Mobility-Aware Vehicle Selection Strategy for Federated Learning in the Internet of Vehicles

Federated Learning (FL) in the Internet of Vehicles (IoV) is a distributed machine learning technology that allows Intelligent Connected Vehicles (ICVs) to collaboratively train models without sharing raw data. In vehicular FL, the selection of participating nodes directly impacts the efficiency and accuracy of the model. Specifically, selecting appropriate ICVs ensures that participating nodes provide high-quality data while possessing sufficient computational resources and communication capabilities. While some existing studies have focused on optimizing resource allocation, improving data quality, and proposing ICV selection strategies, research on the comprehensive impact of ICV mobility on these factors remains limited. This paper proposes a dynamic optimization-based ICV selection algorithm aimed at effectively addressing the impact of mobility on FL performance in IoV. Our strategy combines factors such as ICVs' geographic location, speed, and data quality, selecting the optimal subset of ICVs within learning time constraints to ensure that participating ICVs can provide high-quality data and complete training tasks on time. Simulation results show that this method significantly improves the convergence speed and accuracy of FL models, validating the effectiveness of the proposed approach.