Differential Privacy-Aware Generative Adversarial Network-Assisted Resource Scheduling for Green Multi-Mode Power IoT

Abstract

The low-carbon and efficient operation of smart parks requires high-precision and real-time energy management model training. Multi-mode power Internet of Things (PIoT) consisting of open radio access networks (O-RAN) and power line communications (PLC) can effectively improve the model training performance. However, the negative effects of network threats, such as model inversion attacks, cannot be neglected. To solve this problem, we propose a diFferential pRivacy-aware gEnErative aDversarial netwOrk-assisted resource scheduling algorithM (FREEDOM). Firstly, we integrate a differential privacy mechanism with the energy management model training process and the related system model. Then, a joint resource scheduling optimization problem is constructed, the goal of which is to minimize the weighted sum of the global loss function, total energy consumption, and differential privacy cost under the long-term differential privacy constraint. The original problem is converted based on virtual queue theory and addressed by the FREEDOM. FREEDOM leverages a deep Q-learning network (DQN) to learn the resource scheduling strategy via differential privacy awareness. It improves optimization and convergence performances with the assistance of generative adversarial network (GAN). Simulation results show that FREEDOM can achieve excellent performances of global loss function, total energy consumption, differential privacy cost, and privacy preservation.