A Model Parameter Identification Method for Battery Applications

Due to growing interest in hybrid and electric vehicles, the battery, being one of the critical components, is receiving a lot of attention from designers and researchers. Two battery-modeling approaches, though seemingly different, share the same mathematical challenge of robust non-linear curve-fitting. The two methods are battery equivalent circuit model and battery system level thermal modeling using the linear time-invariant (LTI) method. Both modeling approaches involve curve-fitting testing data or data from advanced models to identify four parameters in a circuit model consisting of two pairs of RC elements. Such curve-fitting is mathematically a non-linear least-squares (LS) problem. Standard methods like the Levenberg-Marquardt (LM) method can be used for non-linear curve-fitting, but the LM method is known to be sensitive to initial conditions. Due to the unique features of the two pairs of RC values in the model, the curve-fitting problem can be reformulated into a linear LS problem. Solution from the linear LS problem can then be used as an initial condition for the LM method for greater accuracy. Since the initial conditions from the linear LS problem are already close to the minimum, the sensitivity issue associated with the LM method is mitigated.