Statistical models for the dynamic monitoring of arch dam deformation via a Bayesian dynamic linear model

Abstract

Accurate real-time analysis of deformation dynamics in arch dam structures and the development of a robust online deformation monitoring framework are imperative for ensuring the safe operation of such structures. However, existing paradigms for monitoring arch dam deformation rely predominantly on static methodologies, which inadequately account for the continuous variations in model parameters induced by the progressive ageing of the dam. To enhance the monitoring of arch dam deformation behaviour, this study transcends traditional statistical models by employing the Bayesian dynamic linear model (BDLM) framework. The BDLM integrates trend, seasonal and regression components individually or in combination to scrutinize monitoring data and extract temporal characteristics essential for forecasting the deformation behaviour of arch dams. The proposed dynamic safety monitoring model facilitates real-time updates based on continuous monitoring data, thereby enabling a quantitative analysis of the temporal evolution of dam behaviour. This approach introduces a novel paradigm for assessing the safety status of dams subjected to deformation. The findings from the case study indicate that the dynamic safety monitoring model, grounded in the BDLM, proficiently captures the inherent temporal variability of arch dams. Furthermore, it adeptly updates and quantifies the uncertainty associated with predictive outcomes. Compared with static statistical models, the dynamic safety monitoring model within the BDLM framework demonstrates superior modelling capabilities and predictive performance.