Research On Prediction Of Lithium-ion Battery Life Based On Neural Network Model

The lithium-ion cycle life prediction is always a key issue in the battery health management systems,so it is of great significance to find a fast and accurate prediction method for devices using lithium-ion batteries as power supply unit.For the prediction of the cycle life of lithium-ion batteries,many researches have been carried out at home and abroad nowadays,which mainly based on physical models and data-driven methods.The lithium-ion battery cycle life prediction with particle filter depends on the physical or empirical model.However,in observation equation based on model,the adaptability and accuracy for individual battery under different operating conditions are not fully considered.At the same time,because of the complexity of the electrochemical reaction inside the battery,it is difficult to measure the value of each parameters accurately through the sensor,which limits the development of the physical model.In addation,because the traditional data-griven method does not consider the connection between the data before and after,the prediction results of the lithium-ion battery cycle life is often not accurate.Therefore,this paper studies the cycle life prediction of lithium-ion batteries based on the neural network model.The main contents are as follows:Firstly,based on the in-depth analysis of the working principle and basic characteristics of lithium-ion batteries,many factors affecting the capacity decay of lithium-ion batteries are determined and confirmed that the discharge temperature and discharge voltage are the main factors.Besides,it also states that the actual capacity of the battery is an important parameter to characterize the health of the battery.Secondly,BP network and NAR network are used as the representatives of static neural network and dynamic neural network to predict the lithium-ion battery remaining useful life and compare the prediction results.The prediction results show that dynamic neural networks have batter prediction effects than static neural networks for time series data.Aiming at the problems of no memory,poor prediction results and bad adaptability to data in shallow neural network,this paper proposes a method for predicting the remaining cycle life of lithium-ion battery based on the improved LSTM model.The LSTM model is used to learn the loss trajectory of lithium-ion battery capacity,and can predict the remaining cycle life of lithium-ion batteries at different prediction starting points accurately.Then,in order to verify the accuracy of the LSTM prediction algorithm,compare it with the traditional data-driven prediction algorithm.The prediction results of the model show that the LSTM model not only predicts accurate in the early stage of battery degradation,but also gets small error in the middle and late stage of battery degradation at the same experimental conditions and prediction starting point,which verifies the superiority of the model againFinally,this paper designs and implements a residual cycle life prediction software for lithium-ion batteries.Based on the historical capacity data of lithium-ion batteries,the software accurately predicts and verifies the remaining cycle life of lithium-ion batteries by dynamically selecting different network paprmeters.