Multi-Fidelity Near-Optimal on-Line Control of a Parallel Hybrid Electric Vehicle Powertrain

On-line optimal control is a crucial issue in the development of hybrid electric vehicles (HEVs). In this paper, optimal on-line control policies for a parallel HEV are firstly derived from off-line optimization. Then, two plant models with different grades of fidelity are considered for the on-line forward HEV simulation. An optimal calibration methodology is proposed to adapt the control policies previously extracted to the specific plant model. A multi-fidelity procedure can thus be established in adopting a low-fidelity plant model to extract a first estimation of optimal on-line control policies, while subsequently refining them through an high-fidelity plant model. Obtained results demonstrate the effectiveness of the proposed approach and measure the impact of the considered model fidelity level on the fuel consumption estimation, the computational time required and the specifically extracted control rules.