Online Parameter Identification for Equivalent Circuit Model of Li-Ion Batteries Using the Eigenvalue Addition/Deletion Least-Squares Method With a Forgetting Factor

It is necessary to identify the parameters for an equivalent circuit model of Li-ion batteries and understand their degradation conditions in electric vehicles and consumer electronic devices. As a parameter identification method, the recursive least-squares method with a forgetting factor (FRLS) has been used in several applications of battery management systems (BMS). However, the inverse covariance matrix of FRLS is singular and a covariance matrix does not exist depending on the current pattern (in one example, the current is maintained at zero). In this case, the parameter identification using FRLS is divergent. One approach to solving this challenge is to realize FRLS for the Moore-Penrose inverse (MIFRLS). However, when tested with MIFRLS, certain parameters became almost zero when the current was maintained at zero, and the challenge still remained. In this study, we proposed a new method to solve this challenge and demonstrated improvements in identification. First, we verified that a certain parameter becomes almost zero when the current was maintained at zero. We found that this is caused by noise and the divergence term in the covariance matrix owing to the rank reduction. To solve this challenge, we propose a new method called “an eigenvalue addition/deletion MIFRLS” for BMS which removes eigenvectors with large eigenvalues from the covariance matrix to avoid the divergence term. Lastly, we describe the proposed method to consistently identify parameters, as confirmed by fake numerical tests with known true values, and the average computation time was only 1.47 times longer than that of FRLS.