Estimating battery state of health using impedance spectrum geometric health indicators and recurrent deep sigma point process

Abstract

Characterizing the state of health for lithium-ion batteries remains a formidable challenge due to their highly complex and nonlinear behavior. Frequency-domain impedance measurements can unveil multiple electrochemical processes within the battery, enabling a comprehensive and accurate depiction of the battery’s dynamic characteristics. This study utilizes impedance spectroscopy geometric health indicators and proposes a hybrid state of health estimation method based on recurrent deep sigma point processes. First, based on prior knowledge of battery dynamics, several geometric health indicators that are approximately linearly correlated with state of health were extracted from the electrochemical impedance spectroscopy Nyquist plot. Then a deep sigma point process model with recurrent structure is designed to implement state of health estimation, its hyper-parameters are recognized using population based training. Finally, the proposed estimator has been validated on battery experimental data under different aging stresses, and the results indicate that the proposed method has good estimation accuracy and general applicability. Furthermore, beneficial attempts and discussions have been made to determine the optimal timing and scope for implementing electrochemical impedance spectroscopy measurements, showcasing the potential of this hybrid method in online applications.