State-of-health estimation for lithium-ion batteries under complex charging conditions based on SDE-BiLSTM model

Accurately forecasting the state of health of lithium-ion batteries is crucial to guarantee their safe utilization. Under complex cycling conditions, conventional models struggle to produce sufficiently accurate results, while consumers are unable to evaluate the results' dependability. To assess the battery health status on an interval basis, this study integrated a stochastic differential equation (SDE) network with a bi-directional long short-term memory (BiLSTM) network, which could improve the accuracy of conventional models. Based on an open-source dataset, the state of health of 12 datasets with different charging strategies of batteries was estimated through conventional models and the interval prediction model, respectively. The root mean square error of state-ofhealth estimation was less than 0.83 %. Furthermore, the lack of observed battery data under practical scenarios poses challenges in establishing an accurate prediction network. Therefore, part of the new dataset was used to represent the data scarcity, and the state of health was estimated by a transfer learning (TL) method. The estimation results of the state of health of the BiLSTM-SDE-TL network agree well with the actual experimental data from the dataset. The comprehensive and reliable lithium-ion batteries' health information was obtained by the BiLSTM-SDE-TL network with simple training process despite a lack of observed data, which has potential for state-of-health prediction in practical complex application scenarios.