Cyberphysical Fault Detection for Electric Vehicle Platoons using Graph Convolution Network

Smart city infrastructure requires increasing deployment of electric vehicle supply equipment (EVSE) to match projected electric vehicle (EV) adoption. However, connected EV platoons pose physical and cyber-risks to charging stations—and by extension the smart grid—by the nature of their size and network surface. This work tackles the problem prophylactically by proposing a health-monitoring technique for EV platoons to identify vehicles that have a battery fault or have been exposed to cyberattacks prior to charging. Our method uses measurements acquired during vehicle operation and leverages the underlying communication structure of the platoon by using a spatiotempo-ral graph neural network (GNN). The GNN utilizes alternating layers of spatial feature extraction using Chebyshev convolution and time-series feature extraction using the attention mechanism. We compare our results with those obtained using learning-based methods that do not exploit the graph structure of data and demonstrate that our spatiotemporal GNN outperforms the baseline models after training on cyberphysical fault scenarios. Additionally, we discuss our future work into the advantages offered by GNNs for platoons.