Research On State Of Health Estimation Of Lithium-ion Battery Based On Wavelet Long Short-term Memory Neural Network

As a clean technology to solve carbon emissions,electric vehicles have been diffusely used in modern vehicles.Lithium-ion batteries have become the main energy storage equipment of electric vehicles because of its light weight,long life,high energy density and low self discharge.Therefore,the evaluation of the state of health(SOH)of lithium battery is significant for the long-term use of electric vehicles.In order to predict the SOH of the battery effectively and accurately,the research on the SOH estimation of lithium-ion battery has come to a hot-button topic.For the SOH prediction algorithm,the feature extraction of input data is of great important for the accuracy of the SOH estimation.Therefore,this paper firstly introduces two kinds of feature extraction methods,and then designs a novel neural network based on wavelet neural network(WNN)and long short-term neural network(LSTM)to predict SOH.Finally,the extended Kalman filter is integrated with the designed neural network to realize the real-time estimation of the SOH of the battery.The primary research contents and innovations of this paper are as follows:(1)Two feature extraction methods were designed.The first feature extraction method is built on constant current charging curve.It extracts the time cost of constant current charging of the battery as the input feature.Compared with other manual feature extraction methods,this method can be implemented simply and easily.The second method is based on convolutional neural network(CNN).It takes advantage of the ability of CNN to extract the spatial features of the data matrix,which can automatically extract the required data features only by inputting the original battery charging data.(2)The prediction algorithm for the SOH of battery of Wavelet neural network-Wavelet Long Short-Term Memory neural network(WNN-WLSTM)is proposed,which is based on WNN and LSTM.The input data is extracted from the constant current charging curve.The experimental simulation shows that the designed method has prominent learning and prediction capacity.(3)The prediction algorithm for the SOH of battery of Convolutional neural network-Wavelet neural network-Wavelet Long Short-Term Memory neural network(CNN-WNN-WLSTM)is designed.This method integrates CNN and WNN-WLSTM,and uses the original battery charging data as the input data.The preeminent prediction performance of the proposed method is confirmed by the experimental simulation.(4)An online real-time estimation algorithm for the SOH of battery of Dual Extended Kalman Filter-Wavelet neural network and Wavelet Long Short-Term Memory neural network(DEKF-WNN-WLSTM)is designed.Based on the WNN-WLSTM,the Extended Kalman Filter(EKF)is integrated and the segment data are used as the input to realize the real-time estimation of the SOH of battery.The effectiveness of this algorithm in real-time SOH prediction is verified by experimental comparison,and the prediction accuracy is improved.