Health Status Prediction Of Lithium Batteries Based On Deep Learning

Abstract

Aiming at the shortcomings of traditional State of Health (SOH) prediction methods in nonlinear modeling and temporal dependence handling, this paper proposes a hybrid CNN-GRU model integrated with the Dung Beetle Optimization (DBO) algorithm (denoted as DBO-CNN-GRU) for lithium battery SOH prediction. Indirect health factors strongly correlated with SOH are extracted from the NASA public dataset, and their effectiveness is verified using Pearson and Spearman correlation coefficients. A CNN-GRU model is designed: the convolutional neural network (CNN) is used to capture local features, and the gated recurrent unit (GRU) is combined to model the temporal dependence of capacity degradation. Furthermore, the DBO algorithm is introduced to optimize the model’s hyperparameters, enhancing the global search capability. Experiments show that the DBO-CNN-GRU model achieves significantly better test performance on the NASA dataset than the single CNN, GRU, and LSTM models.