A review of lithium-ion battery state of charge estimation based on deep learning: Directions for improvement and future trends

With the rapid growth in productivity, the need for fossil has increased, spurring research and development of new energy sources. In the automotive industry, the development of green-car has made great achievements on the way to promoting climate change and green development. And battery-only electric vehicles will be the mainstream of the future automotive industry. For electric vehicles, lithium-ion batteries are now the dominant energy system. By monitoring and maintaining them through a battery management system (BMS), the safety and reliability of electric vehicle operation can be ensured. The state of charge (SoC) represents the remaining charge of the battery and is an important parameter in BMS. It also assesses the stamina of the electric vehicle. In particular, due to the development of deep learning in other areas such as image processing, automatic speech recognition, natural language processing, etc. Deep neural networks have been widely used in the field of battery state estimation. According to the relevant work, this review classifies recent state of charge estimation methods based on deep learning into structured adjustment and unstructured improvement. Trends in the application of network structures over time are shown in the article. Various key implementation factors of deep neural network methods are also reviewed in terms of feature engineering, data augmentation, learning rate strategies, optimization functions and optimal hyper-parameters. In addition, the theory and key techniques of existing methods are also reviewed. The results of the estimation methods are analyzed and summarized. Finally, the review discusses potential future directions for lithium-ion battery state of charge estimation methods in electric vehicles. •A summary of the progress of SOC estimate approaches based on deep learning.•Learn about the development of structured improvements to estimation methods.•Understanding the many techniques to improving estimating accuracy in an unstructured manner.•Provide fresh ideas for estimating SOC utilizing deep learning methodologies and technical approaches for researchers.