Robust parameter identification method for accurate estimation of state of charge of lithium-ion batteries considering time-varying ambient temperature

The parameter identification (PI) of lithium-ion batteries (LIBs) is susceptible to coupled time-varying temperature and complex noise interference. Traditional online recursive least squares (RLS) method delivers large errors under such interference, leading to unreliable online parameters. To enhance robustness for accurate state of charge (SOC) estimation, this paper proposes a robust PI method considering time-varying ambient temperature. Firstly, a discrete Thevenin model considering the capacity temperature sensitivity was built, overcoming the static defects of conventional temperature dependent equivalent circuits. Then, the temperature sensitive key factors governing resistance-SOC variation are extracted through physical analysis. On this basis, manifold regularized least squares support vector machine (MR-LSSVM) method was proposed to identify the battery parameters. Unlike single-improved LSSVM, it integrates manifold learning and error adjustment to strengthen anti-interference ability. Lastly, a state-space model is established, and unscented Kalman filter (UKF) realizes online SOC estimation. Lyapunov convergence analysis verifies its validity. Tests conducted under varying temperature and noise conditions have shown that this framework outperforms other methods on both types of LIBs.