.png){kind=icon}
.png){kind=icon}
Return
iS3Tunnel: Highway-tunnel domain RAG-enhanced LLM and its hallucination-quantified agent design
.png){kind=icon}
OA
AI
X
H
L
A
H
X
C
M
DOI：10.1016/j.undsp.2025.12.007.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper addresses the need for intelligent decision support in highway tunnel engineering, where complex challenges exceed the capabilities of general large language model (LLM) due to limited domain knowledge. The core question is how to develop a tunnel-specific LLM and integrate it into an AI agent capable of autonomous decision-making while minimizing hallucinations. An LLM called iS3Tunnel is developed via retrieval-augmented generation, integrated into an autonomous agent, and augmented with cloud model theory to quantify and mitigate hallucinations. In a case study, the agent performed risk evaluations with minimal human input, and was closely matched expert cognition. The findings demonstrate that iS3Tunnel achieves a comprehensive accuracy rate of 93% in highway tunnel engineering tests, significantly outperforming GPT-4o (79%). The iS3Tunnel agent improves decision-making efficiency while effectively reducing hallucinations. Future research will focus on expanding the knowledge base and developing a hallucination-quantified AI agent in construction domain.
Keywords:
AI agent
Highway tunnel
Large language model
Hallucination quantization
Retrieval-augmented generation
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
8.3
Papers:
668
Citations:
2.9K
Organization
No organization information available
Cited Papers