Federated Learning With Local Differential Privacy for Autonomous Electronic Vehicles: Enhancing Security and Performance

The rapid advancement of autonomous electronic vehicles (AEVs) has revolutionized the transportation industry, offering enhanced safety, efficiency, and convenience. However, the collaborative nature of AEVs, which involves exchanging data and models among vehicles and central servers, raises significant concerns regarding privacy and security. Federated learning (FL) has emerged as a promising approach to enable collaborative model training while preserving data confidentiality. Nevertheless, the vulnerability of FL to inference attacks and the potential presence of malicious participants pose substantial challenges. To address these issues, we propose a novel FL framework with local differential privacy (LDP-FL) tailored specifically for AEVs. Our framework incorporates LDP mechanism to protect individual AEV data by perturbing the model updates before sharing them with the central server. Moreover, we introduce a performance loss constraint mechanism to minimize the impact of the privacy-preserving noise on the model’s performance. The LDP-FL algorithm ensures secure and efficient collaborative learning among AEVs while preserving data privacy. Extensive experiments using the Udacity dataset demonstrate the superiority of LDP-FL compared to state-of-the-art baseline methods for global accuracy, performance loss, running time, energy consumption, communication latency, and robustness against malicious attacks. The LDP-FL framework achieves a remarkable balance between privacy, security, and performance.