State of Health Estimation of Lithium-Ion Batteries via Electrochemical Impedance Spectroscopy and Machine Learning

Estimating the state of health (SoH) of lithium-ion batteries (LIBs) is an attractive and challenging task since they face complex aging mechanisms, environmental sensitivity, and poor safety issues. This paper aimes to develop an effective data-driven approach capable of accurately predict battery capacity degradation. Using the strategy of integrating electrochemical impedance spectroscopy (EIS), a novel nonlinear grey wolf optimization (NGWO) and support vector regression (SVR), the proposed model can successfully estimate battery capacity under single and multiple temperature conditions. On the basis of the identical data, SVR combined with GWO, particle swarm optimization (PSO) and genetic algorithm (GA) respectively, as well as the common SVR as comparisons are employed to further evaluate the actual performance of the presented model. The outcomes indicate that NGWO-SVR tends to perform faster, more accurate and stable among these methods. This paper provides a flexible approach for developing data-driven models using EIS spectra under different temperature conditions, which is potentially to be applied to the practice implementation of battery SoH routine monitoring.