Nowadays, transportation relies heavily on vehicular cyber-physical systems (VCPS), which improve intelligent transportation systems (ITS) with advancements like real-time traffic control and self-driving vehicles. Because the data that these devices handle is sensitive, they not only make it possible for automobiles, roadside units (RSUs), and base stations to connect seamlessly, but they also present serious security issues. Intrusion can result in dangers to safety of the public, monetary losses, and a decline in confidence in these vital services. This paper presents a novel federated learning design intended to improve data security in VCPS in order to overcome these issues. Federated learning guarantees the privacy of raw data by enabling decentralised model training within individual vehicles or RSUs. In order to additionally protect privacy when aggregating local models into a global one, the framework includes the Laplace method to add noise into model updates. RSUs, vehicles, and a centralised server collaborate in the secure framework to stop leaks of information that occur during communication and model training. The suggested method beats conventional cryptography techniques when tested using the CICIDS2017 dataset, preserving significant levels of confidentiality and safety without sacrificing computing speed or accuracy of the model. Developing such sophisticated security measures will be essential to maintaining the integrity and dependability of transportation systems as VCPS develops, which will eventually result in improved safety and efficiency in transportation.
Babbar et al. (Tue,) studied this question.