面向发电设备故障辅助诊断的知识增强大模型方法
Knowledge-Enhanced LLMs for Assisted Fault Diagnosis of Power Generation Equipment
温荣臻 1周芃睿 1李盛祖 1冯琢 1赵乔乔1
作者信息
摘要
发电设备故障辅助诊断需要结合设备结构、运行状态、规程和历史案例，而通用大语言模型直接生成的判断可能缺少可核对的依据。本文提出一种知识增强大模型辅助诊断方法，将设备与故障关系、规程条款和案例组织为带适用范围、来源、版本及审核状态的知识单元；以诊断问题和当前状态共同驱动结构化筛选与语义检索，经语义相似度值域映射、信息覆盖检查和分类选取构建上下文；按异常概况、可能原因、诊断依据、检查建议、风险提示和待确认信息生成草稿，并设置对象一致性、知识支持性、引用有效性和安全边界校核，以及有限重试与人工接管规则。以轴承温度异常的示意案例说明知识绑定、校核和两类复核路径。本文为方法设计研究，未开展定量验证。该方法服务于专业人员的资料整理和辅助分析，不替代保护逻辑、运行规程及现场判断。
Abstract
Assisted fault diagnosis of power generation equipment requires information on equipment structures, operating states, procedures, and historical cases. Judgments generated directly by general-purpose large language models may lack verifiable support. This paper proposes a knowledge-enhanced method that organizes equipment and fault relations, procedure provisions, and cases into knowledge units with applicability, provenance, versions, and review status. The diagnostic question and current state jointly drive structured filtering and semantic retrieval. Semantic-score range mapping, information-coverage checks, and category-based selection produce the diagnostic context. The model drafts six fields: anomaly summary, possible causes, diagnostic evidence, inspection recommendations, risk alerts, and information requiring confirmation. Object consistency, knowledge support, citation validity, and safety-boundary checks are specified, together with bounded retries and human escalation. An illustrative bearing-temperature scenario explains evidence binding, validation, and two review paths. This is a method-design study without quantitative validation. The method supports professional information organization and analysis; it does not replace protection logic, operating procedures, or on-site judgment.关键词
发电设备/故障诊断/大语言模型/知识增强/检索增强生成Key words
power generation equipment/fault diagnosis/large language models/knowledge enhancement/retrieval-augmented generation引用本文复制引用
温荣臻,周芃睿,李盛祖,冯琢,赵乔乔.面向发电设备故障辅助诊断的知识增强大模型方法[EB/OL].(2026-10-16)[2026-10-17].http://www.paper.edu.cn/releasepaper/content/202610-11.学科分类
发电、发电厂
