A robust identification method for inherent parameters of lithium battery considering thermal distribution and state of charge

In charge/discharge processes, the internal resistances of lithium-ion batteries (LIBs) change dynamically with the current, temperature and state of charge (SOC). However, these resistances are usually assumed to be constant after their value is offline identified, and the influence of the factors, especially SOC and temperature distribution on them is ignored. Besides, noise is less considered during the identification, resulting in poor robustness of the model. Aiming for these problems, a robust identification approach is proposed for the inherent parameters of soft pack LIBs in consideration with the effect of temperature distribution and state of charge as data is polluted by noise. First, to solve the modeling problem of the thermal process with unknown dynamics and unknown boundaries, a spatiotemporal least square support vector machine (LS-SVM) model is developed to reconstruct the temperature distribution of the LIB. Then, a robust adaptive neuro fuzzy (R-ANFIS) approach is developed to construct the model between the internal resistances and the external parameters, i.e., SOC, current and distributed temperature. It develops the kernel strategy to reduce the model dimension and minimizes both mean and variance of error to improve the model robustness. Using experiments at different discharge rates, the effectiveness of this method is tested and verified.