State of Health Estimation of Lithium-Ion Batteries Using a Gramian Angle Field-Convolutional Neural Network-Temporal Convolution Network Hybrid Model

To address the issues of low estimation accuracy and insufficient capture of time series features in the existing battery state of health(SOH)estimation, a Gramian angle field-convolutional neural network-temporal convolution network(GAF-CNN-TCN)hybrid model is proposed. The model converts incremental capacity(IC)curves of varying lengths into image data using the GAF algorithm and extracts features from them through a convolutional neural network. Additionally, a feature fusion network is introduced to integrate the image features extracted from the image by a two-dimensional convolutional neural network with the temporal features extracted from the IC sequence by a one-dimensional convolutional neural network. The integrated features are fed into the temporal convolutional network model for training, leading to precise SOH estimation. Validation of the model is conducted using lithium-ion battery data sets from NASA and University of Oxford. The results show that in comparison to the long short-term memory(LSTM)model, the GAF-CNN-TCN hybrid model reduces the mean absolute error(MAE), mean absolute percentage error(MAPE), and root mean square error(RMSE)between the estimated SOH and the true SOH by 85.65%, 86.12%, and 84.0%, respectively. Similarly, compared to the CNN-LSTM model, the reductions in MAE, MAPE, and RMSE are 83.24%, 83.75%, and 82.27%, respectively. Furthermore, compared to the TCN model, the reductions in MAE, MAPE, and RMSE are 76.92%, 77.19%, and 76.01%, respectively. © 2024 Xi'an Jiaotong University. All rights reserved.