Multiple measurement health factors extraction and transfer learning with convolutional-BiLSTM algorithm for state-of-health evaluation of energy storage batteries

State-of-health (SOH) is an important indicator for evaluating battery’s performance. However, most of the current data-driven SOH estimation models feature extraction is complex and only applicable to the same type of battery and the same operating conditions. To address this limitation, this paper proposes a multiple measurement health factor extraction method and a transfer learning-convolutional-bidirectional long short-term memory (TL-CNN-BiLSTM) algorithm for SOH evaluation of energy storage batteries. This feature extraction method directly uses the measured values of experimental data as health factors to characterize the degradation characteristics of batteries, which can simplify the calculations. The TL-CNN-BiLSTM algorithm introduces a transfer learning strategy, which learns the general knowledge of battery degradation and fine-tunes the model according to different situations, so that the trained model is suitable for SOH estimation of the different battery under different operating conditions. Using publicly available NASA and CALCE battery datasets for validation, the results show that the extracted multiple measurement features can be used for SOH estimation, and the proposed TL-CNN-BiLSTM algorithm can improve the accuracy of SOH estimation. The root mean square error (RMSE) and mean absolute error (MAE) of the transfer model results between different cells are less than 1%. In addition, the proposed algorithm also performs well in SOH evaluation across data domains.