A MLP-based transfer learning model using EIS health features for state of health estimation of lithium-ion battery

Lithium-ion battery state of health (SOH) estimation remains a significant challenge in battery management systems due to the sophisticated electrochemical processes within the battery. As a model-free method, data-driven-based method has shown great potential in battery SOH estimation. However, the existing data-driven approach requires a large dataset and shows low model adaptability in SOH estimation among different battery samples. To address the issues, this paper proposes a transfer learning (TL)-based technique coupled with the multi-layer perceptron (MLP) and Spearman analysis to realise battery SOH estimation. Firstly, it extracts health features using Spearman analysis based on early-age data of the battery. Next, it builds the basic MLP model relying on the extracted features. Then, the TL model is developed by retraining the MLP model based on the partial data from the target battery. Finally, the retrained model is used to estimate the battery SOH in the rest of the aging cycles. The results demonstrate the high accuracy performance of the proposed method in the battery SOH estimation with an R_score of 0.9733 and RMSE value of 0.53 % in a full-charge stage, implying the prospect of battery SOH estimation using the TL technique.