Simulation Analysis of an Intelligent Operations and Monitoring Data Indicator Warning and Diagnosis System Based on a Large Language Model-Enhanced Transformer

With the widespread application of cloud computing and modern microservice architectures, the data from intelligent operations and maintenance (AIOps) for information technology systems exhibit complex characteristics such as high dimensionality, massive volume, and dynamic correlations. Traditional warning methods based on static thresholds or simple statistical models are insufficient for the early detection of subtle anomalies and the precise diagnosis of complex faults. To address this challenge, this paper proposes an intelligent operation and monitoring data warning and diagnosis framework based on a Large Language Model enhanced Transformer (LLM-Enhanced Anomaly Transformer, LEAT). This framework adopts a twostage structure: In the first stage, the cutting-edge Anomaly Transformer model is utilized to perform efficient unsupervised anomaly detection by leveraging its innovative “Association Discrepancy” metric to capture early deviations in system operation patterns. In the second stage, for the detected anomalous data segments, carefully designed Prompt Engineering techniques are used to encapsulate their contextual information and numerical dynamics, and a Large Language Model (LLM) is invoked for Zero-Shot inference to achieve interpretable Root Cause Analysis. To validate the effectiveness of the proposed framework, comprehensive simulation experiments were conducted on the public large-scale Server Machine Dataset (SMD). The results show that the LEAT framework achieves an F1-score of 0.930 in anomaly detection, which is on par with the current state-of-the-art. In the more critical task of root cause diagnosis, its Top-1 diagnosis accuracy (Precision@1) reaches 0.891, significantly outperforming baseline methods. This research confirms the great potential of the LEAT framework in enhancing the automation and intelligence of intelligent operations and monitoring.