Research On SOH Prediction Of Lithium Battery Based On LSTM Neural Network

For electric vehicles,State of Health estimation method of power lithium battery is a very important aspect of battery management system.Accurately estimating the aging state of lithium batteries and predicting the remaining battery life is of great significance for the stable and safe operation of electric vehicles.Based on the idea of data-driven method,this paper analyzes the historical data of lithium ion battery life,and establishes the health status evaluation model of lithium battery through deep learning method.In the selection of the model,the long-short-term-memory reccurent neural network(LSTM RNN)was used to learn the battery life attenuation process,and the lithium battery SOH evaluation model was generalized through transfer learning.Firstly,this paper introduces the research background and significance of SOH estimation of power battery,discusses the SOH estimation methods at home and abroad in detail,divides them into three categories and conducts in-depth research and discussion respectively.On this basis,the advantages and disadvantages of model-based estimation method and data-driven method are analyzed,and the main research contents of this topic are proposed.Secondly,this paper analyzes the working characteristics of lithium ion battery and has a deep understanding of the research object.Firstly,the working principle of lithium ion battery is introduced.Then,the main technical indicators of the battery,such as capacity,internal resistance,voltage,charge and discharge rate,cycle life,are described.Then,with samsung ternary lithium batteries as experimental object,based on the new will NEWARE BTS2000 high-performance battery test equipment,structures,battery test platform,the battery test platform based on this,the capacity,internal resistance testing of target cells,and impulse response test,self-discharge storage test and storage features research work;Finally,the aging of battery performance is systematically analyzed based on the experimental data.Then,the single SOH evaluation model of lithium battery.First,the LSTM neural network model is introduced,and its principle and characteristics are described.Then,the performance degradation factor was extracted from the single lithium battery data of the NASAPCo E research center in the United States,and its solution was formed into the time series data and input into the LSTM neural network model to produce the evaluation results of the health status of lithium batteries.Finally,in order to explore and solve the generalization problem of different types of lithium battery SOH evaluation models,the deep neural network transfer learning method was adopted to transfer the model applied to NASA lithium battery data to CALCE battery data,so that the same model could be reused efficiently and the generalization ability of data-driven model was enhanced.