A Spatial Data-Driven Vehicle Speed Prediction Framework for Energy Management of HEVs Using Multi-Horizon MPC with Non-uniform Sampling

Long-horizon economic optimization via model predictive control (MPC) could result in a near optimal solution for multi-timescale energy systems. However, there are multifaceted challenges associated with the implementation of MPC over relatively long horizons, including the large computational footprint and high sensitivity to look-ahead information. While hierarchical and multi-horizon MPCs have addressed the computational burden to some extent, there are more fundamental questions to answer, including the required accuracy, resolution, and length of look-ahead information, and how to obtain that information. In this paper, we explore the vehicle speed prediction for the integrated power and thermal management (iPTM) of hybrid electric vehicles (HEVs) via multi-horizon MPC (MH-MPC). To that end, firstly, we perform a sensitivity analysis to assess the impact of data sampling time and prediction horizon length on the iPTM performance. Next, we propose a new MH-MPC with non-uniform sampling to achieve a proper trade-off between minimizing energy and managing computational demand. Finally, patterns in vehicle speed for recurrent trips are explored in the spatial and time domains to inform speed predictions. By using real-world traffic data collected from a test vehicle, we demonstrate the benefits of using vehicle speed prediction in the spatial domain in combination with non-uniformly sampled MH-MPC, and provide insights into the impacts of uncertainties in the look-ahead information on the iPTM performance.