大语言模型融合知识图谱的风电运维问答系统研究

doi:10.3969/j.issn.2097-0706.2024.09.008

综合智慧能源 ›› 2024, Vol. 46 ›› Issue (9): 61-68.doi: 10.3969/j.issn.2097-0706.2024.09.008

大语言模型融合知识图谱的风电运维问答系统研究

陈庆(), 柳雨生^*(), 段练达, 梁好, 孙启涛, 鲁纳纳

明阳智慧能源集团股份公司，广东中山 528400

收稿日期:2024-07-03 修回日期:2024-08-13 出版日期:2024-09-25
通讯作者: *柳雨生（1996），男，数据分析工程师，硕士，从事风电智能运维、风电机组异常预警等方面的研究，liuyusheng@mywind.com.cn。
作者简介:陈庆（1997），男，数据分析工程师，硕士，从事风电智能运维、旋转机械图像异常识别等方面的研究，chenqing@mywind.com.cn。
基金资助:
明阳智慧能源集团股份公司科技项目(M0B00022271)

Research on a wind power operation and maintenance Q&A system based on large language models and knowledge graphs

CHEN Qing(), LIU Yusheng^*(), DUAN Lianda, LIANG Hao, SUN Qitao, LU Nana

Mingyang Smart Energy Group，Zhongshan 528400，China

Received:2024-07-03 Revised:2024-08-13 Published:2024-09-25
Supported by:
Science and Technology Projects by Mingyang Smart Energy Group(M0B00022271)

摘要/Abstract

摘要：

风电场运维工作高度依赖现场实践经验，而行业内人员高流动性带来经验传递难题，传统知识库和问答系统日益暴露出其局限性。为提高问答系统在专业领域的适用性和可靠性，设计了一种融合大型语言模型和知识图谱的风电运维问答系统，通过语义理解和关联性分析，整合结构化与非结构化数据，提供全面、准确的专业回答。主、客观评价表明，该专业问答模型的准确性、连贯性及信息量均优于某中文大语言模型及ChatGLM模型，不仅提升了风电运维的效率，也为行业知识传承和更新提供了解决方案。

关键词: 风电运维, 大语言模型, 知识图谱, 双基座问答系统, ChatGLM模型

Abstract:

The operation and maintenance of wind farms heavily rely on on-site practical experience， while the high turnover rate in the industry poses challenges to the impartment of such experience. Traditional knowledge bases and Q&A systems are increasingly revealing their limitations in this regard. To enhance the applicability and reliability of Q&A systems in professional domains， this paper designs a wind farm operation and maintenance Q&A system that integrates large language models （LLMs） with knowledge graphs. Through semantic understanding and correlation analysis， the system combines both structured and unstructured data to provide comprehensive and accurate professional responses. Both subjective and objective evaluations indicate that the accuracy， coherence， and informativeness of this specialized Q&A model surpass those of a certain Chinese LLM and the ChatGLM model. This not only improves the efficiency of wind farm operation and maintenance but also offers a solution for knowledge transfer and updating within the industry.

Key words: wind operation and maintenance, large language model, knowledge graph, dual-based Q&A system, ChatGLM model

中图分类号:

TK81：TP24

陈庆, 柳雨生, 段练达, 梁好, 孙启涛, 鲁纳纳. 大语言模型融合知识图谱的风电运维问答系统研究[J]. 综合智慧能源, 2024, 46(9): 61-68.

CHEN Qing, LIU Yusheng, DUAN Lianda, LIANG Hao, SUN Qitao, LU Nana. Research on a wind power operation and maintenance Q&A system based on large language models and knowledge graphs[J]. Integrated Intelligent Energy, 2024, 46(9): 61-68.

