To satisfy the ever growing demand for energy efficiency and pollution reduction, lithium ion batteries have attracted widespread attention as a mainstream option for large-scale energy storage in electronics and electric vehicles. However, the voltage hysteresis, widely observed during the battery charging and discharging, present a key challenge for further development and application of these batteries. Despite much effort have been made from the manufacturing perspective to mitigate the hysteresis, it can not be eliminated. Therefore, an accurate battery modeling method considering the hysteresis is of great importance for further application.

Existing methods, such as the Plett single state model and the Prandtl-Ishlinskii model, employ differential equations or operators for OCV hysteresis modeling. Although both are capable of simulating hysteresis under relatively ideal conditions, their generalization ability under more realistic scenarios is limited.

In this poster, a experimental scheme is designed to quantify OCV hysteresis, considering its dependency on SOC path and temperature for lithium iron phosphate/graphite batteries. Based on the experimental data, an accurate and robust hysteresis model utilizing a stacked LSTM network is proposed for the first time to simulate OCV hysteresis with high precision. The most significant advantage of our proposed model over existing hysteresis models lies in its robust generalization ability across different temperatures. By including these two factors as model inputs, the proposed model maintains its accuracy even when deployed under varied operating conditions. The modeling results reveal a RMSE of 2.42 mV during interpolation and 2.61 mV during extrapolation across different temperatures, which outperformed the Plett single state model (4.66 mV) and the Prandtl-Ishlinskii model (4.47 mV).