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

The global economy relies heavily on fossil fuels.However,fossil fuels are a kind of non-renewable energy sources.Once exhausted,the energy economy based on fossil fuels will suffer a fatal impact.As a new energy source,lithium-ion batteries are receiving more and more attention worldwide.It has the advantages of long life,stable electrochemical performance,high efficiency,small size and so on,and has bright prospects in many application fields.However,the decline in performance is common in the real world,as is the case with lithium-ion batteries during recycling.In order to reduce the accidents and losses caused by the decline of lithium ion battery performance,the prediction of its remaining service life(RUL)has attracted the attention of many researchers.The main work carried out by the subject in exploring effective prediction methods are:Aiming at the problem of insufficient utilization of feature information implied in the life decay signal of lithium-ion batteries,a multi-scale feature signal decomposition method is proposed to extract features from the original life decay signal.Introduce and discuss a variety of multi-scale feature signal decomposition methods,and choose empirical wavelet transform(EWT)to decompose the life decay signal of lithium ion batteries.In the frequency sub-layer signal obtained by decomposition,the low-frequency sub-layer signal can reflect the transformation trend of the original signal,so it is selected as the characteristic of the original signal.Aiming at the problem of inefficiency in setting the hyperparameter method of the deep neural network prediction model by human experience,a Bayesian optimization method is proposed to automatically obtain the optimal hyperparameter of the network model.The Bayesian optimization method is used to automatically optimize the hyperparameters such as the number of network layers,the learning rate,and the number of hidden layer neurons to obtain the optimal prediction performance of the model.Aiming at the problem that the prediction performance of a single prediction model is limited by its own characteristics,an ARIMA-Bi＿LSTM hybrid prediction model is proposed.The model first uses Auto Regressive Integrated Moving Average(ARIMA)to predict the linear characteristics of the original sequence signal,and then predicts the nonlinear characteristics through the bidirectional Long Short Term Memory Network(Bi＿LSTM)model to predict the nonlinear characteristics,so that the model has strong linear and nonlinear prediction performance.The optimal hyperparameters of the ARIMA model and Bi＿LSTM model are obtained through Auto ARIMA and Bayesian optimization methods,respectively,to improve the prediction performance of the hybrid model.In order to verify the effectiveness of the proposed method,the experiment was conducted with NASA lithium-ion battery decay data.The original sequence data was first decomposed by EWT to extract low-frequency sub-layer signals as feature signals,and then the feature signals were predicted by the ARIMA-Bi＿LSTM mixed model.The predicted results and the high-frequency sub-layer signal are subjected to EWT inverse transformation to obtain the prediction results of the life decay signal.Through comparison experiments,the model is verified to have good prediction performance in predicting the life decay signal of the lithium ion battery.And by analyzing the end-of-life point,an accurate prediction result of lithium ion battery RUL is obtained.