A novel energy efficiency improvement framework based on data-driven learning and energy online decoupling for fuel cell hybrid buses

Abstract

Accurate scenarios prediction and effective energy management are crucial for the automobile industry to achieve electrification and reduce regional carbon emissions. However, the lack of data dimensions and the high complexity of energy coupling relationships constrain the practical application of the advanced strategy in electric vehicles. In this study, a real-time optimal framework is proposed based on data-driven learning prediction and energy online decoupling rules for fuel cell hybrid buses. To be specific, a deep neural network is employed to obtain the complete running conditions by using a piece of state curve as the input as well as the learning simples including original vehicle signals and corresponding environmental information. To improve the energy conversion efficiency, the mechanical - electrical - hydrogen energy is decoupled layer by layer based on energy decoupling rules, thus obtaining the control sequences through hierarchical optimization. Finally, the rotating hub tests are conducted to verify the accuracy and economic efficiency of the proposed framework. Compared with existing methods, the framework reduced the total cost in the test cycle to 42.2 USD/100 km with a decrease rate of 11.7 %, and the optimized carbon emissions to 5.15 kg/cycle with a decrease rate of by 34.1 %. Based on the annual operating data of all target vehicles, the proposed framework is predicted to reduce the carbon emissions by 25,247 kg per year in Beijing.