Research For Battery State Of Health Assessment Based On Small Sample Data

Lithium-ion batteries have been widely used in electric vehicles,energy storage systems,aerospace and other fields due to their high output density and power output,high coulomb efficiency,long cycle life and other advantages.However,degradation of electrochemical components can lead to loss of battery capacity and power during the operation of lithium-ion batteries.It is necessary to accurately estimate the state of health(SOH)of lithium-ion batteries,which ensures the safe operation of lithium-ion batteries.In the present study,SOH estimation is mainly based on the data-driven method,which can be realized only by using the data collected during the charging/discharging process of lithium-ion battery,and the capacity is usually used as the measurement of SOH estimation.However,in practical applications,the charging/discharging cycle of lithium-ion batteries takes a long time and spends high collection cost,and the charging/discharging data samples are often insufficient.In addition,under the conditions of vehicle driving and energy storage peak and frequency regulation,the charging/discharging starting point of lithium-ion batteries is relatively random,which leads to the incomplete charging and discharging data of lithium-ion batteries.Therefore,it brings great challenges to the application of existing SOH estimation methods for lithium-ion batteries.In view of the above problems,this thesis aims at single lithium-ion battery and the same type of lithium-ion battery,carries out an in-depth study on SOH estimation of lithium-ion battery based on incomplete charging and discharging data,and proposes a common method for SOH estimation of lithium-ion battery by analyzing the insufficient data and incomplete data characteristics of lithium-ion batteries in practical applications.The main research contents are as follows:(1)Aiming at the difficulty of unifying characteristics of lithium-ion battery under different charging/discharging conditions,a method for extracting charging/discharging characteristics of lithium-ion battery based on different data characteristics is proposed.For the complete charging/discharging data of a single battery,feature points are randomly extracted from the complete data by using Latin hypercube sampling.Due to the characteristics of latin hypercube sampling,feature points are evenly distributed on the charging/discharging curve,which enables the extracted features to accurately represent the charging/discharging curve.According to the incomplete charging/discharging data of different batteries,the characteristics of charging/discharging curves are extracted through the equal voltage interval,and the information of battery aging and performance degradation is added to the neural network for training,which lays the foundation for the subsequent reconfiguration of charging/discharging curves.(2)Aiming at the problem of insufficient samples required for SOH estimation of lithium-ion batteries,a data enhancement method based on improved generative adversarial network model(CGAN-HQOA)is proposed.Pseudo-samples that are closer to the real samples are generated based on the improved generative adversarial network.In the process of generating samples,HQOA algorithm is added to avoid individual samples that deviate from the real sample distribution according to the difference between the generated samples and the real sample distribution in terms of volatility and stability,so as to ensure the diversity of samples on the basis of improving the quality of generated samples.By comparing the data enhancement effects of different generative adversarial networks,the method constructed in this thesis can generate higher quality samples with minimum capacity evaluation error.(3)Aiming at the difficulty of accurate SOH estimation of lithium-ion battery due to incomplete charging/discharging curve,a SOH estimation method of lithiumion battery based on the combination of deep neural network and equivalent circuit model(Phys-DNN)is proposed.Firstly,the method is based on deep neural network,which takes incomplete charging/discharging curve as input and complete charging/discharging curve as output to realize the reconstruction of charging/discharging curve.Then,due to the embedded equivalent circuit model in deep neural network,the model can provide necessary information for the charging/discharging process of networked lithium-ion batteries.In this way,the direction of iterative optimization in the training process of deep neural network is constrained to make up for the driving ability of data on the network when samples are insufficient,thus reducing the dependence of the neural network on samples,and meanwhile,a complete charging/discharging curve can be reconstructed with high precision.By comparing the prediction effect of the deep neural network before adding the model,the method constructed in this thesis can realize the prediction and reconstruction of different lithium-ion battery charging/discharging curve data by using small sample charging/discharging data of a single lithium-ion battery,so as to carry out efficient SOH estimation of lithium-ion battery,and also verify the generalization of the method.To sum up,aiming at the difficulties of SOH estimation of lithium-ion batteries caused by insufficient and incomplete charging/discharging data of lithium-ion batteries,this thesis proposes data enhancement methods based on CGAN-HQOA and SOH evaluation methods based on Phys-DNN respectively to achieve accurate SOH estimation of lithium-ion batteries.It provides key method supports for the operation safety of lithium-ion batteries in practical application scenarios.