Online accurate voltage prediction with sparse data for the whole life cycle of Lithium-ion batteries in electric vehicles

Big data platforms for electric vehicles (EVs) provide data and network advantages for the online battery management system (BMS) algorithms. However, storage costs lead to platform degradation of data quality, which increases the error of online BMS algorithms. Voltage is the raw data for many BMS algorithms, and improving its accuracy and prediction for use in EVs big data platforms has application prospects. In this paper, a high-precision voltage prediction method for the whole life cycle of batteries is proposed. The self-attention network can accurately predict voltage of a sparse dataset by pre-extracting features through transfer learning. A new loss function is proposed to address the problem of overfitting and distortion caused by using sparse data to train the model. Moreover, a training dataset update strategy is proposed to adjust the model according to internal and external factors of battery. The superiority, stability, and robustness of the proposed method are verified by experimental datasets and real-world datasets. The mean-absolute-percentage-error (MAPE) of predicted voltage is less than 0.5%, and it is 1–5% lower than traditional training methods. The proposed method has the advantage of high accuracy, low data requirements, and low computational amount. This study provides the foundation for implementing whole life cycle BMS algorithms.