图/表 9

图1

图2

图3

表1

图4

图5

图6

图7

表2

参考文献 23

[1]	马义松, 武志刚. 基于Neo4j的电力大数据建模及分析[J]. 电工电能新技术, 2016, 35(2):24-30.
	MA Yisong, WU Zhigang. Modeling and analysis of electric power big data utilizing Neo4j[J]. Advanced Technology of Electrical Engineering and Energy, 2016, 35(2):24-30.
[2]	张兴. 风电运营大数据关键技术[J]. 电子技术与软件工程, 2019(10):154-156.
	ZHANG Xin. Critical technologies for big data analytics in wind energy operations[J]. Electronic Technology & Software Engineering, 2019(10):154-156.
[3]	刘瑞宏, 谢国强, 苑宗港, 等. 基于知识图谱的智能故障诊断研究[J]. 邮电设计技术, 2020(10):30-35.
	LIU Ruihong, XIE Guoqiang, YUAN Zonggang, et al. Research on intelligent fault diagnosis based on knowledge graph[J]. Designing Techniques of Posts and Telecommunications, 2020(10):30-35.
[4]	吕梦平. 基于知识图谱技术的风电数据管理与应用研究[D]. 湘潭: 湘潭大学, 2021.
	LYU Mengping. Research on management and application of wind power data based on knowledge graph technology[D]. Xiangtan: Xiangtan University, 2021.
[5]	吕梦平, 段斌, 蒋海辉, 等. 基于知识图谱技术的风电数据管理与应用研究[J]. 电力系统保护与控制, 2021, 49(6):167-173.
	LYU Mengping, DUAN Bin, JIANG Haihui, et al. Research on management and application of wind power data based on knowledge graph technology[J]. Power System Protection and Control, 2021, 49(6):167-173.
[6]	刘梓权, 王慧芳. 基于知识图谱技术的电力设备缺陷记录检索方法[J]. 电力系统自动化, 2018, 42(14): 158-164.
	LIU Ziquan, WANG Huifang. Retrieval Method for defect records of power equipment based on knowledge graph technology[J]. Automation of Electric Power Systems, 2018, 42(14):158-164.
[7]	CHATTERJEE J, DETHLEFS N. Automated question-answering for interactive decision support in operations & maintenance of wind turbines[J]. IEEE Access, 2022, 10:84710-84737.
[8]	萨日娜, 李艳玲, 林民. 知识图谱推理问答研究综述[J]. 计算机科学与探索, 2022, 16(8):1727-1741. doi: 10.3778/j.issn.1673-9418.2111033
	SA Rina, LI Yanling, LIN Min. Survey of question answering based on knowledge graph reasoning[J]. Journal of Frontiers of Computer Science & Technology, 2022, 16(8):1727-1741.
[9]	CHOWDHERY A, NARANG S, DEVLIN J, et al. PaLM: Scaling language modeling with pathways[J]. arXiv preprint arXiv, 2022:2204.02311.
[10]	GUU K, LEE K, TUNG Z, et al. REALM: Retrieval-augmented language model pre-training[J]. International Conference on Machine Learning, 2020, 119:3929-3938.
[11]	BROWN T B, MANN B, RYDER N, et al. Language models are few-shot learners[C]// Conference and Workshop on Neural Information Processing Systems. NIPS, 2020, 33:1877-1901.
[12]	MAYNEZ J, NARAYAN S, BOHNET B, et al. On faithfulness and factuality in abstractive summarization[C]// Annual Meeting of the Association for Computational Linguistics. ACL, 2020:1906-1919.
[13]	赵俊华, 文福拴, 黄建伟, 等. 基于大语言模型的电力系统通用人工智能展望:理论与应用[J]. 电力系统自动化, 2024, 48(6):13-28.
	ZHAO Junhua, WEN Fushuan, HUANG Jianwei, et al. Prospect of artificial general intelligence for power systems based on large language model: Theory and applications[J]. Automation of Electric Power Systems, 2024, 48(6):13-28.
[14]	董俊, 束洪春, 刘瑞, 等. 大语言模型赋能场景生成和双层优化的多农业园区供电-灌溉-蓄水耦合运行[J]. 高电压技术, 2024, 50(7):2906-2917.
	DONG Jun, SHU Hongchun, LIU Rui, et al. Large language models empowering scenario generation and dual-layer optimization for coupled operations of power supply, irrigation, and water storage in multiple agricultural parks[J]. High Voltage Engineering, 2024, 50(7):2906-2917.
[15]	陈艳波, 方哲, 张宁, 等. 基于大语言模型绿电预测和绿电交易的园区综合能源系统集群多目标协同运行方法[J]. 高电压技术, 2024, 50(7):2849-2863.
	CHEN Yanbo, FANG Zhe, ZHANG Ning, et al. Multi-objective collaborative operation method for park-level integrated energy system cluster based on large language model for green electricity prediction and trading[J]. High Voltage Engineering, 2024, 50(7):2849-2863.
[16]	胡志强, 潘鑫瑜, 文思捷, 等. 结合多模态知识图谱与大语言模型的风机装配工艺问答系统[J]. 机械设计, 2023, 40(S2):20-26.
	HU Zhiqiang, PAN Xinyu, WEN Sijie, et al. Assembly process question answering system of wind turbines combining multi-modal knowledge graphs with LLMs[J]. Journal of Machine Design, 2023, 40(S2):20-26.
[17]	DOUZE M, GUZHVA A, DENG C Q, et al. The Faiss library[J]. arXiv preprint arXiv, 2024:2401.08281.
[18]	DU Z X, QIAN Y J, LIU X, et al. GLM: General language model pretraining with autoregressive blank infilling[C]// Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics. ACL, 2022.
[19]	马艺洁, 赖海光, 刘子威, 等. 实体识别技术研究进展综述[J]. 太赫兹科学与电子信息学报, 2024, 22(5):503-515.
	MA Yijie, LAI Haiguang, LIU Ziwei, et al. Journal of terahertz science and electronic information technology[J]. Journal of Terahertz Science and Electronic Information Technology, 2024, 22(5):503-515.
[20]	RONG X. Word2vec parameter learning explained[J]. Computing Research Repository, 2014:1411.2738.
[21]	MARTIN G. Word2vec, node2vec, graph2vec, X2vec: Towards a theory of vector embeddings of structured data[C]// Proceedings of the 39th ACM SIGMOD-SIGACT-SIGAI Symposium on Principles of Database Systems. ACM SIGACT-SIGMOD-SIGART, 2020:1-16.
[22]	YESIR S, SOGUKPINAR I. Malware detection and classification using fastText and BERT[C]// 2021 9th International Symposium on Digital Forensics and Security.ISDFS, 2021: 1-6.
[23]	QIU Y F, LI H Y, QU Y Q, et al. DuReader_retrieval: A large-scale Chinese benchmark for passage retrieval from web search engine[C]// Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. EMNLP, 2022:5326-5338.

