上海科技大学知识管理系统

This paper proposes a capacity trajectory prediction framework for the onboard batteries equipped in electric buses. The framework is a sequence-to-sequence (Seq2Seq) structure based on bi-directional long short-term memory (BiLSTM) neural networks. To develop and evaluate the framework, this paper utilizes the field data of 200 electric buses of two models (denoted as A and B) during the period of January 2019 to December 2019, which is collected by the National Big Data Alliance of New Energy Vehicles (NDANEV), Beijing, China. To more accurately estimate and label the battery capacity, a mileage correction method is developed based on the discharging segments of the electric bus field data. Compared to the Coulomb counting method typically adopted for capacity estimation, this method takes into account the effects of the mileage on the capacity. With tailored feature engineering, 9 features closely related to the capacity are selected for each of the two bus models. Evaluation results show that the maximum prediction errors for the 12 buses in the two test sets are 2.4941 Ah in terms of the mean absolute error (MAE) and 3.0153 Ah in terms of the root mean square error (RMSE), respectively. If they are normalized with respect to the rated battery capacity of 320 Ah estimated for the buses of model B, the maximum normalized prediction errors are 0.78% and 0.94%, respectively, which clearly demonstrate the effectiveness of the proposed framework. © 2025 Elsevier Ltd