Spatial–temporal transformer-based ecological car-following strategy for connected electric vehicles in dynamic environments

This study proposes a learning-based ecological car-following strategy for connected and autonomous vehicles (CAVs). Leveraging advanced vehicle-to-vehicle and vehicle-to-infrastructure technologies, CAVs on roads can receive updated traffic flow information and predict the speed profile of the preceding vehicle using a macro–micro fused spatial–temporal transformer. To handle the discrepancies between the predicted and actual speeds of the preceding vehicle and achieve less conservative results, a robust learning-in-the-loop model predictive control algorithm has been developed. Furthermore, to enable real-time computation for practical applications, the system models are transformed from the time domain to the spatial domain, integrating a fictitious control input design for system convexification. Finally, a comprehensive assessment of the proposed strategy is conducted through numerical simulations and on-road vehicle experiments. • Convex reformation of the car-following problem in spatial domain. • A spatial-temporal transformer is developed for preceding vehicle speed forecasting. • Discrepancies are handled using a novel adaptive tube-based MPC algorithm. • An on-road experiment is designed to verify the applicability of the strategy.