State of Health Estimation Method for Lithium-Ion Batteries Based on BiLSTM-Transformer and Fusion Features

As the core of modern energy storage technology, lithium-ion batteries are already widely applied in mobile devices and renewable energy systems. Battery state of health (SOH) assessment is critical to the safe and stable operation of a lithium battery management system. To tackle the problem of inaccurate SOH estimation. In this paper, fusion features and a hybrid prediction model combining Bidirectional Long Short-Term Memory (BiLSTM) and transformer are proposed to effectively deal with the difficult problem of long-term sequence prediction and inaccurate SOH prediction. Using BiLSTM to capture changes in short time series, and then combines transformer coding-decoding layer and a feedforward neural network to effectively weigh important parts of the input sequence to predict battery health. First, the incremental capacity analysis curve and differential thermal voltammetry curve are obtained from the discharge data of the battery. Then, extracted health factors after Gaussian filtering. Finally, the health factors with high correlation are selected as the input of the prediction model by Pearson correlation analysis. The proposed algorithm is validated using NASA and Oxford open source datasets, the proposed hybrid model achieves MAE and RMSE values below 1.12×10 –2 , exhibiting high accuracy and robustness for battery SOH estimation, as demonstrated by experimental results.