SOH prediction of lithium-ion batteries using a hybrid model approach integrating single particle model and neural networks

Abstract

The prediction of battery state of health (SOH) plays a vital role in battery management systems. A fusion model framework was proposed by integrating an improved single-particle model (SPM) with data-driven deep learning algorithms to enhance predictive accuracy and further elucidate the intrinsic mechanisms of battery aging. First, seven electrochemical features were extracted by the improved SPM, which exhibits a significant reduction in computational complexity compared to conventional electrochemical models. The validity of the extracted features was further verified through the utilization of differential voltage analysis (DVA). Second, a hybrid model was constructed which combines temporal convolutional network (TCN) and bidirectional long short-term memory network (BiLSTM). The effectiveness and superiority of the proposed model was demonstrated, with the full electrochemical features, on Oxford University dataset. Finally, experimental measurements were conducted on five different batteries with two different electrode materials combinations to further study SOH estimation across battery types. To address the forecasting challenges arising from data scarcity for a new type of battery, transfer learning was introduced. The results highlight the potential of this fusion framework to achieve more efficient and accurate SOH prediction.