Traffic prediction-based long-term energy management approach incorporating engine transient control for HEVs

The advancements in intelligent transportation systems and learning-enhanced control methods have opened new possibilities for exploring the energy-saving potential of connected and autonomous hybrid electric vehicles (CAHEVs). In this paper, we propose a multi-scale optimal energy management (EM) framework for CAHEVs, centered around addressing the EM challenge over long preceding horizons. The design framework is divided into three phases based on the relevant time scales. In the long-term phase, the motion of the CAHEV is planned to achieve efficient and safe driving, based on traffic predictions from a Gaussian process regression predictor. Trained on past data covering a much longer time period, the predictor outputs predictions with high confidence for the preceding horizon. For the short-term phase, we develop a model-free integral reinforcement learning-enhanced engine control approach. This method addresses the uncertainty in the engine’s thermal dynamics and improves its transient performance especially during start-up process. It enables the engine reaching desired operating points smoothly and quickly. Finally, the mid-term phase tackles the EM problem. Taking the planned driving cycle as constraints, an optimal control problem is formulated. An engine on–off strategy is implemented to solve it, whose performance is guaranteed by the developed engine control approach. In summary, the proposed framework offers a practical and comprehensive EM solution for CAHEVs over long horizons. Simulation results using a high-fidelity simulator based on real-world driving data demonstrate the effectiveness of the developed framework.