Profiling on EV Chargers: Attack Surface Assessment and a Deep Learning-Based Approach

The electric vehicle (EV) charging communication system typically relies on common security measures to protect against cyber attacks. However, little attention has been given to the privacy of the communicated data of the chargers. This paper presents a new technique for profiling EVs using an arbitrary time window of measured data from EV chargers, allowing an attacker to identify an EV with a minimal amount of information. The attack surface is first explored, showing how a profiling attack can be performed under different threat models. This assessment is considered across all the components of the EV charging infrastructure communication system. A deep neural network-based architecture is then constructed out of multiple smaller models for best possible prediction. These models are then trained using datasets of real EV charging sessions. Results of randomized test cases are then used to evaluate the trained models showing a relatively high prediction accuracy. This study signifies the privacy threat in the existing charging infrastructure and proposes general recommendations to protect the drivers' privacy.