Detection and isolation of battery charging cyberattacks via Koopman operator

The cloud-controller for electric vehicle supply equipment (EVSE) relies heavily on receiving accurate sensor measurements from the battery of the electric vehicle (EV) and sending precise charging actuation to achieve optimal charging while maintaining grid stability. However, an adversary can corrupt these signals as they are transmitted via a communication network. This motivates our research to detect cyberattacks on both the actuation signal and the sensor measurements, as well as isolate or distinguish between them. While the problem of attack isolation based on system model and measurement is generally ill-posed, it is crucial for dispatching appropriate remedial measures following an attack detection. Thus, we propose a model-free Koopman operator-based cyberattack detection and isolation algorithm via residual generation. Our proposed algorithm is trained online using rate-limited voltage, temperature, and charging current data from high-fidelity battery charging experiments in Python Battery Mathematical Modeling (PyBaMM) and the ‘liionpack’ package. In our extensive case studies of prevalent cyberattacks, the algorithm consistently generates real-time detection-isolation flags without any prior model knowledge or historical data of the system. The algorithm is easily generalizable and can be seamlessly applied to various levels of data granularity (cell or pack-level), cell chemistries, and pack configurations, without requiring significant modifications.