数据来源	描述	数据/条
故障工单&预警工单	包含风电机组触发的故障运维、预警工单信息，包括故障现象、处理手段等经验知识	10 725
质量反馈单	涵盖现场到场处理的工单信息和物料备件信息	9 424
故障代码手册	涵盖各类风电机组的触发逻辑条件	1 962
故障处理手册	涵盖风电机组运维积累的故障分析树知识	368
风电机组电子档案	包含风电机组基础信息，如部件品牌、规格等	250 000

模型	余弦距离中位数	相似度中位数	R₅₀/%
某中文LLM	0.819	0.530	52
ChatGLM	0.815	0.590	66
专业问答	0.821	0.885	90

大语言模型融合知识图谱的风电运维问答系统研究

Research on a wind power operation and maintenance Q&A system based on large language models and knowledge graphs

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 9

参考文献 23

相关文章 2

编辑推荐

Metrics

本文评价

[1]	蔺家骏, 闫玮丹, 胡俊华, 郑一鸣, 邵先军, 郭炳延. 多模态知识图谱在电力运检中的应用与展望[J]. 综合智慧能源, 2024, 46(1): 65-74.
[2]	王永真, 韩艺博, 韩恺, 韩俊涛, 宋阔, 张兰兰. 基于知识图谱的数据中心综合能源系统研究综述[J]. 综合智慧能源, 2023, 45(7): 1-10.