Deep learning method for online parameter identification of lithium-ion batteries using electrochemical synthetic data

Abstract

Accurately sensing the internal state of lithium-ion batteries and identifying parameters is crucial for developing effective battery safety and health management strategies. With the advancement of artificial intelligence, the integration of deep learning (DL) and electrochemical techniques has ushered in new avenues for high-level battery management. However, gaining proprietary battery datasets for neural network (NN) training is difficult at limited data scenarios. To address this problem, a DL method based on synthetic data and convolutional neural network (CNN) is proposed to achieve online parameter identification. Firstly, an electrochemical model (EM) considering the impact of solid electrolyte interface (SEI) film growth on battery performance is established. Then, nine features strongly correlated with electrochemical parameters are extracted from charge and discharge segments for the construction of CNN. Based on offline identified parameters, substantial electrochemical parameter-feature datasets are further generated through polynomial fitting and model calculations, followed by the application of principal component analysis and density-based spatial clustering of applications with noise (PCA-DBSCAN) to filter synthesized data for network training. Finally, the developed online parameter identification method is experimentally validated using full life cycle data of three cobalt-acid lithium (LiCoO2) batteries, confirming reasonable accuracy of the identified battery electrochemical parameters.