Privacy protection method for power customers based on federated learning model

With the rapid development of the power Internet of Things (IoT), protecting the privacy of electricity customers has become a critical challenge. To address the risks of privacy leakage in traditional centralized data training methods, this study proposes a privacy protection method for electricity customers based on a federated learning framework. By integrating federated learning with transfer learning techniques and gradient encryption mechanisms, the proposed method enables localized model training and global model aggregation, ensuring that user data remains "usable but invisible" while improving prediction performance. The proposed model achieves a accuracy of 89.85%, surpassing other models and highlighting the strengths of the federated learning framework in balancing user privacy protection and performance optimization.Simulations further verify the framework’s effectiveness in power IoT scenarios, enhancing model generalization while reducing the risk of privacy leakage.