Abstract
In the Network Function Virtualization (NFV) environment, dynamic network traffic prediction is crucial for efficient resource orchestration and service chain optimization. When network service providers and resource providers collaborate to deliver services, traditional centralized prediction models face risks such as single points of failure, data privacy leaks across merchants, and malicious node attacks. To address these issues, we propose a blockchain based decentralized federated learning method for network traffic prediction, incorporating a node incentive mechanism. By dynamically adjusting node roles (e.g., "Aggregator," "Worker," " Residual node," and "Evaluator") based on prediction accuracy and network migration performance, the framework effectively mitigates the impact of malicious nodes. Additionally, the decentralized federated learning architecture eliminates reliance on central servers, enhancing collaborative training efficiency among multiple service providers while ensuring data privacy. Simulation experiments conducted on an NFV platform using real-world traffic datasets demonstrate that the proposed method improves prediction accuracy, reduces VNF migration overhead, and maintains prediction stability even in scenarios with a high proportion of malicious nodes.
All Related Versions

This publication has 23 references indexed in Scilit: