Novel Multi-Physics Informed Neural Network based Coordinated Control Strategy for Mode Transitions of PHEV Considering Torsional Oscillation

This paper proposes a novel multi physics informed neural network-based model predictive control (MPINN-MPC) strategy to address torsional vibrations and torque fluctuations during electric mode to hybrid mode (EM-to-HM) transitions in plug-in hybrid electric vehicles (PHEVs). The powertrain’s mode transition process (MTP) is divided into five distinct stages, where physics-informed neural networks (PINNs) are introduced to capture transient nonlinear dynamics in each stage, ultimately constructing an MPINN model that accurately represents the transient MTP behavior. By embedding physical laws into neural network training, the MPINN framework combines the interpretability of physical equations with the adaptability of data-driven learning, enabling precise prediction of system states under uncertainties. Integrated with the MPC architecture, the proposed strategy accurately coordinates the clutch oil pressure and motor torque during the MTP. Experimental results demonstrate that, compared to Non-control and MPC based on hybrid model (H-MPC) strategies, MPINN-MPC achieves greater reduction of vehicle jerk (by 92.4% and 49.3%, respectively) and clutch slip energy loss (by 6.9% and 1.6%, respectively). This method provides a robust solution for improving ride comfort and energy efficiency during complex powertrain mode transitions.