End-to-end deep learning powered battery state of health estimation considering multi-neighboring incomplete charging data

Abstract

Accurately monitoring the State of Health (SOH) of lithium-ion batteries is one of the key technologies critical for the safe and reliable operation of Electric Vehicles (EVs). However, the random charging–discharging behavior of EV users results in differentiated, incomplete, and limited health information in single charging–discharging data, leading to ongoing challenges in SOH estimation. Firstly, an efficient data preprocessing algorithm is designed to automatically handle data slicing, cleaning, alignment, and recombination. Secondly, recognizing the relatively slow change in SOH, multi-neighboring incomplete charging segments are aligned and used as inputs to the SOH estimation model, and then the average capacity of these neighboring charging data is computed as SOH labels. Furthermore, based on the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) structure, an end-to-end SOH estimation model is constructed. This model initially utilizes a multi-channel CNN to fuse health information from multi-neighboring incomplete charging segments and then employs LSTM for SOH prediction. Finally, on a dataset composed of 20 EVs, the proposed SOH estimation method is rigorously validated using K-fold cross-validation. The results demonstrated that the Mean Absolute Error (MAE) is within 2.13%, and the Root Mean Square Error (RMSE) is below 2.74%, highlighting the model’s high estimation accuracy.