基于大语言模型知识蒸馏的微服务系统根因定位方法
By employing knowledge distillation of large language models and training graph neural networks with multimodal data, the problem of balancing accuracy and efficiency in microservice fault root cause localization is solved, achieving efficient and accurate fault localization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ANHUI UNIV
- Filing Date
- 2026-04-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing methods for locating the root cause of microservice failures based on graph neural networks struggle to balance accuracy and efficiency. Furthermore, large language models suffer from high inference costs and significant latency in real-time fault diagnosis, making them difficult to apply directly.
We employ a knowledge distillation approach based on a large language model. By constructing a knowledge distillation architecture for student and teacher models and training it with multimodal data, we utilize graph neural networks for fault root cause localization, including a multimodal graph encoder and a root cause locator. We adopt progressive training and hard sample selection strategies to reduce training costs and improve accuracy.
While ensuring lightweight models and fast inference speed, the accuracy and efficiency of microservice fault root cause localization have been improved, achieving high-precision and efficient automated root cause localization.
Smart Images
Figure CN122044942B_ABST