State of lithium-ion battery health evaluation based on reconstructed health features and DBO-CNN-GRU

Abstract

Abstract In the feature extraction process of battery health state estimation, redundant features will lead to the increase of model complexity and exponential growth of training parameters, while a single feature will lose local features and lead to the decrease of battery health state estimation accuracy. Therefore, it is very significant to choose appropriate features to represent the health state. In this paper, a CNN-GRU model combining reconstructed health features and a DBO optimization algorithm is proposed for battery health state estimation. Firstly, the health features are extracted from the voltage and current data of the lithium-ion battery during charging and discharging. Secondly, the optimal variational mode decomposition is used to decompose the health features, and the DBO is introduced to optimize the hyperparametric of the CNN-GRU model. Finally, the NASA battery dataset is selected to design comparative experiments of different models. The results show that SOH can be predicted more accurately based on the reconstructed health features and DBO-CNN-GRU model, which verifies the feasibility of the proposed method.