HPPC-Based ECM Parameter Optimisation of Lithium-ion Battery: A Comparative Analysis of Non-Linear Least Squares Methods

This paper investigates the optimization of lithium ion (Li-ion) battery internal parameters for an Equivalent Circuit Model (ECM), a critical factor for accurate State of Charge (SoC) and State of Health (SoH) estimation. Four nonlinear least squares (NLLS) methods, including Gauss-Newton (GN), Levenberg-Marquardt (LM), Trust-Region Reflective (TRR), and Dogbox (DB), were used to optimize the internal parameters for first-order RC(1RC) and second order RC (2RC) ECMs. The parameters are then used to model the cycling data to produce evaluation metrics. The optimized parameter for each NLLS is compared within itself and with the non-optimized parameter to evaluate the performance. Key evaluation metrics such as computational time, iteration count, residual analysis, RMSE, R2 score, and tracking error percentage are used to compare algorithm performance using real-world driving cycle data. The results reveal significant improvement in the parameter estimation using optimisation techniques. It also reveal that within the NLLS group there is significant variations in the performance of the algorithm in different models. For the 1RC model, the GN method performs the best with a composite score of 2.857142, while the DB method performs better with a score of 1 for the 2 RC model. The study highlights the critical role of the selection of the NLLS algorithm in accurate parameter estimation and highlights the need for careful consideration in practical applications.