Anomaly detection of adversarial cyber attacks on electric vehicle charging stations

Abstract

The electrification of the transportation sector involves the widespread adoption of electric vehicles (EVs), to achieve global decarbonization. However, the increasing deployment of EV charging infrastructures (EVCI) introduces cybersecurity challenges, particularly concerning the different vulnerabilities associated with them, leading to cyberattacks. Charging ports are the crucial vulnerable points in the EVCI, which are connecting points between the EVs and EVCI. Intruders pose potential risks to the security, reliability, and functionality of the EVCI, by spoofing the data through these charging ports leading to anomalies in the data. This paper proposes an effective approach in detecting anomalies in the current magnitude of charging ports. An EVCI system is simulated in the MATLAB/SIMULINK environment for various scenarios of data generation. A Long Short Term Memory (LSTM) based autencoder model is used for predicting the charging port current magnitudes that capture the temporal dependencies in the sequential EVCI data. For generating the abnormalities in the data, the Fast-Gradient Sign Method (FGSM) is used, through which adversarial inputs are obtained, and these adversarial inputs are fed to the proposed LSTM autoencoder to obtain the anomalous data. To detect anomalies, the distributions of the sliding windows of the predicted and the observed charging port current magnitudes are compared through Kolmogorov–Smirnov (KS) test. The results demonstrate the model’s robust performance and predictive capabilities in forecasting the current magnitudes and identifying anomalies in them with an accuracy of 98.5%, enhancing the security and reliability of EVCI.