State of Health Estimation of Lithium-Ion Batteries with Feature Interpretability Based on Partial Charge Curves

Accurately estimating the state of health (SOH) of lithium-ion batteries is crucial for ensuring their safety and reliability. The “black box” nature of neural networks makes it difficult to interpret prediction results, while the prediction accuracy of neural networks relies on the reliability of feature extraction. This study proposes a method that utilizes Layer-wise Relevance Propagation (LRP) to explain the importance of features, weights the features based on their relevance scores, and estimates SOH using the weighted features. Savitzky-Golay smoothing filter is applied to denoise aging feature data, enhancing the feature correlation of the smoothed data. Additionally, an LRP-LSTM model is employed to capture time-series information related to SOH. An interpretable model not only explains features but also provides feedback to the model, improving its generalization ability. The proposed method achieves an average RMSE of 1.345% and 1.347% on two datasets, respectively.