Abstract
Lithium-ion (Li-ion) batteries play a crucial role in modern energy storage systems, with their performance and longevity heavily dependent on accurately assessing their State of Health (SOH). Electrochemical Impedance Spectroscopy (EIS) has emerged as a powerful technique for SOH evaluation, capturing the battery's intricate electrochemical properties. However, practical EIS implementation poses challenges due to the need for expensive equipment and controlled testing conditions. This study introduces a data-driven approach to estimate the SOH of Li-ion batteries using EIS data. An autoencoder backpropagation neural network (AE-BPNN) was developed for unsupervised processing, dimensionality reduction, feature extraction, and SOH estimation. Two optimization algorithms-Scaled Conjugate Gradient (SCG) and Resilient Backpropagation (RBP)-were utilized to tune network weights and enhance performance. Experiments were conducted on eight Eunicell cells across six operational states (I, II, III, IV, V, IX) at various temperatures (25 °C, 35 °C, 45 °C). The AE-BPNN model demonstrated significant advantages over Gaussian Process Regression (GPR) and Support Vector Regression (SVR), yielding lower Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), alongside higher R² scores. Across all evaluated states, the AE-BPNN achieved the lowest average RMSE values of 0.0192 and 0.0176 for the 35C02 and 45C02 cells, respectively, compared to GPR (0.0429, 0.0485) and SVR (0.0404, 0.0334), thereby confirming its superior accuracy in estimating the state of health of Li-ion batteries.