Efficient and flexible management for industrial Internet of Things: A federated learning approach

Abstract In this paper, we devise an efficient and flexible management for mobile edge computing (MEC)-aided industrial Internet of Things (IIoT), from a federated learning approach. In the considered IIoT networks, all devices have some computational tasks to be computed with the help of some computational access points (CAPs). Although the performance of the IIoT networks can be optimized by using the resource allocation based on some centralized schemes, such solution is neither efficient nor flexible. To address this issue, we use a deep reinforcement learning (DRL) algorithm based federated learning algorithm to adjust three parameters: the task offloading ratio, bandwidth allocation ratio and transmit power. The optimization can minimize the normalized system cost, while reduce the communication cost in the optimization process. Moreover, simulation results are demonstrated to verify that the proposed federated framework can achieve an efficient and flexible management for the IIoT networks.