Research On Data-driven-based Remaining Useful Life Prediction Of Lithium-ion Battery

The performance of a lithium-ion battery,a kind of complex electrochemical system,will degrade owing to continuous usage.Estimating the remaining useful life(RUL)of the lithium-ion battery has become one of the hot topics in the electronic systems research,as a key technology in battery management system(BMS).In recent years,data-driven approaches not based on accurate modeling of the physics of a system have been used;instead,hidden information has been mined through various data analysis methods.The main advantage of data-driven approaches for batteries is that they do not require extensive knowledge on battery chemistry and failure mechanisms.Thus,data-driven approaches have become the mainstream research method for predicting the RUL and the corresponding analysis of lithium-ion battery nonlinear system performance degradation.It is difficult to measure the lithium-ion internal resistance and battery capacity in practical applications,so it is impossible to illustrate the degradation without these available parameters.At the same time,dynamic and environmental load conditions could cause uncertainty in battery degradation illustration.To solve these problems,this work investigates lithium-ion battery RUL methods using data-driven methods from multiple performance single model multi-step prediction and single performance multi-model one-step prediction.The main contributions of this dissertation can be summarized as follows:Firstly,to overcome the low precision problem in long-term prediction,this work proposes a flexible support vector regression(F-SVR),combined with the lithium-ion battery working temperature and energy efficiency analysis algorithm.The most contemporary current research setup studies lithium-ion battery capacity degradation model without considering multiple performance parameters.This paper analyzes the physical performances of lithium-ion batteries and their internal contacts,and the effects of different performance parameters on the lithium-ion battery RUL,and proposes an F-SVR using the lithium-ion battery working temperature and an energy efficiency analysis algorithm.The experimental results show that this method has higher prediction accuracy and is more reasonable than the traditional methods without considering the energy efficiency and battery working temperature.Secondly,to overcome lithium-ion battery RUL prediction uncertainty problems(such as lithium-ion battery multistep prediction error accumulation,capacity decline,model uncertainty,uncertainty in sensor measurements,and uncertainty in the battery monitoring system operation and loading condition),this paper gives an experience model for determining lithium-ion battery capacity degradation with a particle filter(PF)algorithm,and solves the uncertainty involved in lithium-ion battery RUL predictions.This method not only helps in the RUL prediction of the lithium-ion battery but also affords lithium-ion battery RUL probability statistics.These results vividly describe the uncertainty in predicting the lithium-ion battery RUL.Thirdly,to overcome the low precision and reduce the uncertainty arising from not considering the state equation parameter estimation,this paper proposes a lithium-ion battery capacity degradation model parameter estimation method based on the PF algorithm.For parameter estimation,it uses the PF method twice.PF is first used to estimate the parameters of the state equation,and then to predict the RUL of the lithium-ion battery.In other words,this paper first estimates the parameters based on the current empirical model using the PF method,and then one-dimensional lithium-ion battery capacity degradation model through mathematical polynomial derivation as the state equation,again based on PF,is used for predicting the RUL of the lithium-ion battery.The experimental results show that the uncertainty of predicting the lithium-ion battery RUL is reduced by using PF twice,which is parameter estimation.Fourthly,most methods for predicting lithium-ion battery RUL use the single model that lay different situations.Thus,the predicted results reflect poor generalization ability.To address this problem,we use a novel data-driven prognostic approach in this study,the interacting multiple model particle filter(IMMPF),to predict the RUL of lithium-ion batteries.The IMMPF not only considers the characteristics of different capacity degradation models but also combines the predictions of each model.There are three types of typical capacity degradation models based on the IMMPF approach: the polynomial,exponential,and Verhulst models.The experimental results show that we improve the generalization ability of predicting the lithium-ion battery RUL.In this paper,the data-driven is adopted while flexible support vector regression(F-SVR)and particle filter(PF)are the basic approaches for RUL prediction due to its uncertainty expression capability.The aim of our research is to solve these problems on prediction efficiency,accuracy,and adaptability through algorithm research and lithium-ion battery performance characteristics.Thus,this paper finds a new method for predicting lithium-ion battery RULs.