Research On Remaining Useful Life Prediction Method Of Lithium Batteries Considering Uncertainty Factors

With the sharp increase in energy demand from countries around the world,global fossil fuel reserves are rapidly decreasing,leading to a large number of energy crisis and environmental protection issues.Due to the long service life,high energy density,and environmental friendliness,lithium batteries have been widely used in many fields such as new energy vehicles,energy storage systems,and various consumer electronics systems.However,the energy conversion efficiency of lithium batteries will continuously decrease during their use,leading to problems such as over discharge,over charging,and rapid heating.In order to ensure the smooth operation of lithium-ion battery systems,it is of great significance to monitor the health status of lithium-ion batteries and develop relevant maintenance and support strategies.However,in practical engineering,due to the uncertainty of the working environment,task requirements,and individual differences in lithium batteries,there are often significant difficulties and challenges in accurately modeling the degradation of lithium batteries.In addition,due to the mutual influence between the internal chemical reaction mechanisms of lithium batteries,the nonlinear coupling effect between the uncertainty factors involved in their degradation process will also limit the accuracy of the remaining useful life prediction of lithium batteries.Based on the above issues,this dissertation mainly focuses on the research work on various uncertainty factors involved in the degradation process of lithium batteries,and constructs corresponding models for predicting the remaining useful life of lithium batteries under the conditions of considering various uncertainty factors.The main work and innovative points of this dissertation are as follows:1.Starting from the operating mechanism and degradation mechanism of lithium batteries,analyze the uncertainty factors contained in the internal failure mechanism of lithium batteries,and explore the internal reasons for the formation of uncertainty factors in lithium batteries.Based on independently conducted lithium battery cycling aging experiments and open-source data provided by various research institutions,the characterization parameters of the health status of lithium batteries were analyzed,and the external manifestations of uncertainty factors in lithium batteries were obtained.Based on the analysis of failure mechanisms and related aging experimental processes,the types of uncertainties in the modeling process of lithium battery degradation were summarized.From the perspectives of model uncertainty,data uncertainty,and the simultaneous inclusion of two types of uncertainty,research work was conducted on the modeling methods of lithium battery degradation considering uncertainty factors.2.A Monte Carlo method based on data reconstruction is proposed to address the model uncertainty in the modeling process of lithium battery degradation caused by a small number of historical sample conditions.Firstly,a data reconstruction algorithm based on the Wiener process is used to expand the historical lithium battery samples,providing sufficient prior information for degradation modeling.Then,a parameter update method based on Bayesian theory was proposed to update the posterior parameter distribution of reconstructed historical lithium battery samples and models,improve the consistency between model training samples and target lithium battery degradation data,and reduce model uncertainty caused by a small number of historical lithium battery sample conditions.Based on the updated model training data,an extreme learning machine is used to obtain the remaining useful life prediction results of the target lithium battery.Finally,a comparative experiment based on the lithium battery dataset provided by the Massachusetts Institute of Technology and Stanford University in the United States shows that the proposed method can greatly reduce the impact of model uncertainty on the prediction of the remaining useful life of lithium batteries,and therefore has high prediction accuracy.3.A transfer learning model based on capacity regeneration phenomenon monitoring algorithm is proposed to address the data uncertainty brought by capacity regeneration phenomenon in modeling the degradation of lithium batteries.Firstly,a non-parametric capacity regeneration phenomenon detection algorithm was proposed for online detection of capacity regeneration phenomena that occur during the degradation process of lithium batteries,and quantitative correction was made for prediction errors caused by different capacity regeneration phenomena,thereby greatly reducing the data uncertainty caused by capacity regeneration phenomena.Then,the proposed data reconstruction algorithm and parameter update scheme are adopted to further reduce the uncertainty brought by the capacity regeneration phenomenon on the degradation trajectory of lithium batteries.Finally,based on the reconstructed lithium battery capacity degradation data,a transfer learning model based on long short-term memory neural network is adopted to predict the remaining useful life of the target lithium battery.Comparative experiments based on lithium battery aging samples provided by NASA show that the method proposed in this dissertation can effectively reduce data uncertainty caused by capacity regeneration phenomenon and achieve higher accuracy in residual life prediction than existing methods.4.In response to the data uncertainty and model uncertainty brought about by the dynamic nonlinear characteristics of lithium batteries in the degradation modeling of lithium batteries,this dissertation proposes a Wiener process model based on polynomial fitting.Firstly,a Wiener process model based on polynomial fitting was constructed to adapt to the complex nonlinear degradation characteristics of lithium batteries,thereby reducing the data uncertainty brought by nonlinear characteristics in modeling lithium battery degradation.Then,the prior parameters of the proposed model are obtained through the least square method and the maximum likelihood estimation algorithm,and the Bayesian algorithm is applied to update the posterior parameters of the proposed model to adapt to the dynamic changes in the degradation characteristics of lithium batteries,thereby reducing the model uncertainty caused by dynamics in modeling lithium battery degradation.After obtaining updated model parameters,a general expression for the probability density function of predicting the remaining useful life of lithium batteries based on n(n = 1,2,3...)polynomial Wiener processes was derived,which intuitively characterizes the uncertainty brought by the dynamic nonlinear degradation characteristics of lithium batteries to degradation modeling.Finally,the theoretical correctness of the proposed model was verified using simulation data,and the effectiveness of the proposed model in practical engineering was verified using actual degradation data of lithium batteries provided by the University of Maryland and the experiments conducted by this dissertation.