Seite - 759 - in Book of Full Papers - Symposium Hydro Engineering

COMMISSION INTERNATIONALE DES GRANDS BARRAGES ------- VINGT-SIXIÈME CONGRÈS DES GRANDS BARRAGES Autriche, juillet 2018 ------- A COMPARISON BETWEEN HYDROSTATIC-TIME-SEASON AND BAYESIAN DYNAMIC LINEAR MODELS FOR MONITORING BEHAVIOR OF DAMS* Ianis GAUDOT1, Luong Ha NGUYEN2, James GOULET3 1Postdoc fellow, 2Ph.D. student, 3Assistant Professor POLYTECHNIQUE MONTREAL Benjamin MIQUEL Ing., Ph.D., HYDRO-QUEBEC CANADA ABSTRACT Monitoring the long-term behavior of dams is key for addressing safety and serviceability issues. One challenge is to enable the detection of changes in the baseline response of the structure (i.e. anomalies) in real time, while mitigating false alarms. The most popular data interpretation method for monitoring the long-term behavior of dams is the Hydrostatic-Season-Time model (HST; [1]). The HST performs a regression analysis to decompose a time series into three sub- components related to the hydrostatic effects (H), the thermal effects (S), and the baseline response of the dam (T). The main drawback of HST models, is that they are not dynamic; it means that once the model is built using a training dataset, it stops evolving as new data are collected over time. This is a limitation because the main objective of dam instrumentation/auscultation is to interpret non-stationary (i.e. abnormal) behavior in the baseline response. Moreover, the * Comparaison entre les modèles Hydrostatique-Saison-Temps et les modèles Bayésiens Linéaires Dynamiques pour le suivi temporel du comportement des barrages 759