Towards temperature-dependent linear parameter-varying models for lithium-ion batteries using novel experimental design

This paper proposes a comprehensive approach to identifying temperature-dependent battery models using the input--output model representation in the linear parameter-varying (LPV) framework. The proposed model structure considers a simultaneous dependence of the model parameters on the battery state-of-charge (SOC), current magnitude, current direction and temperature using a suitable set of basis functions formulated using available physical and experimental knowledge. Additionally, a temperature profile design is proposed that can be used along with a current profile design to excite the relevant temperature-dependent battery dynamics during the identification experiments. Moreover, an algorithm to combine multiple identification experiments is presented so that the computational complexity of the regression problem is of the same order as that of a single experiment. Finally, several battery models with varying model order and basis-function complexity are identified for a 2.85-Ah NMC battery, which are subsequently validated using a test dataset resembling a real drive-cycle scenario under varying temperature conditions. The corresponding root-mean-squared error (RMSE) values for the model exhibiting the best voltage simulation performance are found to be 19.31, 11.93 and 6.95 mV for the `cold', `normal' and `hot' temperature conditions, respectively.