基于预训练大模型的业务流程实体与关系的自动提取和评估
Automated extraction and evaluation of business process entities and relations via pre-trained models
- 电信科学 2026年42卷第3期 页码：169-179
- 作者机构：
1.四川中电启明星信息技术有限公司，四川 成都 610213
2.国网信息通信产业集团有限公司，北京 100052
3.国网辽宁省电力有限公司电力科学研究院，辽宁 沈阳 110001
- 作者简介：
[ "万向（1980- ），男，四川中电启明星信息技术有限公司工程师，主要研究方向为电力营销与数据应用。" ]
[ "李杉（1995- ），男，国网信息通信产业集团有限公司研发中心工程师，主要研究方向为电力信息化。" ]
[ "刘泽三（1985- ），男，国网信息通信产业集团有限公司研发中心高级工程师，主要研究方向为电力信息化。" ]
[ "张彬（1981- ），男，博士，国网辽宁省电力有限公司电力科学研究院高级工程师，主要研究方向为电力人工智能及数字化新技术。" ]
[ "佟帅辰（1998- ），男，国网辽宁省电力有限公司电力科学研究院助理工程师，主要研究方向为人工智能。" ]
[ "苟艺（1988- ），男，四川中电启明星信息技术有限公司工程师，主要研究方向为能源数字化。" ]
[ "赵淑曼（1998- ），女，四川中电启明星信息技术有限公司业务咨询师，主要从事数据分析工作。" ]
- 基金信息：国家电网公司科技创新项目“作业流程与标准的数字化融合技术研究与应用”(5700-202319836A-4-2-KJ)
DOI：10.11959/j.issn.1000-0801.2026031
中图分类号： TP311.52收稿：2025-06-03，
修回：2025-07-11，
录用：2025-09-25，
纸质出版：2026-03-20
- 稿件说明：
移动端阅览
万向,李杉,刘泽三等.基于预训练大模型的业务流程实体与关系的自动提取和评估[J].电信科学,2026,42(03):169-179.
Wan Xiang,Li Shan,Liu Zesan,et al.Automated extraction and evaluation of business process entities and relations via pre-trained models[J].Telecommunications Science,2026,42(03):169-179.
万向,李杉,刘泽三等.基于预训练大模型的业务流程实体与关系的自动提取和评估[J].电信科学,2026,42(03):169-179. DOI： 10.11959/j.issn.1000-0801.2026031.
Wan Xiang,Li Shan,Liu Zesan,et al.Automated extraction and evaluation of business process entities and relations via pre-trained models[J].Telecommunications Science,2026,42(03):169-179. DOI： 10.11959/j.issn.1000-0801.2026031.
摘要
在数字化转型背景下，精准建模电力行业业务流程对实现智能运营与标准化管理具有重要意义。为此，提出了一种面向电力工作流程标准化的领域自适应语言模型ElecBPM-LoRA。该模型通过引入低秩微调（low-rank adaptation，LoRA）技术与结构化提示词机制，能够将自然语言准确转化为结构化语句，并自动生成符合BPMN 2.0规范的业务流程图。实验表明，ElecBPM-LoRA在电力行业流程要素的识别任务中的表现均优于目前主流大语言模型，且在流程结构建模的完整性与语义一致性方面表现出显著优势，为关键基础设施领域的流程自动化与标准化建模提供了可靠支撑。
Abstract
In the context of digital transformation
accurate modeling of power industry business processes is considered crucial for achieving intelligent operation and standardized management. To this end
a domain-adaptive language model named ElecBPM-LoRA was proposed for power workflow standardization. By incorporating low-rank adaptation (LoRA) and a structured prompting mechanism
the model was enabled to accurately transform natural language into structured statements and automatically generate business process diagrams compliant with the BPMN 2.0 specification. Experimental results demonstrate that ElecBPM-LoRA outperforms mainstream large language models in identifying power industry process elements
and exhibits significant advantages in the completeness of process structure modeling and semantic consistency. The proposed approach provides reliable support for process automation and standardized modeling in the field of critical infrastructure.
