Key points are not available for this paper at this time.
Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for battery safety management. Traditional prediction methods often struggle to balance prediction accuracy with model complexity, and frequently lack interpretability. This paper presents an enhanced ShuffleNet model with physical constraints for lithium battery RUL prediction. The model utilizes a parallel-branch architecture. One branch, the feature enhancement branch, constructs a triple-pathway enhanced ShuffleNet to improve the model’s ability to capture battery degradation patterns. The other branch, the physical constraint branch, establishes a dual-layer constraint mechanism based on the Arrhenius degradation equation and Coulombic efficiency effects, leveraging various dimensions of physical information to enhance the reliability and practical utility of the model’s predictions. A triple loss function incorporating prediction error, physical parameter constraints, and measured data consistency is then designed to optimize model training. The proposed model is verified on the CALCE dataset and the XJTU dataset which contains irregular charge and discharge. Experimental results demonstrate that it achieves prediction errors within 2 cycles on the CALCE dataset and 3 cycles on the XJTU dataset. The model, consisting of 486,636 parameters, completes training in 128 s and predictions in 0.08 s, showing improvements of 30.9%–79.7% over existing methods, thereby advancing both theoretical understanding and practical implementation of lithium-ion battery RUL prediction.
Feng et al. (Sat,) studied this question.