Coupled Continuous-Time Markov Chain–Bayesian Network Model for Dam Failure Risk Prediction

Abstract

Hydropower dams are critical infrastructure systems characterized by their complex, dynamic, and stochastic behaviors. The frequent variation in the hydrological and meteorological variables poses a higher probability of dam failure, highlighting the need to improve pertinent risk assessment approaches to predict failure risks, considering the uncertain states of such variables. Bayesian networks (BN) analysis has been a key risk assessment tool for decades; however, BN’s static acyclic nature is a recognized drawback. In this paper, a continuous-time Markov chain (CTMC) is coupled with a BN model to enable the dynamic assessment of dam failure risk. In this respect, BN is used to represent the interrelation among the system variables and simulate the propagation of uncertainties throughout the system, whereas the CTMC is adopted to describe the continuous transition of the system variables over their respective states. To demonstrate its applicability, the developed coupled BN-CTMC model was employed to predict the probability of failure of the Daisy Lake Dam in the province of British Columbia, Canada, under the uncertainty of reservoir water level, inflow, and wind speed states. The developed BN-CTMC modeling approach can aid in the development of reliable dam operation schemes and risk mitigation strategies through (1) adequately representing the propagation of the hydrological (e.g., inflow and reservoir water level) and meteorological (e.g., wind speed) variable uncertainties through dam system dynamical processes; (2) effectively quantifying dam failure risk under different operational conditions and failure scenarios; (3) accurately specifying the critical periods of dam system operational safety; and (4) providing in-depth understanding of the relationships between the dam system’s failure and associated variables over time.