Deep Transfer Ensemble Learning-Based Diagnostic of Lithium-Ion Battery

Dear editor, State of health (SOH) estimation is critical for the management of lithium-ion batteries (LIBs). Data-driven estimation methods are appealing with the availability of real-world battery data. However, time- and data-costly training for batteries with different chemistries and models barriers their efficient deployment. Motivated by this, a novel deep transfer ensemble learning method is proposed to estimate the SOH with limited sampling data. Specifically, the convolutional neural network (CNN) is employed for model training based on available data. With the new batteries, the trained CNN model is adapted using only a small proportion of samples with the model selection and parameter-sharing transfer learning (TL). The weighted average ensemble learning (EL) is further incorporated to enhance the estimation performance, giving rise to a novel CNN-EL-TL model. Experimental results suggest that the proposed CNN-EL-TL model can realize accurate SOH estimation even though the model is used among different batteries. The comparison with the CNN, CNN TL, and CNN-EL models validates the effectiveness of introducing the transfer and ensemble learning.