Dynamic Graph Structure Modeling for Anomaly Detection in Complex Cloud Platforms
- Authors
-
-
Guoze Feng
Syracuse University, Syracuse, USA
Author
-
- Abstract
-
To address the challenges of anomaly detection in complex cloud platforms, stemming from the continuous evolution of service dependencies, complex anomaly propagation paths, and high coupling of multiple operational states, this paper proposes a dynamic graph structure modeling method that integrates service topology evolution perception with deep representation learning from graph neural networks. This method first abstracts the complex cloud platform into a dynamic attribute graph at different time steps. By unifying the representation of service nodes, interaction relationships, and operational characteristics, a time-varying structural description framework for anomaly detection is constructed. Based on this, a topology evolution description mechanism is introduced to model service dependency reconstruction, relationship changes, and local structural perturbations at adjacent time steps, thereby enhancing the model's ability to perceive dynamic structural changes. Subsequently, a relationship graph aggregation strategy is used to perform deep propagation and semantic refinement of high-order dependency information between nodes, improving the identification of correlated and propagated anomalies. Furthermore, a temporal fusion mechanism is used to jointly model the current observation state and historical representation information, enhancing the model's ability to distinguish between persistent anomalies, gradual anomalies, and short-term fluctuations. Finally, the constructed method can effectively characterize and accurately identify anomaly risks in the dynamic environment of complex cloud platforms. Comparative experimental results show that the proposed method outperforms the other method in overall detection performance, verifying the effectiveness and applicability of the proposed dynamic graph structure modeling framework in anomaly detection tasks on complex cloud platforms.
- Downloads
- Published
- 2025-07-01
- Issue
- Vol. 5 No. 7 (2025)
- Section
- Articles