A Privacy-Preserving Prediction Model for Individualized Electric Vehicle Energy Consumption

Electric vehicles (EVs) have been gaining popularity in recent years but range anxiety of drivers and the ability to predict the energy consumption of EVs remains an important problem. While machine learning offers promising solutions for energy consumption prediction, it also introduces privacy challenges, especially when handling sensitive user data. As data-driven models become ubiquitous, ensuring the privacy and security of user information is paramount. This paper not only presents an innovative approach for EV energy prediction but also emphasizes the importance of privacy considerations in machine learning applications. We use a system that integrates key parameters such as ambient temperature, road gradient, and vehicle load to simulate real-world EV usage. By utilizing an innovative transformation layer that enables minimal low-level feature sharing and maintains maximum independence between groups, the system is designed to produce multiple simultaneous predictions for individual EVs while protecting privacy. Empirical results validate the system's ability to concurrently generate accurate predictions, outperforming conventional single-output models. Additionally, it provides a granular accuracy analysis across diverse EV models. We advocate for a balanced approach, harnessing data's potential while upholding stringent privacy standards and our experimental results show that the proposed model is robust against various attacks that seek to compromise user privacy.