Towards an Error Correction Model for dam monitoring data analysis based on Cointegration Theory

The Multiple Linear Regression (MLR) model, which can only be used for stationary time series, is one of the most widely used models for the analysis of dam monitoring data. However, dam monitoring data are mostly nonstationary time series. Spurious regression may be observed while using an MLR model without testing the stationarity of the series, thus decreasing forecast precision. The authors of this paper carried out the stationary test for monitoring data before estimating the forecasting model. The Augmented Dickey–Fuller test was adopted in order to verify the stationarity of monitoring data based on Cointegration Theory, followed by the Engle–Granger cointegration test to determine the cointegration relationship among the monitoring variables. An Error Correction Model is proposed in order to represent the long-term equilibrium and short-term disequilibrium relationships of variables so as to improve the fitting accuracy and forecast precision if the cointegration relationships exist among the variables under analysis. An analysis of the deformation monitoring data of an arch dam was undertaken as a case study. Nonstationarity was found to exist in the arch dam monitoring data, and cointegration relationships were found between the dam deformation data and influence factors such as hydrostatic pressure, concrete temperature changes and time-effects. The Error Correction Model displays better fitting accuracy and forecast precision than the MLR model.

▸ For nonstationary time series, spurious regression may result from MLR analysis. ▸ The necessity of stationarity testing for monitoring data series is proposed. ▸ An Error Correction Model to improve forecasting accuracy is proposed. ▸ Cointegration test and error correction model can avoid spurious regression. ▸ Deformation of arch dam as a means of verification for the model's effectiveness.