A Novel Learning-Based Model Predictive Control Strategy for Plug-In Hybrid Electric Vehicle

The multisource electromechanical coupling renders the energy management of plug-in hybrid electric vehicles (PHEVs) highly nonlinear and complex. Furthermore, the complicated nonlinear management process highly depends on knowledge of driving conditions and hinders the control strategies efficiently applied instantaneously, leading to massive challenges in energy-saving improvement of PHEVs. To address these issues, a novel learning-based model predictive control (LMPC) strategy is developed for a serial-parallel PHEV with the reinforced optimal control effect in real-time applications. Rather than employing the velocity-prediction-based MPC methods favored in the literature, an original reference-tracking-based MPC solution is proposed with strong instant application capacity. To guarantee the optimal control effect, an online learning process is implemented in MPC via the Gaussian process (GP) model to address the uncertainties during state estimation. The tracking reference in the LMPC-based control problem in PHEV is achieved by a microscopic traffic flow analysis (MTFA) method. The simulation results validate that the proposed method can optimally manage energy flow within vehicle power sources in real time, highlighting its anticipated preferable performance.

Automated summary

In this study, a novel learning-based model predictive control strategy is developed for serial-parallel plug-in hybrid electric vehicles (PHEVs). This method addresses uncertainties in state estimation through a Gaussian process model, achieving optimized management of energy flow with strong real-time applicability.