Early prediction of lithium-ion battery lifetime via a hybrid deep learning model

Abstract

Accurately predicting the lithium-ion battery lifetime in the early-cycle stage is vital for the optimization of in-use batteries, and also speeds up the development of new batteries. However, traditional methods are incapable of solving nonlinear and negligible capacity fade in early cycles. In this study, a hybrid deep learning model combining a convolutional neural network and a long short-term memory network is proposed to evaluate battery lifetime. Firstly, owing to negligible capacity fade in early cycles, the cycle-to-cycle evolution of capacity-voltage curves is proposed to reflect the potential aging characteristics. Secondly, spatial features and temporal information are extracted by the parallel convolutional neural network extractor and long short-term memory network extractor independently. The complementarity of spatiotemporal information can effectively improve prediction accuracy and stability. Lastly, the output of two extractors is integrated to map into battery lifetime. Experimental results show that the proposed model outperforms other baseline models in accuracy and stability. The end-to-end characteristic makes the model more conducive to deploying in an offline system than traditional approaches.