Bayesian network-based approach for dam safety diagnosis

Certain frontline dam safety management personnel lack the ability to diagnose dam hazard issues and often find it difficult to identify potential risks in practice. Existing causal analysis methods for dam safety diagnosis struggle to provide specific reasoning paths and quantify risk probabilities. To address this gap, this study proposes a dam safety diagnostic method based on Bayesian networks (BNs). First, historical cases of dam safety hazards were collected and classified to extract various types of hazard issues, abnormal manifestations, and underlying causes, which were used as nodes within the BN. Correlation analysis was then performed to identify relationships among the nodes, enabling the construction of directed edges that form the BN structure. The degree centrality algorithm was employed to analyze the prior probabilities of parent nodes, while Bayes’ theorem was applied to calculate the conditional probabilities of the child nodes, generating conditional probability tables for all nodes within the network. Using the BN’s posterior probability inference method, the probabilities of hidden hazards in a target dam were calculated, facilitating accurate diagnosis and root cause tracing of potential risks. Finally, a case study involving a hidden hazard in a domestic earth-rock dam was used to validate the proposed method. The results demonstrate that the method efficiently utilizes a large number of scattered dam hazard cases, is less affected by subjective factors, provides clear reasoning links and risk probabilities, and can accurately identify dam hazard issues and trace their root causes, offering technical support for dam operation and safety management personnel.