Research On Dynamic Capacity Estimation Of Lithium-ion Batteries Based On Deep Learnin

Nowadays,as an important energy storage technology,lithium-ion batteries are widely used in all aspects of life.In order to ensure that the battery is in a safe and reliable state of energy supply,it is necessary to monitor the SOH of lithium-ion battery.The aging process of lithium-ion battery is a dynamic coupling process.Its internal electrochemical mechanism is very complex,and the parameters such as battery materials are difficult to obtain,which greatly increases the difficulty of establishing the mechanism model.The data-driven method establishes the relationship between the battery capacity and the external characteristic information from the perspective of data information,and can realize the SOH estimation of lithium-ion battery.The method based on deep learning has gradually become the mainstream of data-driven methods.The data currently used in the SOH monitoring method of lithium battery based on deep learning are obtained by CC-CV charging and discharging experimental conditions or by CC-CV charging and discharging experiments under determining capacity stage in random charging and discharging experimental conditions.When the random charge and discharge conditions or experimental conditions are changeable,there are still many difficulties in predicting SOH of lithium ion batteries.In response to the above problems,this paper proposes two deep learning methods for SOH estimation of batteries under random charge and discharge aging experimental conditions,namely dynamic conditions,using measured fragment real-time data,and verifies the proposed method using the National Aeronautics and Space Administration lithium ion battery random use data set.The main research contents are as follows.Aiming at the problem that it is difficult to extract the health feature from the measured data of lithium-ion battery dynamic conditions.A capacity estimation method for lithium-ion battery dynamic conditions based on multi-domain features and deep learning algorithm is proposed.Firstly,the 9～12 batteries in the NASA’s lithium-ion battery random use data set were selected as the research object,the health feature extraction and analysis of battery voltage,current,temperature and time measured data are carried out from the perspective of time domain,and the time domain feature is obtained.The wavelet transform method is used to extract and analyze the health characteristics of the battery voltage and temperature data from the time-frequency domain,and the time-frequency domain characteristics are obtained.The inherent characteristics and detailed characteristics of the measured data of the battery are comprehensively excavated.Secondly,the extracted multi-domain feature matrix is used as the input variable of CNN-GRU to construct the battery capacity estimation model.Thirdly,the capacity estimation model is used for model structure and comparative analysis,model hyper-parameter impact analysis,input feature analysis and other experiments to prove the effectiveness of the method.Aiming at the problems of feature redundancy and limited feature information in feature extraction method.A lithium-ion battery capacity estimation algorithm based on 2D feature image of battery voltage,current and temperature measured data under dynamic conditions of lithium-ion battery and deep learning is proposed.Firstly,23 batteries in the NASA’s lithium-ion battery random use data set were selected as the research object.The voltage,current and temperature measured data of the battery were transformed from one-dimensional signal to two-dimensional image according to the unified format.Secondly,the Res-CNN and GRU-RNN are used to build the battery capacity estimation model.Res-CNN is used to extract the feature of the 2D feature images of the battery measured data,and GRU-RNN is used to establish the nonlinear mapping relationship between the inherent characteristics extracted by Res-CNN and the battery capacity.Thirdly,Res-CNN-GRU capacity estimation model is used to carry out model input impact analysis,model structure impact analysis,model performance verification based on large sample data,and model performance verification based on small sample and complex working condition data.Experiments prove the generalization and practical significance of the method.In general,by making full use of the measured data of dynamic random working conditions of lithium ion battery to estimate the battery capacity,a feasible method is put forward to improve the reliability of automobile health status monitoring.