A novel state-of-health estimation for the lithium-ion battery using a convolutional neural network and transformer model. (1st January 2023)

• Record Type: Journal Article
• Authors: Gu, Xinyu; See, K.W.; Li, Penghua; Shan, Kangheng; Wang, Yunpeng; Zhao, Liang; Lim, Kai Chin; Zhang, Neng
• Is Part Of: Energy. Volume 262:Part B(2023)
• Publication Date: 2023-01-01
• Subjects: Lithium-ion battery -- State-of-health (SOH) -- Convolution neural network (CNN) -- Long short-term memory (LSTM) -- Transformer; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.125501
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: State-of-health (SOH) estimation of lithium-ion batteries is crucial for ensuring the reliability and safety of battery operation while keeping maintenance and service costs down in the long run. This study suggests a novel SOH estimation based on data pre-processing methods and a convolutional neural network (CNN)-Transformer framework. In data pre-processing, highly related features are selected by the Pearson correlation coefficient (PCC). Principal correlation analysis (PCA) is also employed to minimize the computational burden of the estimation model by eliminating redundant feature information. Then, all the features are normalized by the min-max feature scaling method, which will speed up the training process to reach the minimum cost function. After pre-processing, all the features are fed into the CNN-Transformer model. The dataset of the battery from the NASA is employed as a training and testing dataset to build the proposed model. The simulations indicate that the proposed performance, proven by absolute estimation errors for each dataset, is within 1%. The estimation performance index is proven by mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) are held within 0.55%. These show that the proposed model can estimate the battery SOH with high accuracy and stability. Graphical abstract: Image 1 Highlights: Three data preprocessing methods are used, which are PCC, PCA, and feature scaling. Transformer structure was firstly applied for SOH estimation. CNN-Transformer can inherit the advantages of both CNNs and Transformers. The dynamic process of PCA and feature extraction are visualized.