Robust Predictive Energy Management of Connected Power-split Hybrid Electric Vehicles Using Dynamic Traffic Data

This research focuses on the predictive energy management of connected human-driven hybrid electric vehicles (HEV) to improve their fuel efficiency while robustly satisfying system constraints. We propose a hierarchical control framework that effectively exploits long-term and short-term decision-making benefits by integrating real-time traffic data into the energy management strategy. A pseudospectral optimal controller with discounted cost is utilized at the high-level to find an approximate optimal solution for the entire driving cycle. At the low-level, a Long Short-Term Memory neural network is developed for higher quality driving cycle (velocity) predictions over the low-level's short horizons. Tube-based model predictive controller is then used at the low-level to ensure constraint satisfaction in the presence of driving cycle prediction errors. Simulation results over real-world driving cycles show an improvement in fuel economy for the proposed controller that is real-time applicable and robust to the driving cycle's uncertainty.