Reliable Federated Learning-Based Wireless Computing Power Scheduling for Efficient Edge Computing Networks

With the advent of the 6G and the rapid development of various AI technologies, the demand for computing power has increased significantly in edge computing networks. The wireless computing power network (WCPN), as an emerging technology, can connect various ubiquitous and heterogeneous devices and computing resources, to enable flexible and unified management and scheduling of computing power. However, the openness of WCPN makes it vulnerable to being attacked by malicious nodes, leading to severe consequences. Additionally, data islands and the strong demand for data privacy protection make traditional centralized machine learning methods difficult to implement. To address these issues, this paper proposes a trusted wireless computing power network scheduling scheme using federated learning. A reputation evaluation mechanism is proposed to assess the reputation of each computing node, ensuring the security and trustworthiness of the participating nodes. A computation resource optimization problem is formulated to minimize the system delay by comprehensively considering energy consumption, reputation, and computing power constraints. We further propose a reputation-based deep reinforcement learning (RDRL) method to solve the optimization problem. Experimental results show that our scheduling mechanism significantly reduces the total delay of WCPN system and enhances the accuracy of task execution.