State Of Charge Estimation Of Lithium Battery Based On Deep Learning

Driven by both climate change and sustainable development,the global energy transition is being carried out in the direction of "electrification of energy and clean power".In this regard,China is also actively working to develop various policies to promote global environmental improvement.In 2012,China began the transformation and upgrading of the automotive industry,adhering to the strategic orientation of pure electric drive,the development of new energy vehicles.As one of the most important components of electric vehicles,the maturity and development of battery technology is an effective way to enhance the market competitiveness of electric vehicles.At present,lithium battery is considered as an ideal power source for electric vehicles by its high energy density,low self-discharge rate,near-zero memory effect,high opencircuit voltage,and long service life.State of Charge(SOC)estimation is an important part of the BMS,and it is crucial to accurately estimate the battery SOC for the safe and reliable operation of the on-board battery.However,the SOC of the battery cannot be measured directly by the in-vehicle system and can only be estimated indirectly through other measurable data during the battery operation.Considering the nonlinear nature of the relationship between the measured value and the corresponding SOC of a lithium battery during operation,a data-driven deep neural network model can be used to achieve the estimation of the battery SOC.In this paper,we take the lithium battery of electric vehicles as the research object and study the deep-learning based SOC estimation method of lithium batteries around the core problem of battery SOC estimation.Firstly,the current research status of the problem of lithium battery SOC estimation at home and abroad is explained,and based on this,the research direction is analyzed and determined.A reasonable network structure is designed,and the model is trained and validated by data under different temperature and operating conditions,which proves its good estimation accuracy and robustness and enriches the deep learning-based SOC estimation method in the new network structure.The introduction of migration learning techniques to transfer the knowledge learned by the model among different battery datasets demonstrates the promise of migration learning in reducing training time,improving SOC estimation accuracy,and reducing the amount of training data required.Through the analysis of lithium battery operation data,a U-Net-based data preprocessing structure is proposed for extracting shallow and deep information from the measured data during lithium battery operation and fusing them to provide richer information for the estimation model.The designed U-Net and TCN-based Liion battery SOC estimation model can capture both the temporal correlation in the Li-ion battery operation data series and extract the spatial feature relationship between different parameters.The proposed data processing structure is proved to be effective in improving the performance of model estimation through ablation experiments.In summary,this work enriches the study of the deep learning-based lithium battery SOC estimation problem in terms of research on novel network structures,alleviating the reliance on large amounts of training data,and exploration of model data processing structures.The designed model achieves an average absolute error of 0.65% and the maximum error is controlled within 4.5% in a total of 12 discharge cycles under various temperature and driving conditions.The average absolute error of SOC estimation is as low as 0.37% in one hybrid drive cycle at 25℃.It is demonstrated that the proposed method can be well applied to the Li-ion battery SOC estimation problem and has good estimation accuracy and robustness under different temperatures and operating conditions.