Research On Data-Driven Prediction Methods Of Lithium-Ion Battery Remaining Useful Life

With the increasing proportion of green energy,the global number of new energy vehicles is increasing.Safety accidents caused by battery failures are becoming increasingly prominent,which seriously threaten the property and personal safety of consumers.Accurate prediction of state of health(SOH)and remaining useful life(RUL)are of great significance to improve the reliability of batteries and avoid safety accidents.Considering that the model-based prediction method is susceptible to external factors and has poor universality,this paper mainly uses data-driven method to realize accurate online prediction of SOH and RUL of lithium-ion batteries.The main research contents and results of this paper are as follows:(1)Considering the nonlinear characteristics of lithium-ion batteries capacity degradation,this paper constructs a lithium-ion batteries RUL prediction method based on improved particle swarm optimization(IPSO)and support vector regression(SVR).Firstly,the introduction of IPSO algorithm can effectively solve the parameters selection problem of SVR.Secondly,a new health indicator(HI)is extracted from the discharge voltage of the batteries,and its effectiveness is verified by Pearson correlation analysis.Finally,the HI is used as the input feature of IPSO-SVR model to realize the online prediction of lithium-ion batteries RUL.The verification results on NASA battery dataset show that the prediction accuracy of IPSO-SVR method is significantly higher than that of SVR method.(2)Aiming at the problem that the traditional machine learning model is difficult to predict the noise and capacity regeneration in the process of batteries capacity degradation,this paper proposes a RUL prediction method for lithium-ion batteries based on empirical mode decomposition(EMD),sample entropy(SE)and deep neural network(DNN).Firstly,EMD is used to decompose the batteries capacity degradation sequence into several subsequences such as main degradation trend and noise.Secondly,considering that there are too many subsequences after decomposition and the fluctuation of some high-frequency sequences is large,SE method is introduced to calculate the entropy of each sequence,and the sequences with similar entropy are superimposed for reconstruction.Then grey relation analysis is used to verify the correlation between HIs and reconstruction sequences.Finally,the HIs are used as the input features of DNN model to predict the reconstructed sequences respectively,and the predicted values of each sequence are added to obtain the final prediction results.The verification results on NASA battery dataset show that the prediction accuracy of EMD-SE-DNN method is significantly better than that of DNN method,EMD-DNN method and IPSO-SVR methods.(3)Aiming at the problem of poor temporal memory ability of SVR model and DNN model,this paper proposes a lithium-ion batteries RUL prediction method based on improved genetic algorithm(IGA)and long short-term memory(LSTM)network.Firstly,the introduction of IGA can effectively solve the problem of selecting the structural parameters of LSTM network.Secondly,two new HIs are extracted from the discharge voltage and discharge current,and their effectiveness is verified by Pearson correlation analysis.Finally,the HIs are used as the input features of IGA-LSTM model to realize the online prediction of lithium-ion batteries RUL.The verification results on CALCE battery dataset show that the prediction accuracy of IGA-LSTM method is significantly improved compared with that of LSTM method.