A Cellular Traffic Prediction Method with Energy Tracking Secure Federated Learning in 5G Scenarios

Predicting cellular traffic is crucial for intelligent design, resource optimization, and anomaly detection in fifth-generation (5G) and future networks, particularly in the realms of power grid management and energy distribution. Accurate traffic forecasts can enhance network performance, improve energy efficiency, and ensure the stable operation of power grids. However, traditional centralized prediction methods pose risks related to data security and privacy breaches, especially given the sensitivity of grid data. This study proposes a cellular traffic prediction method that combines federated proximal optimization aggregation algorithms with differential privacy techniques. The goal is to securely aggregate prediction model parameters across base stations while safeguarding the privacy and security of grid data. This approach allows us to provide power grid operators with accurate traffic forecasts, thereby optimizing energy distribution and demand response strategies while maintaining data privacy. We evaluated the proposed secure federated learning method using a real 5G cellular base station traffic dataset from Barcelona. The results demonstrate its advantages in terms of prediction accuracy and data privacy protection. Additionally, this study considers the environmental impact of federated learning, focusing on the trade-offs between security, accuracy, and energy consumption. By assessing the energy consumption and carbon dioxide emissions of the proposed method, this paper not only offers an energy-saving reference for distributed modeling of 5G base stations but also provides new perspectives for the green and sustainable development of power grids.