SOH estimation method for lithium-ion batteries under low temperature conditions with nonlinear correction

Abstract

The burgeoning growth of green energy in the transportation sector has resulted in increased expectations for battery longevity and safety. However, the capacity of lithium-ion batteries (LIBs) decreases with each successive charge and discharge cycle. And under harsh operating conditions, the capacity decay can exhibit strong nonlinearity. To enable effective battery management under such complex conditions, it is crucial to possess precise understanding of the state of health (SOH) of LIB. In this study, low-temperature aging experiments were designed to obtain capacity attenuation data of LIBs. Then the nonlinear component in capacity decay is identified and transformed into model nonlinear correction. Subsequently, we developed a long short-term memory (LSTM) neural network and extracted incremental capacity (IC) features based on experimentally collected data to train the model. The findings indicate that by training with only one cell aging dataset, the model can estimate the SOH for other samples. Furthermore, we introduced nonlinear correction features with an attention mechanism (AM) and analyzed the improvement of prediction accuracy through feature optimization and network adjustment to achieve SOH estimation with a maximum root mean square error (RMSE) of 0.92 %. This approach provides a potential solution for predicting SOH of LIBs at low temperature.