Privacy-preserving estimation of electric vehicle charging behavior: A federated learning approach based on differential privacy

With the popularity of connected electric vehicles, the openness and sharing of charging data between stakeholders allows a more accurate estimation of charging behavior, which is valuable for optimizing energy systems and facilitating travel convenience. However, to enable such an effective mechanism, the challenge of data security and privacy should be addressed. Federated learning in the vehicular network is appealing for utilizing individual vehicle data while preserving data privacy. We propose an improved local differential privacy-based federated learning approach for modeling charging session prediction problems while preserving user privacy against the threat from a honest-but-curious server. In this approach, all vehicles, within the coordination of a cloud server, collaboratively establish a global regression network through parameter exchange. Meanwhile, the servers may belong to third-party model owners and can be semi-honest when inferring private information on the collected model parameters. Hence, local differential privacy is adopted to perturb the parameters. Additionally, a combination of local and global models via elastic synchronization is proposed to improve the accuracy of the learned noisy global model. Through the test on a real data set, the results show the superiority of the proposed algorithm over traditional noisy federated learning methods. Furthermore, the practical value of the proposed method is validated with a real-world charging case. Such an accurate charging session prediction service for electric vehicle drivers facilitates charging and travel convenience in the green transportation world.