Research on wireless traffic prediction based on federated learning and differential privacy
Xingzhen Gao, Yuhong Zhao · 2024
Wireless traffic prediction in wireless communication networks holds significant application value for optimizing resource allocation. However, precise prediction faces numerous challenges due to traffic complexity, diversity, and dynamics. Complex spatiotemporal dependencies exist between regional base stations, requiring accurate capture. Traditional centralized methods necessitate centralized storage and processing of substantial user data, posing high privacy risks and potential leaks. This paper proposes a wireless traffic prediction model combining federated learning and differential privacy to capture fused spatiotemporal features of base stations and enhance predictive performance. Concurrently, existing federated learning systems have been proven to harbor potential threats during the training phase, endangering data privacy. This paper tackles privacy threats by adaptively pruning to safeguard data privacy, achieving privacy protection. Experimental results demonstrate that this model outperforms mainstream algorithms in prediction accuracy and privacy protection, showcasing broad application prospects.