SOH Estimation Of Lithium-ion Battery Based On One-dimensional Sequential Convolutional Neural Network

When the lithium-ion battery is used to a certain degree of aging,it will no longer be able to continue to use.If it is used,unpredictable faults and dangers will occur.Therefore,the aged lithium-ion battery should be found and replaced in time.The state of Health(SOH)of lithium-ion battery provides the main reference for replacing aging batteries.Therefore,it is of great significance to estimate the optimal performance and safe operation of lithium-ion SOH in real time and accurately.In this paper,a one-dimensional sequential convolutional neural network is designed to estimate SOH of lithium battery.The main research contents are as follows:Firstly,the aging experiment of lithium-ion battery is analyzed systematically.The experimental data are visually analyzed by Python language.It is found that the voltage,current and temperature curves of lithium battery are time series changing with time during a constant current constant voltage charging process.Each charge corresponds to the current SOH value.The one-dimensional convolution neural network is expert in handling time sequence The task of the column is as follows: the network model is one-dimensional sequential convolutional neural network,which defines SOH in capacity mode.The voltage,current and temperature data of the battery in the process of constant current constant voltage full charge are taken as model inputs,and the current capacity of lithium-ion battery is the output of the model.The SOH estimation process is designed,and layers in the kernas high-level API are used in tensorflow.The class and functions of losses,sequential and optimizers are built.A one-dimensional convolutional neural network consisting of 6-layer convolution layer,2-layer full connection layer and 3-layer pool layer is built.Dropout layer is added to the full connection layer to prevent over fitting.In network training,MSE is used as the target function,Adam is selected as gradient optimizer,batch batch batch batch batch batch is 128,and training samples and test are used as the target function The proportion of the samples is divided into 8 to 2,the iteration step is 100times,after 30 epochs,the objective function will not change,the training set and the verification set are consistent,and the network training will be completed.The one-dimensional convolutional neural network model was verified by using NASA lithium ion data set.The error(MAE)of the lithium-ion battery was 1.04%,1.18%,1.41%and 1.32% respectively in the test No.RW13,RW17,RW21 and RW25,and the training parameters were 137381.The results show that the one-dimensional sequential convolutional neural network designed in this paper has high accuracy in the estimation of SOH of lithium battery.In addition,the proposed one-dimensional sequential convolution neural network has high accuracy in the estimation of SOH of lithium battery.In addition,the results show that the one-Compared with the results of the total connected neural network,the one-dimensional convolutional neural network model designed in this paper has the advantages of less training parameters and less memory.Finally,the influence of network input,network structure and data enhancement on network model is analyzed.The experimental results show that when voltage,current and temperature are taken as input,the estimation accuracy is not high;the number of convolution layer and pool layer has obvious influence on network performance;the network model without data enhancement has no data enhancement The network model is robust and the data enhanced network model can better deal with the noise impact.