Research On Capacity Estimation And SOH Prediction For Lithium-ion Battery Based On SOC Interval Stress

Lithium-ion battery is widely used in power storage,electric vehicles and communication systems for its high energy density,long cycle life and high voltage platforms.However,considering the power demand,safety and power performance,lithium-ion battery in electric vehicles or energy storage application scenarios is usually used under particular SOC ranges.The cells under different operating conditions will be affected by many factors in the process of degradation,and show various capacity degradation paths under different interval stresses.Based on this background,the cyclic degradation performance of lithium-ion batteries under different interval stresses is studied in this paper.The specific research contents are as follows:Taking 2.75Ah 18650 NCM material battery as the research object,11 different partial SOC intervals were divided according to battery practical operating interval,phase transition interval and 20%discharge depth(DOD)interval,based on the actual operating conditions and degradation mechanism of the battery.Referring to the general lithium-ion battery test manuals,the test protocol of cycle life test and performance test at different temperatures were designed.28 lithium-ion batteries of the same batch were selected to carry out cycle life and performance tests for 15 months.Based on the test data,the results of capacity degradation among three kinds of SOC interval were compared and analyzed,and the different stages of battery aging were divided.The variation of internal resistance,incremental capacity curve and capacity regeneration phenomena with cycle times are summarized.The mechanism of capacity regeneration is explained from the perspective of electrochemistry.The effects of constant voltage charging process,interval characteristics and phase transition process on battery aging in different SOC ranges were analyzed.The Keras deep learning library in python is used to complete the construction,initialization and training of the cycling aging model of lithium-ion batteries based on LSTM RNN network.And the accurate prediction of battery capacity degradation under arbitrary interval stress is realized.The optimal hyperparametric combination of the model is obtained by parameter search.The LSTM model at different minimum values is fused by snapshot ensembling method to further improve the accuracy of the model.Considering the actual operating conditions of power batteries,two health state characteristic parameters,peak height and peak position voltage of IC curve,were extracted.The contribution of different covariance functions to the output and accuracy of the model is explored by using single kernel function and compound kernel function respectively;the correlation analysis of features and battery capacity is carried out using multiple correlation analysis methods,which verifies the correctness of feature selection and the advantages of multi-output Gauss regression method.In view of the lack of historical data,the relationship between the proportion of training data and the accuracy of estimation was studied.Finally,the residual life prediction of the cell in the battery pack is realized by using the multi-output Gauss process regression algorithm.