关键词
Keywords
references
Grohs M , Abb L , Elsayed N , et al . Large language models can accomplish business process management tasks [C ] // Business Process Management Workshops . Cham : Springer , 2024 : 453 - 465 .
Sola D , Meilicke C , Van D A H , et al . A rule-based recommendation approach for business process modeling [C ] // Advanced Information Systems Engineering . Cham : Springer , 2021 : 328 - 343 .
Beheshti S M R , Benatallah B , Sakr S , et al . Process analytics: concepts and techniques for querying and analyzing process data [M ] . Cham : Springer , 2016 .
Dumas M , La Rosa M , Mendling J , et al . Fundamentals of business process management [M ] . Berlin, Heidelberg : Springer , 2013 .
Beheshti A , Ghodratnama S , Elahi M , et al . Social data analytics [M ] . Boca Raton : CRC Press , 2022 .
Delgado A , Calegari D , García F , et al . Model-driven management of BPMN-based business process families [J ] . Software and Systems Modeling , 2022 , 21 ( 6 ): 2517 - 2553 .
Schunselaar D M M , Slaats T , Maggi F M , et al . Mining hybrid business process models: a quest for better precision [C ] // Business Information Systems . Cham : Springer , 2018 : 190 - 205 .
Manning C , Surdeanu M , Bauer J , et al . The Stanford CoreNLP natural language processing toolkit [C ] // Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations . Stroudsburg, PA, USA : ACL , 2014 : 55 - 60 .
Padró L , Stanilovsky E . FreeLing 3.0: towards wider multilinguality [C ] // Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12) . Kerrville : Association for Computational Linguistics , 2012 : 2473 - 2479 .
Bird S , Klein E , Loper E . Natural language processing with python: analyzing text with the natural language toolkit [M ] . Sebastopol, CA : O'Reilly Media , 2009 .
Bommarito II M J , Katz D M , Detterman E M . LexNLP: Natural language processing and information extraction for legal and regulatory texts [PP ] . V1. arXiv ( 2018-06-10 )[ 2025-04-03 ] . arXiv: 1806.03688 .
Luo R Q , Sun L A , Xia Y C , et al . BioGPT: generative pre-trained transformer for biomedical text generation and mining [J ] . Briefings in Bioinformatics , 2022 , 23 ( 6 ): bbac409 .
Chalkidis I , Fergadiotis M , Malakasiotis P , et al . LEGAL-BERT: the muppets straight out of law school [PP ] . V1. arXiv ( 2020-10-06 )[ 2025-04-03 ] . arXiv: 2010.02559 .
Wang N , Yang H Y , Wang C D . FinGPT: instruction tuning benchmark for open-source large language models in financial datasets [PP ] . V2. arXiv ( 2023-11-11 )[ 2025-04-03 ] . arXiv: 2310.04793 .
DeepSeek-AI , Guo D Y , Yang D J , et al . DeepSeek-R1: incentivizing reasoning capability in LLMs via reinforcement learning [PP ] . V2. arXiv ( 2026-01-04 )[ 2025-04-03 ] . arXiv: 2501.12948 .
Pfister R , Jud H . Understanding and benchmarking artificial intelligence: OpenAI’s O3 is not AGI [PP ] . V1. arXiv ( 2025-01-13 )[ 2025-04-03 ] . arXiv: 2501.07458 .
Guo D , Wu F M , Zhu F D , et al . Seed 1 .5-VL technical report[PP ] . V1. arXiv ( 2025-05-11 )[ 2025-05-23 ] . arXiv: 2505.07062 .
0
浏览量
33
下载量
0
CSCD
- 网络PDF下载
- WORD下载
- XML下载
- Meta-XML下载
- BibTeX下载
- Endnote下载
- RefMan 下载
- NoteExpress下载
- NoteFirst下载
关联资源
相关文章
相关作者
相关机构
- 技术支持由北京北大方正电子有限公司提供 京ICP备09064830号-19
京公网安备11010802024621 - 本系统建议在Chrome、 IE9+ 以上版本浏览器阅读本站内容，360浏览器请切换至极速模式
- Cookies帮助我们提供服务并提供个性化体验。使用本网站，即表示您同意我们使用Cookies
- 总访问量：952130 今日访问量：1298