Adaptive recursive least squares for online parameter identification in Lithium-ion battery models

Abstract

Accurate online parameter identification is fundamental to the reliable operation of battery management systems (BMS) in electric vehicles. This paper proposes an Adaptive Forgetting Factor Recursive Least Squares (AFFRLS) algorithm for real-time identification of equivalent circuit model (ECM) parameters in lithium-ion batteries. A Thevenin-type ECM is adopted and discretized using the bilinear (Tustin) transformation, yielding a linear regression structure amenable to recursive estimation. The identified parameters ohmic resistance R₀, polarization resistance R_P, and polarization capacitance C_P are updated to compute the estimated battery terminal voltage U_est. Three algorithms standard Recursive Least Squares (RLS), Forgetting Factor RLS (FFRLS), and the proposed AFFRLS  are comparatively evaluated under the highly dynamic Urban Dynamometer Driving Schedule (UDDS) current profile applied to a commercial LG 18650HG2 NMC lithium-ion cell. Simulation results show that AFFRLS achieves the lowest maximum absolute voltage error of 2.68%, compared with 4.09% for RLS and 3.45% for FFRLS, while maintaining an RMSE of 0.13%, which is comparable to that of FFRLS at 0.12%. These results indicate that the adaptive forgetting factor mechanism improves worst-case tracking performance during abrupt current transients while preserving competitive overall estimation accuracy. With an explicit saturation mechanism to constrain the forgetting factor within [0.95, 1.0], AFFRLS provides a promising and computationally lightweight candidate for online ECM parameter identification in embedded BMS applications.