Nonlinear Stochastic Degradation Modeling And Remain Useful Life Prediction Of Lithium-ion Batteries

Lithium-ion batteries（LIBs）are widely used in various energy storage and power systems,such as water power,thermal power,wind power,solar power stations,and so on.LIBs provide necessary power for electric bicycles,electric motorcycles,electric vehicles,military equipment,aerospace,etc.However,the safety of lithium-ion battery is still key concern in the field of academic and engineering.In general,the performance degradation of lithium-ion batteries is a nonlinear and time-varying dynamic electrochemical process,which is affected by internal and external environmental loads over time.When the battery capacity exceeds a presetting threshold,the battery may cause catastrophic events,e.g.,electrical fires and overheating.In order to ensure the safe operation of LIBs,accurate prediction of remaining useful life（RUL）of these batteries is a critical issue for the entire equipment,even the power systems.In this research,the data-driven RUL prediction methods are studied to construct an appropriate battery degradation model and a health index;then,its RUL prediction model is constructed to improve the prediction accuracy and practicability of LIBs.The main contents and contributions in this study are summarized as follows:First,due to the nonlinear characteristics in the capacity degradation process of the lithium-ion battery,it is necessary to design an appropriate health index to accurately describe the degradation process and reflect the battery capacity change.In this study,the autoregressive moving average model（ARMA）is used to find the desired HI.In the prediction process,the collected voltage and current data are used to determine the battery capacity degradation,and then the relationship between different parameters and their change of capacity is determined during the batter charge and discharge.After that,the correlation coefficient between each parameter and battery capacity is calculated based on the experimental analyses.By comparing their values,the desired HI can be selected to more accurately reflect the degradation of the battery.Second,considering that the health factor constructed by fusing multi-dimensional features can reflect battery degradation more comprehensively,a new RUL prediction method based on multi-entropy is proposed for LIBs.In this method,the wavelet denoising method is first used to preprocess the collected data and remove the abnormal changes and bad points in the battery data.Then,a prediction model based on relevance vector machine（RVM）and multi-entropy is established.The input of the prediction model is a linear weighting of multiple entropy measures,and the prediction accuracy is taken as the objective function to optimize the weights.The experimental results indicate that the HI based on multi-entropy and its weight optimization can obviously improve the performance of RUL prediction model of the LIB.Finally,in order to improve the accuracy and real-time performance of LIB RUL prediction,a hybrid RUL prediction model based on unscented Kalman filter（UKF）and RVM is proposed in this study.First,because the estimation using the UKF has high accuracy,strong robustness and low computational complexity for the nonlinear system,its state can be estimated and adjusted iteratively by the UKF to accurately track the degradation process of the LIB.Second,the RVM model is trained with the obtained battery data sets,and some relevance vectors（RVs）are used to characterize the battery degradation curve.The predicted residual information and UKF iterative estimation are used to update the parameters of the prediction model.Accordingly,accurate multi-stepahead prediction results of LIBs are obtained.Experimental verification and method comparison indicate that the proposed hybrid prediction model can realize the real-time evaluation of the RUL of LIBs and provides an accurate and effective method to guarante the safe and reliable operation of LIBs.