Enhancing real-time degradation prediction of lithium-ion battery: A digital twin framework with CNN-LSTM-attention model

Abstract

Lithium-ion batteries (LIBs) have gained widespread usage in electric vehicles (EVs) due to their high energy density, long cycle life, and environmental friendliness. However, as LIBs undergo repeated charging and discharging cycles, they experience performance degradation. When the rated capacity of LIBs drops to approximately 80 %, retirement becomes necessary. Therefore, accurately determining real-time battery degradation is of paramount importance. This study presents a digital twin framework for analyzing and predicting LIB degradation performance. Within this framework, the back propagation neural network (BPNN) is employed to predict and complete the partial discharge voltage curve of the actual battery cycle. Building upon this, in conjunction with the battery's state of charge (SOC), the convolutional neural networks-long short term memory-attention (CNN-LSTM-Attention) model is utilized to real-time forecast the maximum available capacity of LIBs and reveal the battery's degradation state. Experimental results demonstrate a 99.6 % accuracy in completing the partial discharge voltage. Moreover, the prediction accuracy for maximum available capacity surpasses 99 % with a maximum error of less than 3 mAh. Thus, this research substantiates the efficacy and practical applicability of the proposed approach.