电力物资供应链电子数据信任强度自适应分级方法OA
Adaptive grading method of trust intensity for electronic data in the electric power material supply chain
电力物资供应链电子数据管理需求复杂,传统信任强度分级方法受限于静态规则与单一评价维度,缺乏领域知识引导,难以满足精细化与差异化的应用要求.针对这一问题,提出一种电力物资供应链电子数据信任强度自适应分级方法——Adap-LLM.该方法将专家构建的多维评分体系与分级规则以结构化提示词的形式注入大语言模型,并结合参数高效微调实现领域知识深度融合,从而实现对多种类型电力文档信任强度的自适应分级.实验结果表明,Adap-LLM在私有电力供应链数据集与基于公开数据源构建的跨领域评估数据集上的各项指标均优于对比模型,充分验证了其在电力物资大数据分级管理中的有效性与先进性,具有良好的工程应用价值.
The management of electronic data in power material supply chains is characterized by complex requirements,while traditional trust strength classification methods are constrained by static rules and single evaluation dimensions,and lack sufficient domain knowledge guidance,making it difficult to meet the demands of fine-grained and differentiated applications.To address this issue,an adaptive trust strength classification method for electronic data in power material supply chains,named Adap-LLM,was proposed.In this method,a multi-dimensional scoring system and classification rules constructed by domain experts were injected into a large language model(LLM)in the form of structured prompts,and parameter-efficient fine-tuning was integrated to achieve deep fusion of domain knowledge,thereby enabling adaptive classification of trust strength for various types of power documents.Experimental results demonstrated that Adap-LLM outperforms the compared models across all evaluation metrics on both private power supply chain dataset(PPSCD)and a cross-domain evaluation dataset constructed from public data sources(CDE-PDS),which fully validates its effectiveness and advancement in the hierarchical management of big data in power material supply chains,and exhibits promising engineering application value.
俞春华;郭庆雷;杨珂;柏能;黄宜华
国网江苏省电力有限公司物资分公司,江苏 南京 210036国网区块链科技(北京)有限公司,北京 100077||国网区块链应用技术实验室,北京 100077国网数字科技控股有限公司,北京 100077国网江苏省电力有限公司物资分公司,江苏 南京 210036南京大学计算机软件新技术国家重点实验室,江苏 南京 210023
信息技术与安全科学
供应链电子数据分级大语言模型领域知识提示词工程参数高效微调
supply chain electronic data gradingLLMdomain knowledgeprompt engineeringparameter-efficient fine-tuning
《大数据》 2026 (5)
172-189,18
国家电网公司总部科技项目(No.5700-202418240A-1-1-ZN) State Grid Corporation of China Headquarters Science and Technology Project(No.5700-202418240A-1-1-ZN)
评论