A Multilevel Life Prediction Method For Lithium-ion Batteries Based On Particle Filter And ARIMA

The life prediction provides a theoretical basis for predictive maintenance of power batteries to avoid serious accidents caused by battery aging or battery failure.Life prediction also reminds owners to replace the battery in time to ensure sufficient driving range.Therefore,the study of life prediction has important engineering application value.The power battery on electric vehicles(EVs)is affected by its working condition and environment,thus the state of health(SOH)trajectory is complex,and the local fluctuation of the SOH greatly improves the life prediction difficulty.To improve the life predciton accuracy of lithium-ion batteries on EVs,the detailed research that has been carried out focusing on the ternary lithium batteries on EVs is presented as follows.SOH estimation is the basis of life prediction.Considering that the ambient temperature influences the SOH estimation,a SOH estimation method incorporating the ambient temperature is proposed.Extracting the voltage integration from partial charging voltage curve avoids the vehicle driving phase in which the voltage and current fluctuate drastically,so the voltage integration gains high practical value.Considering the SOH estimation accuracy and the practicality of voltage integration,a proper voltage interval is chosen to extract the voltage integration.With the NCA ternary lithium battery accelerating life test data,inputing the voltage integration and ambient temperatre to the Relevance Vector Machine(RVM)gains SOH,thus SOH can be estimated at various temperatures,and the estiamtion average error is 0.84%.The fluctuation of the SOH series make the life prediction more complex,thus a multilevel lithium-ion battery life prediction technology,PF-ARIMA,is proposed based on the the joint of Particle Filter(PF)and Autoregressive Integrated Moving Average Model(ARIMA).Empirical Model Decomposition(EMD)is used to extract the degradation trend and the fluctuation details from the SOH sequence.The PF and ARIMA are used to predict the degradation trend and fluctuation details,respectively.Adding all the predictions and comoparing the SOH failure threshold yields a multilevel life prediction,thus improving the prediction accuracy.The PF-ARIMA method is verified by the accelerating life test data of NCA ternary lithium battery with average relative error of 3.50%,indicating that the PF-ARIMA method can accurately precdict the RUL of NCA ternary lithium battery.Batteries with different kind of materials have different degradation mechanism,thus the data of NCM ternary lithium battery accelerating life test is used to verify the PF-ARIMA method and an ARIMA-EMD life prediction method is also proposed.The accelerating life test incorporates both constant current discharge and alternating current discharge.Life is predicted by PF-ARIMA method with an average relative error of 2.14%,and the average relative error of ARIMA-EMD is 5.05%,both of which are lower than that of particle filter algorithm,Long Short-Term Memory Neural Network,thus PF-ARIMA and ARIMA-EMD method can accurately predict the life of NCM ternary lithium battery,and the PF-ARIMA method has a stronger robustness and more accurate RUL prediction.The life prediction methods in the literature have not been verified by the batteries' data collected from the electric vehicles,thus the PF-ARIMA life prediction technique is verified with the electric vehicle's data.Voltage integration and ambient temperature are first extracted to estimate the SOH,and then the life is predicted with PF-ARIMA method.The concept of effective driving range is proposed and the life is measured in kilometers,and the cubic spline interpolation method is used to improve the life prediction accuracy.The average relative error of the life prediction is 1.90%for two electric vehicles' power batteries,showing that the PF-ARIMA method can be used to estimate the RUL of batteries on EVs.The research results validate that the multi-level life prediction technology based on PF-ARIMA can accurately predict the life of electric vehicle power battery,thus this technology has important practical application value.In addition,the prediction results of PF-ARIMA method are more accurate than three other methods,indicating that the PF-ARIMA method has certain guiding significance for improving time series prediction accuracy.Therefore,the multi-level prediction method based on PF-ARIMA also has important theoretical value.