Physics-Based Attack Detection for Traction Motor Drives in Electric Vehicles Using Random Forest

With the fast development of electric vehicles and vehicle onboard communication networks, modern electric vehicles suffer from potential threats from cyber networks. In order to secure vehicle safety and reliability, advanced attack detection techniques are in urgent need. In this paper, we propose a physics-based attack detection method using a random forest classifier. The key idea is to extract system features from the trustworthy and easy-to-get electric machine phase current signals, and use a random forest classifier to search a secure boundary to distinguish whether or not the powertrain system is under malicious cyber-attacks. The proposed method is tested and validated by simulation data generated from MATLAB Simulink. The results prove the feasibility of using electric machine phase current signals to represent multiple powertrain system features and accurately detect malicious attacks based on these extracted features.