Volume 44: Energy Transitions toward Carbon Neutrality: Part VII

Robust online parameter identification for batteries using exponential resetting recursive least squares Sheng Wang, E Zhang, Haomiao Li, Kai Jiang, Kangli Wang

Abstract

Parameter estimation methods based on recursive least squares (RLS) are extensively used in the online identification of battery models. Forgetting factor RLS (FFRLS), which can track the parameter changes online, has been regarded as an essential solution to real-time model adaptation. However, under the non-persistent excitation condition, the performance of FFRLS will degrade and the covariance windup phenomenon will be triggered, as a consequence, FFRLS will become numerically unstable and lose the capacity to provide reliable estimates. In this paper, a new scheme named exponential resetting RLS (ERRLS) is proposed to overcome the aforementioned shortcoming, the mechanism of information matrix updating is modified to guarantee exponential convergence towards a non-zero matrix under no excitation. This modification will result in a bounded covariance matrix whether the excitation is persistent or not, which implies an improved robustness in comparison with FFRLS. Experimental results indicate that ERRLS can achieve better performance than FFRLS when the persistent excitation condition is violated.

Keywords battery modeling, equivalent circuit model, parameter identification, least squares, covariance resetting