Knowledge graph-large language model fusion approach for emergency knowledge recommendation in gas tunnels

Area of Science:

• Emergency response
• Knowledge management
• Artificial intelligence

Background:

• Gas tunnel emergencies require rapid, accurate decisions.
• Emergency knowledge is often fragmented across unstructured data sources.
• Existing methods hinder quick access to critical information.

Purpose of the Study:

• To develop a knowledge recommendation approach for gas tunnel emergencies.
• To synergistically combine knowledge graphs (KG) and large language models (LLM).
• To improve the speed and quality of emergency decision-making.

Main Methods:

• Constructing a domain-specific KG from heterogeneous emergency documents using prompt-optimized LLMs.
• Retrieving relevant subgraphs via semantic similarity-based vector search.
• Dynamically augmenting LLMs with retrieved KG knowledge to reduce hallucinations.

Main Results:

• The proposed KG-LLM approach significantly outperforms standalone LLMs and other architectures.
• Demonstrated improvements in accuracy, completeness, and coherence of emergency knowledge.
• The method provides a structured and interpretable approach to knowledge-driven decision support.

Conclusions:

• The integrated KG-LLM method enhances emergency response for gas tunnels.
• This approach offers a novel solution for knowledge accessibility in high-risk scenarios.
• It facilitates more informed and timely decision-making during critical events.