Online State-of-Health Estimation of Lithium-Ion Batteries Based on CNN-LSTM Model with Multi-Head SE Attention

Abstract

Lithium-ion batteries gradually lose capacity and show increased internal resistance during repeated cycling, so reliable state-of-health (SOH) estimation is important for safe and dependable battery operation. In this study, an online SOH estimation method is developed using a convolutional neural network–long short-term memory (CNN-LSTM) model combined with a multi-head Squeeze-and-Excitation (SE) attention mechanism. Four health-related features are obtained from charge–discharge measurements: constant-current charging time, constant-voltage charging time, the maximum value of the incremental capacity curve, and its corresponding voltage. The CNN is used to identify local degradation patterns, whereas the LSTM describes the temporal evolution of battery aging. The SE module further adjusts the contribution of different feature channels. Tests on the Xi’an Jiaotong University (XJTU) dataset indicate that the proposed model gives the lowest mean absolute error (MAE) and root mean square error (RMSE) for Battery-1 and Battery-7 and comparable results for Battery-3. These findings support the effectiveness of the proposed method for online SOH estimation of the tested batteries.