An energy management strategy for plug-in hybrid electric vehicles based on deep learning and improved model predictive control

Abstract

In order to reasonably allocate the power needs and manage the energy of the plug-in hybrid electric vehicle (PHEV) more efficiently, an energy management strategy (EMS) based on deep learning and improved model predictive control was proposed. Firstly, the vehicle energy flow test was carried out for the PHEV, and the multi-physics (mechanical-electrical-thermal-hydraulic) model was constructed and validated. Secondly, six prediction models were built based on different algorithms and the effects were compared and analyzed in detail. Finally, a long short-term memory based improved model predictive control algorithm (LSTM-IMPC) was developed, and the effects of three EMSs based on the charge-depleting charge-sustaining rule (CD-CS), dynamic programming (DP) and LSTM-IMPC were investigated under Worldwide Light-duty Test Cycle (WLTC), New European Driving Cycle (NEDC) and real driving cycle (RDC). The results show that the fuel-saving rates of the LSTM-IMPC-based EMS under these three cycles are respectively 3.81%, 5.6% and 18.71% compared with the CD-CS-based EMS, which prove the good fuel-saving performance and strong robustness of the proposed EMS. The fuel-saving rates of the LSTM-IMPC-based EMS are close to the DP-based EMS, which are the global optimal under these three cycles.