SOH Estimation Of Lithium Battery Based On Time Convolution Memory Network

With the gradual improvement of China’s requirements for vehicle emission standards and the continuous development of new technologies,people began to vigorously develop electric vehicles.As the power source of electric vehicle,the power battery is a research object of great concern in the development process of electric vehicle.At the same time,the research work of battery management system(BMS)has also been paid attention to.State of charge(SOC)and state of Health(SOH)analysis of battery are important contents of battery state analysis in BMS,among which the SOH of battery represents the ability of the maximum battery capacity that can be stored currently,which is very important for the whole battery life cycle.Accurate estimation of SOH can know whether the battery needs to be replaced at present,so as to ensure the normal use of electrical equipment and maximize the battery utilization.Firstly,this thesis introduces the related characteristics of lithium battery,the necessity of research on the health status of lithium battery and the commonly used estimation methods of battery SOH at home and abroad.Secondly,the thesis describes the process and results of the estimation methods based on BP neural network model and two improved methods based on the model: combining with grey model and battery equivalent circuit simulation model.Then,by comparing and analyzing the estimation results and advantages and disadvantages of the existing methods,this thesis proposes a time convolution memory network model for battery health state estimation.Time convolution memory network is a model which combines the two algorithms of CNN(convolution neural network)and LSTM(short-term memory network).It can well extract and learn the feature relationship of battery cycle charge and discharge process in a long time span.Finally,the input and output datasets are sorted out from the experimental battery datasets,which are divided into training datasets,test datasets and verification datasets to train and test the time convolution memory network model.In order to improve the prediction accuracy,this thesis selects exponential smoothing for data denoising and data cutting for data augmentation to improve the estimation accuracy.In the experimental part,the improved algorithm model estimation results are compared with the estimation result the BP neural network model,the gray neural network model,and the battery equivalent circuit model.The results show that the proposed method based on time convolution memory network model is more accurate,MSE is 0.0076.