Integrating knowledge graphs and LLMs for intelligent full-chain safety risk pre-control in large hydropower projects

Construction safety risks (CSRs) in large hydropower projects (LHPs) are difficult to manage due to complex interactions, dynamic evolution, and cascading propagation. This paper investigates how to achieve intelligent CSR knowledge modeling and proactive risk pre-control for LHP construction. An intelligent framework is developed by integrating ALBERT-BiLSTM-Att-CRF-based entity recognition, association-rule-based relation extraction, knowledge graph (KG) construction, and large language model (LLM)-enhanced decision support. The results show that the proposed framework effectively identifies CSR entities (F1 > 90%), multi-hop risk propagation paths, and supports accurate knowledge retrieval and intelligent pre-control recommendations (overall closed-ended question-answering accuracy of 0.8667 and an expert relevance score of 4.21/5.00 for open-ended recommendations). This study provides an interpretable and scalable solution for safety managers and engineers to enhance risk identification and decision-making in complex engineering projects. The findings inspire future research on integrating multimodal data and advanced artificial intelligence techniques for more adaptive construction safety management.