The Internet of Vehicles (IoV) has emerged as a critical paradigm in intelligent transportation systems, generating vast amounts of data that can be leveraged for various applications. However, the distributed nature of vehicular networks, privacy concerns, and the non-IID (non-independent and identically distributed) characteristics of the data pose significant challenges to traditional machine learning approaches. This paper proposes a novel Hierarchical Federated Learning (HFL) architecture for IoV networks, introducing an intermediate tier of regional cloud servers between vehicles and the central server. Our approach addresses key challenges in communication efficiency, data heterogeneity, and model performance. Experimental results demonstrate that HFL achieves up to a 30% reduction in communication overhead and a 15% improvement in model accuracy compared to conventional federated learning methods in IoV scenarios.