The early warning method of Dagangshan high-arch dam risk based on the time series prediction of the multivariate monitoring data

Abstract

In areas with high seismic activity, stability analysis of high-arch dams becomes essential for the safe functioning of the structure. Measured displacement is a common approach used in risk assessment techniques to evaluate the state of the dam; however, there are very limited studies on forecasting and early warning methods involving future unknown data. In light of this, this study proposes a novel early warning framework for high-arch dams based on displacement and stress prediction. First, the radial displacements, along with bolt stress measured from the field were processed to remove any abnormal data before being normalized. Second, the future displacement and stress at different measurement points were predicted using three time-series prediction models (bidirectional long short-term memory network, echo state network, and Transformer) to obtain the data closest to the true value. The overload (by numerical simulation) and extreme conditions (Luding earthquake, Ms = 6.8) methods were then combined to determine the warning threshold at each measurement point. Finally, based on the principle of Bayesian probability, the Dam Risk Index was calculated based on the displacement and stress at all measurement points. This framework considers the interdependence between multiple monitoring factors, avoids the subjective complexity of the weight determination of each factor, prevents the influence of experience and subjective judgment, and makes the decision-making process more objective. This study provides a more reasonable solution for monitoring and controlling dam engineering safety.