Flexible energy storage estimation for electric buses: A hybrid data-driven and physical model-driven approach

Abstract

The large-scale deployment of electric buses contributes to the development of low-carbon transportation systems and carbon neutrality strategies. Effectively predicting the available energy of electric buses and aggregating flexible energy storage plays a crucial role in the operation and scheduling of power grids. This paper proposes a hybrid-driven estimation method for flexible energy storage of electric buses. First, an adaptive graph convolutional network is introduced to predict the arrival times and passenger crowdedness levels for each bus line. This data-driven approach captures the spatiotemporal dynamics and various evolution patterns of different routes, enhancing the prediction accuracy. Subsequently, a physical model-driven approach based on vehicle dynamics equations and first-order thermodynamic equivalents is proposed to forecast the energy consumption of the trip. This model represents energy flow and potential physical processes, providing interpretable insights into the impact of various parameters on traction and air conditioning energy consumption. In addition, we design and conduct the battery test profile to investigate the degradation patterns of batteries under different temperatures. Finally, based on the battery available energy, we develop a power and energy boundary model for electric buses to characterize flexible charging loads. Furthermore, an approach is proposed to estimate the flexible energy storage of a large number of electric buses. Case studies show that the proposed graph convolutional network outperforms baseline models, achieving an MAE of 16.33 s in arrival times. The aggregated schedulable energy storage varies by as much as 27.65% under different temperature conditions.