Data-driven Prediction Study Of Lithium Battery Health Status Assessment

In the state proposed "carbon neutral","carbon peak" in the era of the conditions development,lithium-ion batteries with their own many characteristics in new energy vehicles and other fields are being increasingly widespread applications.As part of the core composition of each device,the health of the lithium battery itself depends on the consistent implementation of the mechanism.As the lithium battery is continuously recycled,some complex chemical reactions will occur inside the battery,which will lead to a continuous decline in capacity,so an accurate estimation and forecasting of its state of health will effectively ensure the safe and consistent functioning of the equipment.Firstly,this paper summarizes each method of current lithium battery health state(SOH)estimation by combining the national and international related literature,and selects a data-driven approach based on the analysis and comparison of various methods to estimate SOH of lithium-ion batteries.Furthermore,the compositions and operating principle of Li-ion batteries are explained,and the internal and external factors affecting the decline of their life are analyzed and explained.In the end,the differences of each battery sample in the public data set of Li-ion batteries provided by research centers with high credibility are explained,and suitable battery samples are selected for the subsequent analysis and research of this paper..Secondly,to deal with the inappropriateness of measuring direct health factorial such as capacity and domestic resistance of lithium-ion battery or the high incorrectness of measurement,we choose to extract suitable independent health factors from the variables that can be directly monitored to such parameters as voltage,current and temperature,and then perform the study of lithium-ion battery health state evaluation on the basis of indirect health factors of lithium-ion battery.In consideration of the actual operational circumstances of Li-ion battery and the elements which affect the lifespan of Li-ion battery,suitable health factors are extracted from three aspects,namely charging process stage,discharging process stage and temperature,to complete the feature extraction project under the data-driven method.Then,in order to intuitively analyze the validity and feasibility of the extracted indigenous health coefficients,the Pearson and Spearman correlation coefficients are selected and verified by this paper.The correlation and validity between the selected feature factors and lithium battery capacity decline.Then again,due to the little available data of lithium-ion battery,some characteristic parameters are difficult to extract,which means that its data has the characteristics of small samples,and the support vector regression machine algorithm(SVR)has significant effect in dealing with some data with small sample characteristics,based on this,this paper proposes the SVR method to establish the estimation model of lithium battery SOH,corresponds the independent variable to the dependent variable,establishes the corresponding mapping relationship between the two,accomplish the model calculation,and estimate the health state of Li-ion battery according to equation(1-2).Meanwhile,the SVR model has the problem of difficult parameter finding,and for this problem,the improved Antlion algorithm(IALO)is chosen to optimize and improve the SVR to solve the problem of difficult model parameter selection.The experimental results show that the IALO-SVR based Li-ion battery SOH estimation model greatly improves the generalization ability and estimation accuracy of the SVR estimation model..Finally,in considering to better guarantee the reliable and secure functioning of the system,the prediction research of the health state of Li-ion battery in the future moment is needed to establish a model to forecast the health state of Li-ion battery for the next50 recharges in multiple steps.Here,SOH series data are used to establish the prediction model by choosing the long short-term memory network(LSTM)for the time sequence,meanwhile,for the capacity regeneration phenomenon existing in Li-ion battery,the Empirical Modal Decomposition(EMD)will be used to process the SOH variation signal of Li-ion battery.And compared with the single LSTM prediction model,the simulation results demonstrated that the prediction market on the basis of EMD-LSTM can deal with the partial capacity rebound phenomenon of Li-ion battery,and the accumulated errors are reduced to some different magnitude,which significantly improved the precision of the model.Figure [31] Table [10] Reference [88]...