Improving plug-in hybrid electric vehicles (PHEVs) fuel economy requires a proper energy management strategy (EMS). Efforts to enhance the energy-saving performance of predictive EMSs have concentrated on advanced speed prediction methods. However, the impact of battery models on the predictive EMS hasn't been investigated. This paper aims to fill the research gap by suggesting a model predictive control-based (MPC) EMS framework that uses a hybrid speed predictor. Firstly, six control-oriented battery electro-thermal-aging models with various terminal voltage and temperature simulation accuracies have been developed and verified. Secondly, it also examines the effects of different battery models on MPC-based EMS through quantitative analysis, including battery dynamics, calculation time, and resultant operating costs, which leads to the suggestions of model selection for the design of the EMS under low- and room-temperature driving scenarios. Finally, a novel hybrid speed prediction model is proposed, where the historical speed is decomposed into strongly periodic intrinsic mode functions (IMFs) by variational mode decomposition (VMD), and backpropagation (BP) neural network is utilized to learn the feature parameter and mapping relationship of each IMF. In addition, a case study is conducted by applying the proposed speed prediction model in an MPC-based EMS method. The simulation results highlight that the proposed hybrid speed prediction model can achieve preferable speed prediction. The operating cost errors (compared with the MPC-based EMS with 100% speed prediction) are reduced to 0.4% and 0.98% at −20 °C and 25 °C driving scenarios, respectively.