A Cloud Computing Anomaly Detection Method Integrating Diffusion Graph Neural Networks with Long- and Short-Term Temporal Dependency Modeling

Abstract

With the continuous expansion of cloud platform scale and the widespread application of microservice architecture, container orchestration, and dynamic resource scheduling mechanisms, system operation states exhibit greater complexity, dynamism, and correlation. Traditional anomaly detection methods relying on single indicator fluctuations or local time window analysis are no longer sufficient to meet the accurate identification needs in complex cloud environments. Addressing the challenges of concealed structural propagation, intertwined temporal dependencies, and uneven influence of key nodes in cloud computing scenarios, this paper proposes a cloud computing anomaly detection method that integrates a diffusion graph neural network and a long short-term temporal dependency capture mechanism. This method first models service instances and their dependencies in the cloud environment as a dynamic graph structure and constructs a unified input representation by combining monitoring indicator sequences to enhance the joint characterization of system topology interactions and state evolution. Subsequently, it extracts high-order service association information through the diffusion graph modeling mechanism, mines potential patterns of anomaly propagation along dependency links, and utilizes the long short-term temporal dependency capture mechanism to learn the dynamic changes in system state over continuous time, thereby improving the ability to identify sudden and gradual anomalies. Based on this, an adaptive attention fusion strategy is introduced to selectively enhance key information from different service nodes and at different times, further improving the discriminative power of anomaly representation and overall detection stability. The results show that the proposed method can effectively integrate relational structure information and temporal evolution information in cloud platforms, demonstrating strong comprehensive recognition capabilities in cloud computing anomaly detection tasks, and providing an effective method for intelligent monitoring and operational status perception in complex distributed systems.