An improved CNN-LSTM model-based state-of-health estimation approach for lithium-ion batteries

Accurate SOH (State of Health) estimation is one of the key technologies to ensure the safe operation of lithium -ion batteries. When predicting SOH, efficient data feature extraction is the premise to ensure accurate prediction. In this work, a feature selection method is proposed to help neural networks train more effectively by removing useless features from the input data during the data preparation step. In addition, the skip connection is added to the convolutional neural network-long short-term memory (CNN-LSTM) model in this work to address the problem of neural network degradation caused by multi-layer LSTM. The presented approach is validated on the NASA and Oxford battery dataset. The results demonstrate that after using the feature selection approach to remove the less significant features, the SOH prediction accuracy is enhanced and the computational load on the neural network is decreased. Compared with other neural network models, the CNN-LSTM-Skip model has better robustness and higher accuracy under different conditions, and the RMSE is below 0.004 on the NASA dataset and the Oxford dataset.

Automated summary

The article proposes a feature selection method to enhance the accuracy of SOH prediction by removing insignificant features from the input data during data preparation. Additionally, a skip connection is added to the CNN-LSTM model to address the degradation of neural networks caused by multi-layer LSTM. Experimental results demonstrate that the feature selection approach improves SOH prediction accuracy and reduces computational load. Compared to other neural network models, the CNN-LSTM-Skip model exhibits better robustness and higher accuracy across different conditions, achieving RMSE below 0.004 on the NASA and Oxford datasets.