Research on predicting the state of health of lithium-ion batteries via back propagation network based on multi-feature combination of electrochemical impedance spectroscopy

Abstract

Lithium-ion batteries are widely used in electric vehicles and portable electronic devices. Accurate estimation of the State of Health (SOH) is essential to guarantee their safe and reliable operation. Electrochemical Impedance Spectroscopy (EIS) can characterize the internal electrochemical aging properties of batteries. However, traditional EIS-based methods only adopt single impedance parameters, which fail to fully describe the coupled aging behaviors and thus suffer from unsatisfactory prediction accuracy. To address this issue, this paper proposes a lithium-ion battery SOH prediction method combining multi-feature combinations of EIS and optimized Back Propagation (BP) neural network. Firstly, we analyze the cyclic aging experimental data of the same type of batteries under different operating conditions. Spearman’s Rank Correlation Coefficient (SRCC) is employed to select valid features highly correlated with capacity degradation, and two sets of EIS multi-feature combinations are established for comparative analysis. Secondly, three optimization algorithms, namely Particle Swarm Optimization (PSO), Genetic Algorithm (GA) and Ant Colony Optimization (ACO), are used to optimize the initial parameters of the BP neural network, which overcomes the drawback that the conventional BP model is prone to falling into local optima. The experimental results reveal that under the operating conditions of 35C01 and 35C02 with insufficient samples and prominent data noise, the BP neural networks optimized by ACO and GA achieve superior prediction performance on the test set, with lower Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). For 45C01 and 45C02 with sufficient samples and low data noise, ACO-BP and GA-BP maintain stable prediction performance. The proposed method realizes the organic integration of electrochemical mechanism analysis and data-driven modeling, and effectively improves the prediction robustness and generalization ability. It provides a high-precision and practical technical solution for the online SOH monitoring of lithium-ion batteries.