An Integrated Framework for ARX Model Identification and Its Application to Lithium-Ion Battery

Abstract

This article addresses the critical problem of real-time parameter estimation in dynamic systems using the autoregressive with exogenous input (ARX) model framework, with a particular focus on applications in lithium-ion batteries. The accurate and efficient parameter estimation is essential for optimizing system performance in various contexts. However, traditional gradient-based methods often encounter limitations, such as poor adaptability and reduced reliability under dynamic conditions. To address these challenges, this study proposes a hybrid gradient-based approach that integrates fractional-order gradients and a convex combination mechanism. This approach enables a more broader exploration of the optimization objective. A kernel function is introduced to dynamically adjust the weights of multi-innovation variables, reducing the impact of random errors in parameter estimation. Furthermore, the proposed ARX-based identification framework is tailored to the second-order RC equivalent circuit model (ECM) for lithium-ion batteries. Experimental results suggest that the proposed identification framework possesses the potential to achieve a more satisfactory performance than traditional methods.