CTARNS: Improving Capacity Estimation of Lithium-Ion Battery by Using Convolutional Transformer with Nested Sequence Models

Lithium-ion batteries are rechargeable and have been used widely in mobile phones and even autonomous vehicles because of their lightweight and high energy density. Therefore, predicting the State-of-Health (SoH) of a battery becomes significant and challenging because of its capacity regeneration. Artificial intelligent methods, especially long short-term memory (LSTM), have been applied and achieved fantastic results. However, previous studies either did not account for possible features in a battery cycle or required a long training time because they were based on LSTM. In this research, we propose a nested sequence model called Convolutional Transformer Autoregression Nested Sequence (CTARNS) to improve accuracy and reduce training time. The results show that our proposed model performed approximately 39.13% (RMSE) and 43.95% (MAPE) better than the baseline on the NASA dataset and 14.67% (RMSE) and 23.78% (MAPE) on the CALCE dataset.