A Landslide Monitoring and Early Warning System with Retrieval-Augmented Generation Enhanced by Knowledge Graph
Wu Runze
Li Hao
Mei Hongbo
Wang Chengbin
Zhu Minyi
Wang Hongqun
Zhang Liang
Hu Guanghong
Ma Mingjie
Wang Zhiwen
Zhang Longlong
Huang Min
Li Zhenhua
Abstract：Aiming at the current problem of knowledge extraction and utilization from text reports in landslide prevention and control work,this study explores a set of methods for constructing and applying a knowledge graph based on geological disaster data and materials,promoting the transformation of geological disaster prevention from data management to knowledge management.Knowledge extraction and knowledge graph construction were conducted on basic data from 182 reports,22 specifications,5 literatures,4 668 landslide points,and 313 governance projects,as well as three years of historical disaster data,current landslide monitoring data,and forecasted weather data.Retrieval-augmented generation(RAG)technology was adopted to integrate and output multi-source knowledge.A knowledge graph system with 12 797 entities,34 873 relationships,and 9 658 chunks was established,enabling 10-day early warning for 4 518 landslides.Knowledge graph technology effectively extracts knowledge from text reports,and RAG technology efficiently integrates multiple knowledge sources,improving the accuracy of knowledge question-answering.
Keywords：Three Gorges reservoir arealandslide monitoring and early warningknowledge graphlarge language modelretrieval-augmented generationengineering geology
Publication Date：2025-10-30
Online Publishing Date：2025-11-11(First online date of this platform, not the publication date of the document)
Pages：12( 4125-4136 )