Linear parameter-varying modeling and identification of lithium-ion battery for control-oriented applications

Abstract Model-based approaches are popularly used for real-time management of lithium-ion batteries. However, building accurate models to characterize the inherent nonlinear dynamic properties of electrochemical processes is still a challenge for control-oriented applications. Instead of building a model independently from physical laws or experimental data, a linear parameter-varying (LPV) modeling method for lithium-ion batteries is proposed in this research. In the LPV modeling structure, the nonlinear relationships are considered to be dependent on the so-called scheduling variables. In combination with the subspace identification algorithm, a global method is proposed to identify the LPV battery model based on a single data set with constantly changing scheduling variables. The validity of this modeling approach is verified by an experimental study of lithium iron phosphate (LiFePO4) cells. Experimental results reveal that state-of-charge (SOC) and SOC change rate are key scheduling variables that determine the model nonlinearities, and the identified model provides good results in terms of prediction accuracy and robustness. Meanwhile, comparison results show that the global LPV modeling approach can describe the nonlinear battery dynamics adequately with first-order model, which makes it a promising method for control-oriented applications.