Secure Parameter Aggregation and Distributed Anomaly Detection in Cross-Cloud Data Center Networks

Abstract

This study proposes an intelligent method based on federated learning to address the problem of anomaly detection in distributed cross-cloud data center environments. The research first analyzes the complexity and risks in multi-tenant shared computing scenarios and points out the limitations of traditional centralized detection methods, including high bandwidth consumption, privacy leakage risks, and poor adaptability to heterogeneous data in large-scale distributed architectures. To solve these issues, cross-cloud data centers are modeled as a collaborative system composed of multiple autonomous nodes. Each node performs local data preprocessing and feature modeling, while secure parameter sharing and global aggregation are achieved through a federated learning framework. In method design, the framework integrates multi-layer representation encoding with local optimization, enabling the model to capture complex cross-tenant interactions without exposing raw data. A global weighted aggregation strategy further improves robustness and generalization under heterogeneous data distributions. The anomaly detection task is formulated as a joint optimization of latent feature representation and reconstruction error, and precise anomaly identification is achieved through global parameter updates. Overall analysis shows that the proposed method balances privacy protection and detection accuracy, alleviates communication burdens in cross-data center collaboration, and provides reliable support for the secure operation of critical systems across multiple industries.