State of health estimation for lithium-ion batteries based on impedance feature selection and improved support vector regression

Accurate state of health (SOH) of lithium-ion batteries (LIBs) provides valuable input for their performance optimization and lifespan extension. While traditional SOH estimation approaches primarily entail external measures such as current, voltage, and temperature, which, however, are prone to fluctuations and external interferences. Electrochemical impedance spectroscopy (EIS) offers a wealth of information on the dynamic processes within batteries, and provides robust reference for SOH estimation. By leveraging optimized features from impedance spectroscopy, an efficient SOH estimation algorithm is established based on the improved support vector regression (SVR). Firstly, the health indicators are extracted from EIS measures using a sequential forward selection strategy and the multi-objective optimization on the basis of ratio analysis to trade off model accuracy and complexity. Then, the sine sparrow search algorithm is integrated into the hyperparameter optimization of the SVR model. By effectively exploring the hyperparameter space and avoiding local optima, this approach considerably promotes the accuracy of battery SOH estimation. The experimental results demonstrate that the feature selection method employed remarkably reduces feature numbers, removes irrelevant features and shortens EIS test time. Under varying temperature conditions, the proposed algorithm leads to the maximum SOH estimation error of 2.58 %.