Construction and application of an automated operation and maintenance knowledge base based on large language models

Traditional automated operation and maintenance relies on rule engine and human experience, facing the severe challenges of poor generalization and low response efficiency in dynamic scenarios. In order to break through the bottleneck of complex fault diagnosis and decision-making, this paper proposes an intelligent operation and maintenance framework FlowXpert, which integrates large language model and hybrid knowledge base, realizes dynamic knowledge fusion through the deep coupling of knowledge map and domain FAQ, optimizes workflow generation strategy by combining reinforcement learning, and introduces preference learning mechanism to screen high-confidence schemes. The framework design covers the construction of knowledge base, multi-modal reasoning engine and lightweight execution module, and adapts to the high concurrency and low latency requirements of the production environment. The verification based on the real scene of Huawei cloud shows that the recall rate of the key steps of the system reaches 92.31%, the manual intervention rate is reduced to 7.83%, and the fault handling efficiency is thousands of times higher than that of the traditional method, which provides a solution with both theoretical innovation and industrial value for the field of intelligent operation and maintenance.