Secure Vertical Federated Learning for Charging Station Recommendation in Connected Vehicles

In this paper, we propose a solution for secure and communication-efficient Vertical Federated Learning (VFL) applied to charging station recommendations in Intelligent Connected Vehicles (ICVs). Our methodology leverages a hybrid security strategy combining threshold partially homomorphic encryption (PHE) and additive secret sharing scheme (SSS) to ensure data privacy and security during model training. The threshold PHE enables encrypted operations, allowing a subset of participants to decrypt results, while SSS further protects gradients by splitting them into multiple shares across participants. To enhance communication efficiency, we employ asynchronous updates and gradient compression techniques, reducing synchronization and communication overhead. The approach is evaluated using three datasets and four baseline models, with results demonstrating strong recommendation accuracy, security, and reduced communication costs. This solution provides a secure and efficient VFL framework for ICV charging station recommendations.