm rate can reach 23.17%-27.94% due to the influence of modeling sequence features and model structure features. And if the HST-ARMA method is used, the false alarm rate can be reduced to 0.00%‍‍‍‍-0.63% due to combine the advantages of different models.

Conclusions The new method can combine the advantag‍es of different models and avoid the disadvantages of the single model, thus effectively reducing the frequency of false alarms. The warning results of HST-ARMA method are more reliable and can more truly reflect the dam displacement behavior. It can provide reference for improving the dam safety management level.