Enhanced online identification of battery models exploiting data richness

The online model parameter identification is essential to ensure the accuracy and dependability of other battery management system (BMS) tasks in the case of battery degradation and operational environment change. Traditional recursive least squares (RLS) algorithms have always been dependent on persistently exciting data, which limits their ability to operate online when this cannot be guaranteed. This paper proposed a modified RLS method that selects the data richest point for parameter identification. In this model, Fisher information matrix and Cramer-Rao bound are utilised to evaluate the data richness. The final algorithms solve the operational limitations of RLS algorithms, enabling a reliable online model parameter identification under real-world dynamic conditions. The identified model parameters from the single cycle dynamic stress test (DST) of an NCM battery are verified by terminal voltage and state of charge (SoC) estimation with the Root Mean Square Error (RMSE) 0.0332 and 0.0131, respectively.