Research On Remaining Useful Life Prediction Method Of Lithium-ion Battery Based On Machine Learning

Lithium-ion batteries are widely used in the field of new energy with their own advantages,and the remaining useful life of the batteries has been widely concerned.It is difficult to establish an accurate degradation mechanism model to predict the remaining useful life of Li-ion batteries in practical applications.Therefore,this paper uses a data-driven approach based on performance degradation parameters.The method uses experimental battery life data,publicly available data sets from NASA and CALCE to conduct research on the remaining life of lithium-ion batteries to improve the accuracy of predictions.This paper accomplishes the following:First,the study designs battery life experiments to obtain historical data on the charge and discharge of lithium-ion batteries.Experiment using the battery capacity tester GN-CD30V10A of Silienergy.Li-ion battery is a 5550mWh 18650 battery with a nominal capacity of 1500mAh.To avoid experimental data homogenization,the NASA public dataset and CALCE public dataset are used for joint validation and provide data support for subsequent experiments.Secondly,for the difficulty of obtaining the capacity of Li-ion batteries online,an intelligent algorithm is used to construct an indirect health Indicator.Six degradation parameters containing the degradation characteristics of Li-ion batteries are extracted from the experimental data.For the existence of some nonlinear features among the extracted 6 parameters,a new HI was constructed by fusing the selected 6 parameters using kernel principal component analysis.Then,the degradation trend of Li-ion batteries contains strong time-series nature,and the LSTM neural network has good prediction ability for time-series data.Therefore,this paper uses the KPCA method to construct HI and then combines LSTM neural network to propose the KPCA-LSTM indirect prediction model for the remaining service life of lithium-ion batteries.The validity of the model was verified after conducting experiments.Finally,considering the local fluctuations in the data due to the different working conditions of lithium-ion batteries in daily use,using this part of the data for prediction will have an impact on the accuracy of the results.Multi-scale decomposition of battery history data using EEMD and EWT methods.The data are divided into trend and error factors,and a prediction model is built for each of them for simulation experiments.The RUL prediction software for lithium-ion batteries is designed according to the prediction model proposed in this paper.The software can dynamically set some parameters of the prediction model and visualize the results,which has certain practicality.Figure[66]Table[25]Reference[81]...