Machine Learning-Based Physical Layer Security for 5G/6G-Enabled Electric Vehicle Charging Network

The rapid deployment of electric vehicle (EV) charging infrastructure, coupled with the integration of 5G/6G and Internet of Vehicles (IoV) technologies, has transformed charging stations into cyber–physical systems that rely on wireless communication for authentication, control, and grid coordination. While existing security standards such as ISO 15118 provide cryptographic protection at upper layers, they are insufficient to address physical-layer threats inherent to wireless connectivity. In particular, wireless active eavesdropping attacks can corrupt channel estimation during the authentication phase, enabling impersonation, unauthorized charging, and disruption of grid operations. This paper proposes a machine learning-based physical layer security (PLS) framework for detecting active eavesdropping attacks in 5G/6G-enabled EV charging systems. By modeling malicious EVs as pilot-spoofing attackers, three discriminative features, namely mean power, power ratio, and angle-based feature, are extracted from received pilot signals at the charging station. Three classifiers are evaluated: single-class support vector machine (SC-SVM), Random Forest (RF), and DNN. Simulation results demonstrate that the SC-SVM maintains a stable accuracy between 94% and 96% across all attacker power levels, while RF and DNN significantly outperform it under stronger attack conditions. Specifically, under strong attacker conditions, RF achieves an accuracy of 99.9%, and DNN reaches 99.8%, both exceeding 99% detection accuracy. By preventing pilot-spoofing-based impersonation during authentication, the proposed framework enhances charging availability, billing integrity, and grid-aware scheduling in intelligent EV charging infrastructure.