1.广东省道路运输事务中心,广东 广州 510199
2.广东省智能交通系统重点实验室 / 中山大学智能工程学院,广东 深圳 518107
易智君(1976年生),女;研究方向:交通运输;E-mail:915184458@qq.com
何兆成(1976年生),男;研究方向:智能交通系统;E-mail:hezhch@mail.sysu.edu.cn
收稿:2025-09-30,
修回:2025-10-31,
录用:2025-10-31,
网络首发:2026-01-02,
纸质出版:2026-01-25
移动端阅览
易智君,韦邦彦,徐永高等.KG存储协同CoT推理:构建车辆风险分析智能体[J].中山大学学报(自然科学版)(中英文),2026,65(01):85-92.
YI Zhijun,WEI Bangyan,XU Yonggao,et al.KG storage and CoT reasoning: Construct the intelligent agent for vehicle risk analysis[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2026,65(01):85-92.
易智君,韦邦彦,徐永高等.KG存储协同CoT推理:构建车辆风险分析智能体[J].中山大学学报(自然科学版)(中英文),2026,65(01):85-92. DOI: 10.11714/acta.snus.ZR20250213.
YI Zhijun,WEI Bangyan,XU Yonggao,et al.KG storage and CoT reasoning: Construct the intelligent agent for vehicle risk analysis[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2026,65(01):85-92. DOI: 10.11714/acta.snus.ZR20250213.
文中提出了一种基于知识图谱与思维链(CoT)引导的大语言模型的车辆风险分析智能体框架。首先,面向重点车辆的安全监管,将分散的风险驾驶行为、交通事故与路网数据等信息建模为统一的风险知识图谱语义网络。然后,设计了基于提示词工程的分级推理CoT架构,引导大语言模型实现理解问题、查询数据、确认结果和修正查询的可溯源推理过程,提升问答的准确性与鲁棒性。实验证明,车辆风险分析智能体能够有效支持自然语言交互下的风险分析与关联挖掘,在多种复杂查询任务中表现良好,并能结合子图分析为交通管理提供高效、智能和可解释的决策分析工具。
This paper proposes a vehicle risk analysis intelligent agent framework integrating knowledge graph technology and large language model(LLM) guided by chain-of-thought(CoT). we construct a risk knowledge graph tailored for critical vehicle safety supervision to unify fragmented data including risky driving behaviors, traffic accidents, and road network information into a cohesive semantic network. And a hierarchical CoT reasoning architecture grounded in prompt engineering is designed to steer the LLM through a traceable inference pipeline comprising query comprehension,data retrieval,result validation and query refinement,thereby enhancing the accuracy and robustness of question-answering systems. Experimental validation confirms that the proposed intelligent agent of vehicle risk analysis effectively facilitates natural language-driven risk assessment and association mining across diverse complex query scenarios,delivering an efficient,intelligent,and interpretable decision-support tool for traffic management with demonstrable superiority in operational efficacy and analytical transparency.
柏峰 , 2022 . 道路运输车辆动态监控工作优化改进研究 [J]. 运输经理世界 , 27 : 64 - 66 .
江旭晖 , 沈英汉 , 李子健 , 等 , 2023 . 社交知识图谱研究综述 [J]. 计算机学报 , 46 ( 2 ): 304 - 330 .
李劲业 , 李永强 , 2024 . 融合知识图谱的时空多图卷积交通流量预测 [J]. 浙江大学学报(工学版) , 58 ( 7 ): 1366 - 1376 .
刘奕含 , 宁念文 , 杨东霖 , 等 , 2024 . 面向城市交通的动态知识图谱综述——构建、表示与应用 [J]. 地球信息科学学报 , 26 ( 4 ): 946 - 966 .
王占中 , 兰若冰 , 杨萌 , 等 , 2025 . 利用知识图神经推荐算法实时评估危险品道路运输风险 [J]. 同济大学学报(自然科学版) , 53 ( 8 ): 1253 - 1261 .
BOUHSISSIN S , SAEL N , BENABBOU F , et al , 2024 . SafeSmartDrive: Real-time traffic environment detection and driver behavior monitoring with machine and deep learning [J]. IEEE Access , 12 : 169499 - 169517 .
CHEN M , TAO Z , TANG W , et al , 2024 . Enhancing emergency decision-making with knowledge graphs and large language models [J]. Int J Disaster Risk Reduct , 113 : 104804 .
CHEN T , ZHANG Y , QIAN X , et al , 2022 . A knowledge graph-based method for epidemic contact tracing in public transportation [J]. Transp Res Part C Emerg Technol , 137 : 103587 .
EDGE D , TRINH H , CHENG N , et al , 2025 . From local to global:A graph RAG approach to query-focused summarization [J/OL]. [ 2025-02-19 ]. arXiv. 2404 . 16130 .
GIJRE S , AGRAWAL R , LADDHA P , et al , 2024 . Enhancing question-answering with knowledge graph retrieval and generation using LLMs [C]// International Conference on Artificial Intelligence and Quantum Computation-based Sensor Application . Nagpur, India : 1 - 6 .
HOGAN A , BLOMQVIST E , COCHEZ M , et al , 2021 . Knowledge graphs [J]. ACM Comput Surv , 54 ( 4 ): 1 - 37 .
HOU Y , SHAO Y , HAN Z , et al , 2025 . Construction and application of traffic accident knowledge graph based on LLM [R]. SAE Technical Paper Series . ht tps://doi.org/10.4271/2025-01- 7139.
JIANG L , XIE W , ZHANG D , et al , 2022 . Smart diagnosis: Deep learning boosted driver inattention detection and abnormal driving prediction [J]. IEEE Internet Things J , 9 ( 6 ): 4076 - 4089 .
LI H , APPLEBY G , ALPERIN K , et al , 2025 . The role of visualization in LLM-assisted knowledge graph systems: Effects on user trust, exploration, and workflows [J]. [ 2025-05-20 ]. arXiv. 2505 . 21512 .
LIU C , YANG S , 2022 . Using text mining to establish knowledge graph from accident/incident reports in risk assessment [J]. Expert Syst Appl , 207 : 117991 .
PAN S , LUO L , WANG Y , et al , 2024 . Unifying large language models and knowledge graphs: A roadmap [J]. IEEE Trans Knowl Data Eng , 36 ( 7 ): 3580 - 3599 .
QI Z , NI H , 2024 . Research on power data analysis algorithms based on knowledge graph and artificial intelligence [C]// The 7th International Conference on Information Systems and Computer Aided Education . Dalian, China : 265 - 271 .
YU D , PENG W , CHEN Y , et al , 2024 . Road traffic accident data management and application analysis based on knowledge graph technology [C]// International Conference on Digital Society and Artificial Intelligence . Qingdao, China : 297 - 302 .
ZENG J , TANG J , 2023 . Combining knowledge graph into metro passenger flow prediction:A split-attention relational graph convolutional network [J]. Expert Syst Appl , 213 : 118790 .
ZENG W , HE Z , LU R , et al , 2012 . Freeway segment speed estimation model based on distribution features of floating-car data [C]// Transportation Research Board 91st Annual Meeting . Washington DC,United States : 1612 - 1618 .
ZHAO Z , 2023 . Quantification of carbon emission technologies based on knowledge graph bert-BiLSTM-attention-CRF model [C]// International Conference on Electronics and Devices , Computational Science.Marseille, France : 42 - 47 .
ZHU C , WANG Y , WANG Q , et al , 2023 . Research on traffic accident prediction based on KG-CWT-RGCNN-BiLSTM [J]. Eng Lett , 31 ( 4 ): 1402 - 1414 .
0
浏览量
71
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621
