Remaining Useful Life Prediction Of Lithium-ion Battery Based On Deep Learning

In the practical application of lithium-ion batteries,the safety of the batteries themselves is very important.If a battery failure occurs during use,it may cause the performance or failure of the corresponding power equipment or system,thereby increasing costs and even causing personal injury or death.Therefore,it is of great practical significance to find a method that can accurately predict the remaining life of lithium-ion batteries.At present,there are many researches on the life prediction of lithium ion in related fields.From the research methods,the methods of life prediction for lithium ion batteries are roughly divided into two categories: model-based methods and data-driven methods.The model-based method mainly models the battery and simulates the behavior of the battery.However,this method often relies too much on experience and the model generalization is not strong.The traditional data-driven method is not good enough for prediction accuracy.In order to further study this problem,this paper studies the public lithium-ion battery data set NASA PCo E data set.In order to make the life prediction model perform better in this data set,a deep learning method is proposed to study the remaining life prediction of the lithium battery.The contents are as follows:First of all,the relevant literature at home and abroad is studied,the current status of domestic and foreign research on lithium ion battery life prediction is summarized,and the NASA PCo E public data set is used to study the charge and discharge parameters of lithium ion batteries in detail.There is a lot of redundant data,so it is necessary to reduce the dimensionality of the original data set to remove the redundant information in the data set.Through observation,it is found that during the charging and discharging process,the curve of each parameter with time can be selected to replace the curve with appropriate parameters.By extracting these parameters,the original data set is reduced in dimensionality.This article is written for different characteristics of different parameter curves the program performs dimensionality reduction on each parameter and performs data preprocessing operations to enable the new data set to be used as input to the neural network.In this way,the input data can retain most of the properties of the original data set,and there will be no situation where the input of redundant data into the neural network will cause the deterioration of the final life prediction result.Secondly,after understanding the existing research methods,a convolutional neural network(CNN)model is constructed to predict the life of lithium-ion batteries.This method uses the convolutional layer in the convolutional neural network to learn the parameters in the battery data set and input the learned advanced features to the output layer of the neural network to calculate the remaining life.The final prediction result shows that the prediction result of the convolutional neural network is better than other machine learning life prediction methods,and the accuracy of life prediction of other methods is quite different from that of this method,that is to say,the convolutional neural network is in terms of battery life prediction has a certain degree of excellence.Finally,considering the time series relationship of the internal data of the lithium battery data set,the long-short-term memory network(LSTM)method is used to predict the life of the lithium battery,and the LSTM network is improved,and a double-layer long-short time is proposed.The memory network(Bilayer-LSTM,B-LSTM)method performs life prediction on the lithium battery data set.This neural network can combine historical input data to help the neural network perform better calculations on the results.The prediction results show that the B-LSTM model has the best prediction effect after using the time series relationship within the data set,and the model performs best under different conditions,which verifies the battery life of the Superiority in forecasting of B-LSTM model.