人工智能大模型在电力设备运维场景中的应用探讨  

Application of Artificial Intelligence Large Language Model in Power Equipment Operation and Maintenance

作　　者：陈晓红 傅文润[1,2] 刘朝明 刘泽洪 李俊朋 胡志亮 胡东滨 Chen Xiaohong;Fu Wenrun;Liu Chaoming;Liu Zehong;Li Junpeng;Hu Zhiliang;Hu Dongbin(School of Management,Xi’an Jiaotong University,Xi’an 710049,China;Xiangjiang Laboratory,Changsha 410205,China;Bussiness School,Central South University,Changsha 410083,China;Global Energy Interconnection Development and Cooperation Organization,Beijing 100031,China)

机构地区：[1]西安交通大学管理学院,西安710049 [2]湘江实验室,长沙410205 [3]中南大学商学院,长沙410083 [4]全球能源互联网发展合作组织,北京100031

出　　处：《中国工程科学》2025年第1期180-192,共13页Strategic Study of CAE

基　　金：中国工程院咨询项目“全球未来产业发展趋势及湖南未来产业布局研究”(2024-DFZD-39);湘江实验室重大项目(23XJ01006)。

摘　　要：电力设备运维是新型电力系统建设的重要环节,以人工智能(AI)大模型技术为代表的AI技术变革为传统电力设备运维的数智化提供了新机遇。本文探讨了多模态AI大模型对电力设备健康状态评估、电力设备运行状态预测、电力设备故障诊断、电力设备寿命预测、电力设备故障检修策略推荐等电力运维具体场景的赋能作用,辨识了数据问题制约电力AI大模型的应用成效、“算法黑箱”影响智能运维辅助决策的透明度与可靠性、环境变化导致电力AI大模型性能衰退等多模态AI大模型赋能电力设备运维的技术难点。着眼攻克相关技术难点,结合知识图谱检索增强生成、多模态对齐、微调和持续学习等大模型应用优化技术,构建了基于多模态AI大模型的电力设备运维系统架构,梳理了多模态AI大模型在电力设备运维场景应用时涉及的需求分析、模型训练、应用部署、运营管理等主要阶段的实现过程,进而提出了持续监控并优化数据质量、采用持续学习算法、建立模型性能反馈循环机制等大模型性能持续优化策略。进一步探讨了多模态AI大模型赋能电力设备运维的应用趋势和发展保障举措,以深化对电力设备智能运维领域的前沿技术认知,推动构建智能化、智慧化的新型电力系统。The operation and maintenance of power equipment is a crucial aspect of the construction of new power systems.The artificial intelligence large language model(AI-LLM)presents significant opportunities for the digital intelligence of traditional power equipment operation and maintenance.This study aims to explore the enabling role of multimodal AI-LLM in health assessment,operational state prediction,fault diagnosis,life prediction,and maintenance strategy recommendation,among other specific scenarios of power equipment operation and maintenance.Additionally,this study analyzes the challenges faced by multimodal AI-LLM in enabling power equipment operation and maintenance,including the varying quality of multimodal data,the“black box”nature of algorithms leading to low transparency in decision-making processes,and model performance deterioration induced by environmental changes.To address these challenges,this study combines knowledge graph retrieval-augmented generation,multimodal alignment,fine-tuning and continuous learning,and other big model application optimization techniques to construct an AI-LLM power equipment maintenance system.It then sorts out the implementation process of multimodal AI-LLM in the operation and maintenance of power equipment,covering four stages:demand analysis,model training,application deployment,and operational management.Furthermore,strategies for continuously optimizing model performance are proposed,including the continuous monitoring and optimization of data quality,use of continuous learning algorithms,and establishment of a feedback loop mechanism for model performance.Finally,this study explores the future directions for multimodal AI-LLM in the field of power equipment operation and maintenance and provides a series of implementation safeguards to promote the intelligent transformation of power equipment operation and maintenance and support the construction of new power systems.

关 键 词：新型电力系统 电力设备运维 多模态AI大模型 检索增强生成 知识图谱 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] D926[自动化与计算机技术—控制科学与工程]