A novel deep neural network model for estimating the state of charge of lithium-ion battery. (October 2022)

• Record Type: Journal Article
• Authors: Gong, Qingrui; Wang, Ping; Cheng, Ze
• Is Part Of: Journal of energy storage. Volume 54(2022)
• Publication Date: 2022-10
• Subjects: Lithium-ion batteries -- State of charge -- Deep neural network -- Deep learning -- Simple recurrent unit; Energy storage -- Periodicals; Energy storage -- Research -- Periodicals; 621.3126
• DOI: 10.1016/j.est.2022.105308
• Languages: English
• ISSNs: 2352-152X

Abstract

Abstract: State of charge (SOC) estimation of lithium-ion battery is an important step in battery management system (BMS). As the internal chemical mechanism of battery is a complex process, there is a strong nonlinear relationship between the external measurable variable of battery and SOC. In order to accurately estimate the SOC, this paper proposes a novel deep neural network (DNN) model based on deep learning, which takes the data unit composed of voltage, current and temperature data of the battery sampled in 10 s as the input, and the SOC estimate as the output. The proposed model consists of convolution layer, ultra-lightweight subspace attention mechanism (ULSAM) layer, simple recurrent unit (SRU) layer and dense layer. The convolution layer can extract the features of the input and get the corresponding feature sequence. The ULSAM layer can highlight key information in the feature sequence. The SRU layer is used to process the feature sequence and transfer historical information. The dense layer is responsible for outputting SOC estimate. The proposed model is simulated by using the two public battery datasets, the test results show that the model has high estimation accuracy, and has good adaptability to battery degradation, ambient temperature and discharge conditions. Highlights: Deep learning method is used to solve the problem of SOC estimation. The hybrid neural network is used to establish the SOC estimation model. ULSAM is redesigned to improve the performance of one-dimensional CNN. SRU is selected as the main structure of the SOC estimation model.