Remaining Life Prediction Of Lithium Battery Based On Deep Learning

Lithium-ion batteries are widely used and their safety should not be ignored.Because the aging of batteries caused by cyclic charging-discharge process may bring safety risks,it is very important to accurately predict the status of health(SOH)and remaining useful life(RUL)of lithium batteries.This paper studied the published NASA lithium-ion battery data set,proposed multi-scale health indicator selection strategy and deep learning method,and carried out research on the remaining life prediction of lithium batteries.The main contents are as follows:Firstly,relevant literatures at home and abroad were studied,the research methods and health factor selection methods were summarized,and a multi-scale health factor selection strategy based on discharge characteristics was proposed.Based on the time-domain expression transformed by Laplace transform from the equivalent circuit model formula of lithium battery and combined with the actual terminal voltage discharge curve,the parameters of the battery internal resistance,polarization internal resistance and polarization capacitance were identified and taken as health indicators.After analyzing the relationship between the time of equal discharging voltage and the selection of voltage drop interval,in order to select the best time interval as the health indicator,the elite genetic algorithm using Pearson correlation function value as fitness was used to select the the time of equal discharging voltage with the highest correlation with the actual capacity of the battery as the health indicator.Then,in order to fit the time series relationship between the health factor and the actual capacity of the battery and the capacity regeneration phenomenon,a SOH prediction model was established on the pytorch platform using the short and long duration memory neural network(LSTM).In order to further improve the prediction accuracy,the PSO algorithm was used to configure the neural network superparameters to improve the LSTM network,so that the PSO-LSTM neural network could learn the nonlinear relationship between the health factor and the actual capacity,and establish the data-driven SOH prediction model.By comparing the influence of different health factor selection strategies and the neural network on the prediction results,To verify the superiority of the proposed health factor selection strategy and LSTM as a prediction model,and prepare for the following lithium battery RUL prediction.Finally,a lithium battery degradation model was established based on deep learning to predict the change trend of health indicators.The future health indicators were estimated by the change trend of health indicators in the historical data,and the predicted health factor sequence was input into the SOH prediction model to predict the end of battery life(EOL)and RUL.On the basis of the above RUL prediction method,in order to avoid obtaining the difficult the time interval of equal discharging voltage as a health indicator,the equivalent circuit model of lithium battery was established by MATLAB Simulink and Simscape module,and the function relationship between the open circuit voltage and the battery SOC was estimated by the least square method with the identified parameters.Using the combination of the open-circuit voltage and the identified internal resistance of the battery to estimate the equal voltage drop and discharge as an alternative health indicator,the practicability of the RUL prediction algorithm is greatly improved.