针对特种设备安全监管中多源异构数据难以统一访问、制约监管效率的问题，本研究提出一种基于智能体专业化分工的特种设备多源异构数据统一处理方法。该方法通过文档-页面-段落3级分层检索机制，平衡检索精度与计算开销，并构建由主智能体、结构化查询语言（Structured Query Language，SQL）子智能体与知识库智能体协同的多智能体系统，实现多源异构数据的统一语义和可解释性访问，同时采用面向特种设备监管领域的加权倒数排名融合（Reciprocal Rank Fusion，RRF）策略进行结果优化。实验结果表明，所提方法在检索准确率、覆盖率和查询成功率上均优于传统单一检索方法与统一智能体架构，且能有效支持特种设备安全监管中的风险评估、检验合规性验证、事故调查等实际监管场景，为提升特种设备安全监管水平提供技术解决方案。
Abstract
To address the difficulty in accessing multi-source heterogeneous data and constraining regulatory efficiency in special equipment safety supervision, this study proposes a unified processing framework based on agent specialization. The approach integrates a document-page-paragraph hierarchical retrieval mechanism to achieve an optimal balance between retrieval precision and computational overhead. A multi-agent system, comprising an orchestrator agent, an structured query language (SQL) agent, and a knowledge base agent, is designed to facilitate coordinated information processing, enabling unified semantic representation and interpretable data access. To enhance result consistency and relevance, a weighted reciprocal rank fusion (RRF) strategy, which customized for the domain-specific characteristics of safety supervision, is employed for effective result fusion. Experimental results show that the proposed method consistently outperforms both conventional single-retrieval approaches and a monolithic unified agent architecture across key performance metrics, including retrieval accuracy, coverage, and query success rate. Furthermore, the method demonstrates strong applicability in critical operational scenarios such as risk assessment, inspection compliance verification, and accident investigation, thereby offering a scalable and robust technical foundation for advancing the effectiveness of special equipment safety supervision.
关键词
特种设备 /
多源异构数据 /
多智能体系统 /
多阶段混合检索 /
统一数据访问
Key words
Special equipment /
Multi-source heterogeneous data /
Multi-agent system /
Multi-stage hybrid retrieval /
Unified data access
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 中华人民共和国特种设备安全法[EB/OL]. (2013-06-29)[2025-12-03]. https://www.gov.cn/flfg/2013-06/30/content_2437160.htm.
[2] 国家市场监督管理总局. 特种设备安全与节能事业发展“十四五”规划[EB/OL]. (2022-12-17)[2026-02-08].https://www.samr.gov.cn/zw/zfxxgk/fdzdgknr/tzsbs/art/2023/art_d9664fb21b4d4167b9bc052c0e96b469.html.
[3] 蓝麒, 曹宏伟, 郝素利, 等. 典型特种设备质控数字化关键技术研究与应用概述[J]. 中国特种设备安全, 2024, 40(1): 3-8.
[4] 金益斌, 张玉媛, 石坤, 等. 基于领域知识增强大模型的特种设备法规智能问答[J]. 物联网技术, 2025, 15(17): 88-92.
[5] LEWIS P, PEREZ E, PIKTUS A, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver: Curran Associates Inc., 2020: 9 459-9 474.
[6] GAO Y F, XIONG Y, GAO X Y, et al.Retrieval-augmented generation for large language models: A survey[PP/OL].V5.arXiv(2023-12-18)[2026-02-13].https://doi.org/10.48550/arXiv.2312.10997.
[7] 赵衍, 程恺. 基于检索增强生成技术的规章制度问答大语言模型的构建[J]. 计算机应用与软件, 2025, 42(7): 192-198.
[8] 齐思洋，胡慧云，李洪冰，等. 融合大语言模型的领域问答系统构建方法[J]. 北京邮电大学学报, 2024, 47(4):50-56.
[9] 朱新立，高志强，姬纬通，等. 武信：一种垂直领域大语言模型系统架构设计与实证[J]. 数据采集与处理, 2025, 40(3): 637-646.
[10] 沈佳楠，崔翛龙，高志强，等. 大语言模型问答任务准确性评价方法及基于微调的垂直领域优化研究[J]. 网络安全与数据治理, 2025, 44(S1): 36-44.
[11] 刘渊, 李尼亚, 井科学, 等. 特种设备风险防控多源异构数据统一语义表征方法研究[J]. 中国特种设备安全, 2025, 41(9): 31-37.
[12] XI Z, CHEN W, GUO X, et al. The rise and potential of large language model based agents: A survey[PP/OL].V3. arXiv(2023-09-19)[2026-02-13].https://arxiv.org/ab5/2309.07864.
[13] 王志远, 张伟, 官炳政, 等. 基于多智能体协同图RAG方法的工业问答研究[J/OL]. 计算机工程, 2025: 1-10[2026-02-08]. https://doi.org/10.19678/j.issn.1000-3428.0252696.
[14] ROBERTSON S, ZARAGOZA H. The probabilistic relevance framework: BM25 and beyond[J]. Foundations and Trends® in Information Retrieval, 2009, 3(4): 333-389.
[15] KARPUKHIN V, OGUZ B, MIN S, et al. Dense passage retrieval for open-domain question answering[C]//Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: Association for Computational Linguistics, 2020: 6 769-6 781.
[16] XIONG L, XIONG C, LI Y, et al. Approximate nearest neighbor negative contrastive learning for dense text retrieval[C]//Proceedings of International Conference on Learning Representations. OpenReview.net, 2021.
[17] 刘腾, 王朗, 孙英杰, 等. 面向垂直领域智能问答的分层检索增强生成方法[J]. 网络安全与数据治理, 2025, 44(S1): 125-131.
[18] HOFSTÄTTER S, ALTHAMMER S, SCHRÖDER M, et al. Improving efficient neural ranking models with cross-architecture knowledge distillation[PP/OL].V2.arXiv(2021-01-22)[2026-02-13].http://arxiv.org/abs/2010.02666.
[19] 厉志安, 李笑, 鲍雨田, 等. 压力管道检验知识图谱溯源系统设计与应用[J]. 计算机应用与软件, 2025, 42(8): 80-85.
[20] 陈明, 杨静, 周涛. 知识图谱在法规标准智能检索中的应用研究[J]. 中文信息学报, 2023, 37(4): 112-120.
[21] CORMACK G V, CLARKE C L A, BUETTCHER S. Reciprocal rank fusion outperforms condorcet and individual rank learning methods[C]//Proceedings of the 32nd International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: Association for Computing Machinery, 2009: 758-759.
[22] SARON S, DEGENARO D, GUALLAR-BLASCO J, et al. MMMORRF: Multimodal multilingual modularized reciprocal rank fusion[C]//Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: Association for Computing Machinery,2025:4 004-4 009.
基金
国家重点研发计划项目（2023 YFC3010401）
