Research On Prediction Of Remaining Useful Life Of Lithium Ion Battery Based On Improved Neural Network

Lithium Ion Batteries(LIBs)have the advantages of high security capability,long cycle life,low environmental pollution,high energy density and low self-discharge rate.They are widely used in the field of electric vehicle(EV)to provide power.Battery Management System(BMS),as a key component of the energy system of EVs,has functions such as diagnosis,control,and protection.Accurately estimating the State of Health(SOH)and predicting the Remaining Useful Life(RUL)of the battery can enable the battery to operate efficiently and stably.Both SOH and RUL are important performance indicators that reflect the degree of battery aging,and play a vital role in the BMS.Battery performance will gradually weaken with continuous use.SOH estimation and RUL prediction have always been the focus of BMS research.This research used car LIBs as the research object,combined with data-driven algorithms to study the battery SOH estimation and RUL prediction algorithms.The main contents are:First,the development process of EVs and LIBs are introduced,and the current domestic and foreign research status of battery SOH estimation and RUL prediction are described.Second,through the battery test system in the laboratory,the lithium manganese oxide battery has subjected to aging experiments to study the influence of the number of charge and discharge cycles on battery performance at a constant temperature.And the LIBs’ data in the NASA Ames Prognostics Center of Excellence database were summarized.Third,the battery data are preprocessed to extract Health Indicator(HI)that can be used for battery SOH estimation and battery RUL prediction.There are many factors that affect battery performance,such as charging and discharging time,charging and discharging current,external environment,and available battery capacity.In the charging phase,the time,isochronous voltage and isochronous current changes in the constant current charging mode,which are easy to measure,are extracted as health indicators.Then,using statistical Spearman correlation coefficient method,the correlation between the extracted HI and battery capacity is analyzed.Fourth,this paper proposed a battery RUL prediction method based on improved Back Propagation Neural Network(BPNN).During the neural network training process,the dynamic exponential decay learning rate is used to train the neural network;and the particle swarm optimization(PSO)with mutation factor is used to initialize the weights and thresholds of the BP neural network.Then,two parallel improved BP neural network methods are used to predict the battery SOH value and RUL value at the same time,the number of charge and discharge cycles when the battery SOH estimate drops to the failure threshold is used as the judgment condition for the end of the battery RUL prediction.Fifth,the neural network for battery RUL prediction cannot give a confidence interval.Therefore,the Gaussian Process Regression(GPR)model is used to estimate the battery SOH,and the SOH estimation range boundary given by the GPR model is used as the neural network’s boundary Input,take the failure threshold confidence interval as the end judgment condition,so that the battery RUL predicted by the improved neural network method also has the upper and lower limit reference range,which increases the reliability of the prediction results.