SOH Estimation For Lithium-ion Battery Based On Charging Process Features And Multi-model Fusion

The problems of environmental pollution and sustainable energy development have been paid more and more attention worldwide.Therefore,electric vehicles have gradually become an important direction of automobile development.However,as the main power source of electric vehicles,the performance of lithium-ion battery will gradually degrade with the increase of charging and discharging times.In order to improve the reliability and effectiveness of lithium-ion battery,it is necessary to manage lithium-ion battery.State of health(SOH)is an important part of battery management system(BMS)monitoring and management.The accuracy of its estimation is directly related to the reliability of electric vehicles operation.A SOH estimation method for lithium-ion battery based on charging process features and multi-model fusion is proposed in this paper.The main contents are as follows:(1)The aging experiment was carried out,and the aging features of the lithium-ion battery were extracted from the constant current(CC)and constant voltage(CV)stages in the charging process of the aging test.The principal component analysis method was used to reduce the feature dimension.(2)A SOH estimation method based on grey correlation degree combination model is designed.Firstly,the grey correlation degree between the estimation results of WOA-LSTM,WOA-Elman and WOA-SVR models and the SOH of the training set in the early,middle and late stages of lithium-ion battery aging was calculated.Then the estimated weights of the three models at different stages were determined.Finally,the SOH were estimated by model weighting.The validation was performed using a laboratory test dataset.(3)A SOH estimation method is designed to improve the patch learning framework.The WOA-LSTM model was selected as the global model,and the WOA-Elman and WOA-SVR estimation models are used as the patch models.The regions with large errors were searched out in the training results of the global model and marked as patch regions.The patch model was used to train the patch area,and the update of patch position was added to the test set.Finally,a patch learning model composed of a global model and a patch model was obtained.The validation was performed using a laboratory test dataset.The innovations of this paper are:(1)The lithium-ion battery aging features are extracted only from the charging process.The selected features were also reduced in dimension using principal component analysis.It maximized the retention of original features information while reducing the redundancy of features.(2)A SOH estimation method based on grey correlation degree combination model is proposed.The weight coefficients of WOA-LSTM,WOA-Elman and WOA-SVR models in different attenuation stages of lithium-ion-ion battery were determined.Finally,the SOH of each stage was estimated by the method of model weighting,which improved the accuracy of model estimation.(3)A SOH estimation method for lithium-ion battery based on an improved patch learning framework is proposed.The patch model was used to reduce the local error of the global model and improved the estimation accuracy and adaptability of the model.