Efficient Anomaly Detection in Distributed Edge-Cloud Systems Using Lightweight Modeling

Abstract

This paper addresses the problem of anomaly detection in edge-cloud collaborative environments and proposes a lightweight model design method. The study first analyzes the challenges of anomaly detection under multi-tenant sharing and heterogeneous resource conditions, including high latency, bandwidth consumption, and privacy risks. To solve these issues, a detection framework combining edge computing with cloud collaboration is developed. At the edge side, input projection and feature encoding are performed, and temporal dependencies are modeled through a lightweight attention mechanism. The cloud side then aggregates and optimizes features from different nodes to achieve global anomaly recognition in distributed scenarios. The method incorporates edge memory caching and reconstruction mechanisms, enabling the model to reduce computation and storage costs while maintaining accuracy. To further enhance performance, the framework applies regularization and privacy constraints to improve robustness against noise and data heterogeneity in distributed environments. Experiments using real cloud workload trace datasets demonstrate that the proposed method outperforms comparison approaches in F1-Score, Precision, Recall, and ACC, confirming its effectiveness and stability in complex conditions. The findings enrich the theoretical system of lightweight detection methods in edge-cloud collaboration and provide feasible technical support for the secure operation of distributed intelligent systems.