| 325 | 0 | 759 |
| 下载次数 | 被引频次 | 阅读次数 |
【背景】轨道交通系统中的突发事件处置不当，将会严重扰乱列车运行秩序，甚至导致乘客伤亡与巨额经济损失。【目标】构建一种融合本地知识库的应急管理大语言模型（EM-LLM），为城市轨道交通行车调度员提供复杂环境下快速且精准的决策支持，保障城市轨道交通系统的安稳运行。【方法】通过文本切分及向量化等技术构建本地知识库，定义统一的“prompt”作为模型响应的全局引导指令，结合LangChain框架构建并部署本地EM-LLM。设计人机现场对抗实验验证EM-LLM的有效性，安排有经验的调度员与通用LLM、EM-LLM在34种典型故障场景下进行同步决策，从真实调度指令的生成质量、响应时效、方案完整性等维度进行多维对比。【数据】收集约1 000万字符的国家、行业及企业行车调度相关标准和突发事件处置历史数据构建为本地知识库。【结果】通用LLM在故障类型判断方面较人工调度员具有一定优势，但在生成长文本形式的事故处置方案时表现不足；相比之下EM-LLM能更有效地应对复杂突发事故场景。【结论】EM-LLM能够为轨道交通系统在动态场景下的智能调度提供可靠支撑，有助于提升系统在突发事件中的应急处置能力与运营组织效率。
Abstract：[Background] Improper emergency management in rail transit systems can significantly disrupt train operations, thus potentially causing passenger injuries and substantial economic losses.[Objective] This study introduces a large language model(LLM) for emergency management(EMLLM) that integrates a domain-specific local knowledge base. Its primary goal is to provide urban rail transit dispatchers with rapid and accurate decision support in complex environments, thereby safeguarding the operational safety and reliability of urban rail transit systems. [Method] Textual data were obtained, segmented, and vectorized to construct a local knowledge base. A unified “prompt”was defined to serve as the global guiding instruction for model responses. The EM-LLM was developed and deployed locally using the LangChain framework. To validate its effectiveness, a comparative human-machine evaluation experiment was designed. In this experiment, three approaches—experienced dispatchers, a general-purpose LLM, and the proposed EM-LLM—were tested in 34 typical emergency scenarios. The performance was evaluated across multiple dimensions, including the quality of the generated dispatch instructions, the response time, and the completeness of the generated decision. [Data] A local knowledge base was constructed using approximately 10 million characters from national, industrial, and enterprise standards related to train dispatching, along with historical emergency-management data. [Result] Although the general-purpose LLM exhibited certain advantages over human dispatchers in terms of fault-type classification, it underperformed in generating comprehensive, long-form management decisions. By contrast, the EM-LLM demonstrated greater effectiveness in responding to complex and unexpected emergency scenarios. [Conclusion] The EM-LLM offers valuable support for intelligent rail traffic management in complex emergency scenarios, thus significantly enhancing the emergency response capabilities and operational efficiency of the system during unexpected disruptions.
[1]韩宝明,李卓一,靳雨馨,等. 2024年世界城市轨道交通运营统计与分析综述[J].都市快轨交通, 2025, 38(1):1-12.HAN Baoming, LI Zhuoyi, JIN Yuxin, et al. Statistical analysis of urban rail transit operations worldwide in2024:a review[J]. Urban Rapid Rail Transit, 2025, 38(1):1-12.
[2]陈星,阴佳腾,高原,等.面向客流聚集风险防控的城轨列车实时调度模型与算法[J].交通运输工程与信息学报, 2024, 22(2):90-103.CHEN Xing, YIN Jiateng, GAO Yuan, et al. Real-time train rescheduling of urban rail transit to optimize safety risks of over-congestions[J]. Journal of Transportation Engineering and Information, 2024, 22(2):90-103.
[3]王宇,陈佳琦.突发事件下列车开行方案调整对铁路客运网络鲁棒性的影响[J].大连交通大学学报, 2024, 45(3):8-15.WANG Yu, CHEN Jiaqi. Impact of train operation plan adjustments under emergency events on the robustness of railway passenger transport networks[J]. Journal of Dalian Jiaotong University, 2024, 45(3):8-15.
[4]王洪德,姜天宇.地铁突发事件应急能力评价指标体系建模[J].大连交通大学学报, 2015, 36(5):92-95.WANG Hongde, JIANG Tianyu. Modeling of an evaluation index system for emergency response capability of metro systems under unexpected incidents[J]. Journal of Dalian Jiaotong University, 2015, 36(5):92-95.
[5]查伟雄,周佳宇,李剑,等.基于到发线优选罚值的高速铁路列车运行调整双层优化模型[J].北京交通大学学报, 2023, 47(6):13-20.ZHA Weixiong, ZHOU Jiayu, LI Jian, et al. Two-layer optimization model of train rescheduling for high-speed railway based on preferred penalty value of arrival-departure track allocation[J]. Journal of Beijing Jiaotong University,2023, 47(6):13-20.
[6]宁正,牛宏侠,张肇鑫.基于排队论的地铁列车运行自动调整优化模型[J].铁道科学与工程学报, 2019, 16(7):1826-1832.NING Zheng, NIU Hongxia, ZHANG Zhaoxin. Automatic operation regulation optimization model of metro train based on queuing theory[J]. Journal of Railway Science and Engineering, 2019, 16(7):1826-1832.
