Adaptive Federated Learning for Digital Twin Driven Industrial Internet of Things
Qiang Song, Shiyu Lei, Wen Sun, Yan Zhang
Abstract
Industrial Internet of Things (IoT) enables distributed intelligent services varying with the complex industrial environment to achieve the benefits of Industry 4.0. In this paper, we consider a new architecture of digital twin empowered Industrial IoT, in which digital twins capture characteristics of industrial devices to assist federated learning tasks of industrial scenarios. A trust-based aggregation is proposed in federated learning to alleviate the effects of digital twins deviation and emphasize the contribution of high-performance clients. Based on Lyapunov dynamic deficit queue and deep reinforcement learning, we propose a federated learning framework that adaptively adjusts the aggregation frequency to improve the learning performance under resource constraints. Numerical results show that the proposed framework outperforms the benchmark in terms of learning accuracy, convergence, and energy saving.