Water Resources and Power

To address the issues of insufficient experience among frontline dam safety managers and the ineffective utilization of industry knowledge, a dam safety decision support method based on knowledge graphs and twin BERT network was proposed. Firstly, by analyzing the content potentially involved in the dam safety decision-making process, a dam safety knowledge system comprising nine entity types and fifteen relationship types was constructed. Next, the collected dam safety-related data were stored and associated based on the knowledge system to form a dam safety knowledge graph. Then, a semantic matching model for dam safety knowledge was constructed based on the twin BERT network, establishing effective connections between natural language retrieval queries and target cases in the graph database. Finally, an intelligent matching method for dam hazard cases based on the knowledge graph were proposed, which enabled the intelligent retrieval and recommendation of the most similar cases. The application effectiveness was validated through the actual case. The results indicate that the proposed method effectively leverages the expertise and typical cases of the dam safety industry, and provides decision support for dam safety managers in the form of industry knowledge base.