[7]VEELENTURF L P, KIDD M P, CACCHIANI V, et al. A railway timetable rescheduling approach for handling large-scale disruptions[J]. Transportation Science, 2016,50(3):841-862.
[8]龙科军,刘嘉,高志波,等.考虑配送紧迫度与公平的应急车辆调度模型[J].中外公路，2024,44(6):51-258.LONG Kejun, LIU Jia, GAO Zhibo, et al. An emergency vehicle scheduling model considering delivery urgency and equity[J]. Journal of China&Foreign Highway,2024, 44(6):51-258.
[9]王宇,王天琪.基于改进SIR模型的地铁站突发事件恐慌传播与干预时间研究[J].大连交通大学学报, 2025, 46(2):27-35.WANG Yu, WANG Tianqi. Study on panic propagation and intervention timing of metro station emergencies based on an improved SIR model[J]. Journal of Dalian Jiaotong University, 2025, 46(2):27-35.
[10]SU S, QU J, CAO Y, et al. Adversarial training lattice LSTM for named entity recognition of rail fault texts[J].IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11):21201-21215.
[11]LIU J, CHEN K, DUAN H, et al. A knowledge graphbased hazard prediction approach for preventing railway operational accidents[J]. Reliability Engineering&System Safety, 2024, 247:110126.
[12]ZENG Y, QIN Y, LIU D, et al. Railway train device fault causality model based on knowledge graph[C]//2020 International Conference on Sensing, Diagnostics, Prognostics, and Control(SDPC). Beijing:IEEE, 2020:385-390.
[13]刘晓磊,邹国建,段征宇,等.面向个体出行的地铁路径提取与行为模式挖掘[J/OL].交通运输工程与信息学报, 2025:1-14.(2025-09-04)[2025-09-10]. https://link.cnki.net/doi/10.19961/j.cnki.1672-4747. 2025. 07.026.LIU Xiaolei, ZOU Guojian, DUAN Zhengyu, et al. Metro route extraction and travel behavior patterns mining at the individual level[J/OL]. Journal of Transportation Engineering and Information, 2025:1-14.(2025-09-04)[2025-09-10]. https://link.cnki.net/doi/10.19961/j.cnki.1672-4747.2025.07.026.
[14]刘学博,户保田,陈科海,等.大模型关键技术与未来发展方向——从ChatGPT谈起[J].中国科学基金,2023, 37(5):758-766.LIU Xuebo, HU Baotian, CHEN Kehai, et al. Key technologies and future development directions of large language models:insights from ChatGPT[J]. Bulletin of National Natural Science Foundation of China, 2023, 37(5):758-766.
[15]DEVLIN J, CHANG M W, LEE K, et al. BERT:pretraining of deep bidirectional transformers for language understanding[C]//North American Chapter of the Association for Computational Linguistics. San Diego:NAACL,2019:3498-4195
[16]LAN Z, CHEN M, GOODMAN S, et al. ALBERT:a lite BERT for self-supervised learning of language representations[EB/OL].(2019-09-01)[2025-08-22]. https://arxiv.org/abs/1909.11942.
[17]RADFORD A, NARASIMHAN K, SALIMANS T, et al.Improving language understanding by generative pretraining[EB/OL].(2018-09-26)[2025-08-23]. https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf.
[18]TOUVRON H, LAVRIL T, IZACARD G, et al. LLaMA:open and efficient foundation language models[EB/OL].(2023-02-27)[2025-07-20]. https://arxiv. org/abs/2302.13971.
[19]RAFFEL C, SHAZEER N, ROBERTS A, et al. Exploring the limits of transfer learning with a unified text-totext transformer[J]. Journal of Machine Learning Research, 2020, 21(1):5485-5551.
[20]LEWIS M, LIU Y, GOYAL N, et al. BART:denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension[EB/OL].(2019-10-29)[2025-07-22]. https://arxiv. org/abs/1910.13461
[21]张慧敏. DeepSeek-R1是怎样炼成的[J].深圳大学学报(理工版), 2025, 42(2):226-232.ZHANG Huimin. How DeepSeek-R1 was created[J].Journal of Shenzhen University(Science and Engineering), 2025, 42(2):226-232.
[22]贺正冰.大语言模型在道路交通领域应用:创新与挑战[J].交通运输工程与信息学报, 2025, 23(1):85-92.HE Zhengbing. Large language models in road transportation:innovations and challenges[J]. Journal of Transportation Engineering and Information, 2025, 23(1):85-92.
[23]钱振伟,钱大琳,张辉.城市轨道交通车站客流应急疏散时间影响要素分析[J].大连交通大学学报, 2017, 38(2):6-10.QIAN Zhenwei, QIAN Dalin, ZHANG Hui. Analysis of influencing factors on emergency evacuation time of passenger flow in urban rail transit stations[J]. Journal of Dalian Jiaotong University, 2017, 38(2):6-10.
基本信息:
DOI：10.19961/j.cnki.1672-4747.2025.09.025
中图分类号:U29-39;U298
引用信息:
[1]冷勇林,张宏伟,阴佳腾,等.大语言模型驱动的城市轨道交通突发事件应急响应方法[J].交通运输工程与信息学报,2026,24(01):102-115.DOI:10.19961/j.cnki.1672-4747.2025.09.025.
基金信息:
国家自然科学基金“优青”项目（72322022）;国家自然科学基金“轨道联合”项目（U2469211）
2026-01-07
2026-01-07
2026-01-07