Knowledge graph-enhanced large language models for decision support in coastal waterway aids to navigation

Effective management of aids to navigation (AtoN) plays a critical role in ensuring maritime safety and maintaining navigational order in coastal waterways. However, decision-making in AtoN management is challenged by fragmented data sources, heterogeneous information, and the need to integrate domain knowledge under complex operational conditions. Traditional rule-based systems and data-driven methods have limitations in integrating heterogeneous knowledge and supporting context-aware operational decision-making. To address this issue, this study proposes a knowledge graph-enhanced large language model (LLM) framework for decision support in AtoN management. The proposed framework integrates heterogeneous data sources and domain knowledge into a unified semantic representation via a knowledge graph. A domain-specific entity extraction model is employed to extract structured information from unstructured incident reports for knowledge graph construction. LLMs are further employed to perform knowledge-guided generation and produce structured decision support outputs for AtoN incident handling. The proposed framework is evaluated using a case study in the Jiaomen waterway, based on AtoN registry data and operational incident reports. The results indicate that the knowledge graph-enhanced framework improves the consistency of generated decision workflows and strengthens their alignment with reference workflows. The study demonstrates the feasibility of integrating structured knowledge representation with LLM-based generation to support traceable and knowledge-guided decision workflows in maritime and coastal safety management.