Learning operando impedance function for battery health with aging-aware equivalent circuit model (2407.06639v3)

Abstract: The wide usage of Lithium-ion batteries (LIBs) requires a deep understanding about battery health. Estimation of battery state-of-health (SOH) is a crucial but yet still challenging task. Pure model-based methods may suffer from inaccuracy and instability of parameter estimations in lifelong aging. While pure data-driven methods rely heavily on quality and quantity of training set, causing lack of generality when extrapolating into unseen cases. In this paper, we propose an aging-aware equivalent circuit model (ECM), which combines model-based and data-driven techniques for SOH estimation. Gaussian process (GP) regression is incorporated in ECM to modelling parameters dependency on operating condition and aging time. The state space formulation of GP is used to enable a computationally efficient co-estimation framework of both parameters and states. Samples from two Open datasets are used to validate model performance, which gives accurate estimation for capacity and impedance. The learnt impedance function can be lined to the shape change of open circuit voltage (OCV) versus SOC curve and thus providing a way to further estimate changes of differential voltage (dV/dQ) curves.