Abstract: In response to the problem that traditional methods are difficult to cope with the complexity and dynamic changes of large-scale networks, review of collaborative Large Language Model (LLM) for network operations and maintenance were summarized. First, the applications of LLM in the network modeling stage were reviewed, and their advantages in dynamic traffic measurement and heterogeneous configuration semantic parsing were analyzed. Secondly, the paths of LLM-empowered network simulation were explored, and their intelligent auxiliary roles in achieving high fidelity, elastic resource allocation, and distributed large-scale simulation were expounded. Thirdly, the roles of LLM in network verification were discussed, especially their capabilities in transforming natural language intents into formal specifications and generating intelligent test cases. Then, the applications of LLM in fault diagnosis and repair were analyzed. Finally, a closed-loop framework for LLM-collaborative network operations and maintenance integrating modeling, simulation, verification, and repair was proposed. The research results show that the heterogeneous configuration semantic alignment problem in network modeling is effectively solved by LLM technologies relying on their powerful semantic understanding and reasoning capabilities. The efficiency bottleneck of large-scale network simulation is broken through, and the level of interactive intelligence is improved. Furthermore, the threshold for using formal verification is significantly reduced through natural language intent translation, and the evolution of fault diagnosis and repair from passive response to intent-driven automated closed-loop is promoted. In conclusion, the traditional operations and maintenance gap is effectively bridged by large models through strengthening semantic reasoning and constructing a full-link closed loop, and core support is provided for moving towards a highly autonomous intelligent network systemAs network scales continue to expand, heterogeneity increases, and service diversity explodes, network operations and maintenance (O&M) play a crucial role in ensuring the efficient, reliable, and intelligent operation of large-scale network environments. However, traditional O&M technologies, which rely on single data-driven models or formal logic methods, struggle to cope with the complexity and dynamic nature of modern networks. Fortunately, artificial intelligence technologies, led by Large Language Models (LLMs), offer new opportunities for the intelligent transformation of network O&M, leveraging their exceptional general semantic understanding, dynamic knowledge generation, and complex reasoning capabilities. Integrating the powerful general-purpose capabilities of large models with the highly precise and rigorous tasks of network management necessitates innovative fusion paths in areas such as network modeling, simulation, validation, diagnosis, and repair. This need is fostering a remarkable new research direction, one that deeply integrates the reasoning capabilities of LLMs with the precision of traditional network technologies, jointly pioneering a new paradigm for intelligent O&M driven by data-logic synergy. This paper focuses on the application of large models in network O&M, reviewing the research progress in the four major stages of modeling, simulation, validation, and repair, and concludes by discussing future directions.

Key words: Large Language Model, Network Modeling, Network Simulation, Network Verification, Network Fault Diagnosis and Repair

摘要： 针对传统方法难以应对大规模网络复杂性与动态变化的问题，综述了大语言模型(LLM)协同的网络运维。首先，梳理了LLM在网络建模阶段的应用，重点分析其在动态流量测量与异构配置语义解析中的优势；其次，探讨了LLM赋能网络仿真的路径，阐述其在实现高保真、弹性资源分配及分布式大规模仿真中的智能化辅助作用；再次，论述了LLM在网络验证中的角色，特别是其将自然语言意图转化为形式化规约及生成智能测试用例的能力；继次，分析了LLM在故障诊断与修复中的应用；最后，提出了一个融合建模、仿真、验证及修复的LLM协同运维闭环框架。现有研究结果表明，LLM技术凭借其强大的语义理解与推理能力，能够有效解决网络建模中的异构配置语义对齐难题 ，突破大规模网络仿真的效率瓶颈并提升交互智能化水平，通过自然语言意图转译显著降低形式化验证的使用门槛，并推动故障诊断与修复从被动响应向意图驱动的自动化闭环演进。综上，LLM通过强化语义推理与构建全链路闭环，有效弥合了传统运维断层，为迈向高度自治的智能网络体系提供了核心支撑随着网络规模的持续扩张、异构性的增加和业务多样性的爆发，网络运维在保障大规模网络环境的高效、可靠、智能化运行中扮演着至关重要的角色。然而，传统依赖于单一数据驱动模型或形式化逻辑方法的运维技术，已难以应对现代网络的复杂性与动态变化。幸运的是，以大语言模型为代表的人工智能技术，凭借其卓越的通用语义理解、动态知识生成与复杂推理能力，为网络运维的智能化变革提供了新的机遇。将大模型强大的通用能力与高度精确严谨的网络管理任务相结合，需要在网络建模、仿真、验证、诊断与修复等环节寻求创新的融合路径。这一需求正催生一个令人瞩目的新研究方向，将大模型的推理能力与传统网络技术的精确性深度融合，共同开创数据与逻辑协同驱动的智能运维新范式。本文聚焦于大模型在网络运维中的应用，回顾了其在建模、仿真、验证和修复这四大阶段的研究进展，并探讨了未来的发展方向。

关键词: 大语言模型, 网络建模, 网络仿真, 网络验证, 网络错误诊断与修复