Edge Hierarchical Asynchronous Federated Learning Model for the Industrial Internet of Things

In the context of the Industrial Internet of Things (IIoT), a substantial volume of data is continually transferred, and the data types usually involve a large number of devices and sensors, which leads to diverse and Non-IID. In addressing the limitations of the conventional federated learning model, which fails to account for the performance disparities among devices and encounters challenges in ensuring privacy in heterogeneous IIoT networks, a novel edge hierarchical asynchronous federated learning model, termed T-FedHA, has been proposed. The Edge Proxy Head Node (EPHN) is proposed as an intermediate layer, and the asynchronous process is improved and optimized accordingly, with the result that it can better adapt to the hierarchical architecture of the Edge Proxy Head Node as the middle layer. In addition, the proposed model is evaluated based on two standard datasets, Fashion-MNIST and CIFAR10, as experimental datasets. The results demonstrate that the proposed approach enhances model accuracy and convergence speed and exhibits superior fault tolerance and stability compared to numerous existing methods.