Real-Time Multi-Ensemble Learning-Based Power Prediction for Energy Management of Dual-Source Electric Vehicles

Electric vehicles (EVs) rely on batteries as their main energy source, so it is important to take steps to reduce their stress to extend their life and maintain their performance. Combining supercapacitors with batteries can be a promising way to reduce battery stress, while also improving global EV performance. This paper proposes a novel multi-ensemble learning (MEL)-based method for designing energy management strategies (called MEL-EMSs) for dual-source EVs (DSEVs), i.e., EVs using a battery and a supercapacitor. A designed MEL-EMS determines in real-time a performant power distribution. Such power distribution is determined solely based on the history of the speed and traction power, without requiring any prior knowledge of the complete driving cycle. MEL-EMS is evaluated at two levels, to assess power distribution and energy consumption respectively. The real-time simulation is carried out under unknown driving cycles based on a validated numerical EV model. Our results demonstrate that MEL-EMS outperforms state-of-the-art methods by enhancing energy efficiency, reducing battery stress with a 4.8% reduction in battery root-mean-square current and a 41.7% decrease in standard deviation of battery current, while ensuring accurate and low-cost real-time power distribution.