Application of Large Language Models in Intelligent Operation and Maintenance of Information Systems
大语言模型在信息系统智能运维中的应用

摘要

信息系统规模扩大与架构复杂化增加了日志分析、告警处置和故障定位的难度。大语言模型具有语义 理解、知识组织和任务生成能力，可用于日志解释、告警归并、根因分析、知识问答及脚本辅助。文章分析 其技术基础与应用方式，构建数据、知识、模型和工具协同的运维架构，并讨论模型幻觉、数据安全、执行 权限与推理成本等约束。研究认为，大语言模型适合作为智能运维的语义中枢和决策辅助工具，生产部署宜 采用检索增强、规则校验、权限隔离和人工确认机制。

Abstract

The expansion of information system scale and the increasing complexity of system architecture have made log analysis, alarm handling, and fault localization more difficult. Large language models have capabilities in semantic understanding, knowledge organization, and task generation, and can be applied to log interpretation, alarm aggregation, root cause analysis, knowledge-based question answering, and script assistance. This paper analyzes their technical basis and application modes, constructs an operation and maintenance architecture integrating data, knowledge, models, and tools, and discusses constraints such as model hallucination, data security, execution permissions, and inference costs. The study suggests that large language models are suitable as semantic hubs and decision-support tools for intelligent operation and maintenance. In production deployment, mechanisms such as retrieval augmentation, rule verification, permission isolation, and human confirmation should be adopted.