Seite - 760 - in Book of Full Papers - Symposium Hydro Engineering

detection of anomalies typically depends on hypothesis testing or threshold- based procedures, which are prone to false alarms. Bayesian Dynamic Linear Models (BDLMs) are a class of state-space models which are well suited for sequential inference [2]. BDLMs are capable to decompose time series into a set of sub-components, as HST does. In contrast with the HST technique, the behavior of the sub-components can vary with time, without the need to retrain the model. Recent applications have demonstrated the potential of BDLMs to track time-varying baseline responses of civil infrastructures from real dataset [3], and to detect anomalies [4]. This paper compares the HST and BDLMs techniques for the long-term monitoring of dams. This study presents a specific class of BDLMs based on the switching Kalman filter (BDLM-SKF). Two examples of application based on real displacement data recorded on a dam in Canada and simulated data show that BDLM-SKF is robust towards false alarms, and enable to interpret non-stationary time series. One feature of BDLM-SKF is its capacity to interpret the baseline after an anomaly occurs (i.e. when the dam returns to its normal behavior), without retraining the model. In contrast, HST model is prone to false alarms, and is unable to provide information about when an anomaly stops. This study demonstrates that Bayesian Dynamic Linear models (BDLMs) outperform Hydrostatic-Time-Season (HST) modelling for monitoring the long- term behavior of dams, while providing the same level of interpretability. The main advantages of BDLM over HST model is its ability to interpret non-stationary time series and to detect change in the baseline response of the dam (i.e. anomalies) without being prone to false alarms. REFERENCE [1] WILM G., BEAUJOINT N., 1967. Les méthodes de surveillance des barrages au service de la production hydraulique d'Electricité de France. Problèmes anciens et solutions nouvelles, IXe Congrès CIGB, Istambul, Q34, R30 [2] WEST M. AND HARRISON P. J., 1997, Bayesian Forecasting Dynamic Models. Springer Verlag, 2nd edition [3] GOULET J.-A., 2017, Bayesian dynamic linear models for structural health monitoring, Structural Control and Health Monitoring, https://doi.org/10.1002/stc.2035 [4] NGUYEN L.H. AND GOULET J.-A., 2018, Anomaly Detection with the Switching Kalman Filter for Structural Health Monitoring, https://doi.org/10.1002/stc.2136 760