Research On Power Battery Life Prediction Method Based On Machine Learning

The state advocates energy conservation and carbon reduction,and promotes the orderly development of new energy.Lithium-ion batteries are widely used in electric vehicles,aircraft and aerospace because of their long cycle life and green environmental protection.At present,the life of lithiumion batteries gradually decreases with the increase of the number of charge and discharge cycles,and there are battery short circuit and explosion and other safety risks.The existing prediction methods are affected by the noise of battery experimental data and the defects of a single algorithm model,and the phenomenon of gradient disappearance and gradient explosion will occur,resulting in low accuracy of prediction methods and inaccurate results.Therefore,a data preprocessing method and an algorithm fusion of lithium-ion battery remaining useful life prediction method is proposed to efficiently solve the above problems.This paper studies the prediction method of the remaining useful life of lithium-ion batteries from the following aspects:(1)Data preprocessing method based on remaining useful life prediction of lithium-ion batteries.Aiming at the problems of high dimensionality and data redundancy of original data sets,a data preprocessing method for feature extraction and feature fitting of original data sets for the prediction of the remaining useful life of lithium-ion batteries was proposed.This method uses the 18650 lithium-ion battery test data set disclosed by NASA.According to the changes of various battery indicators during charging and discharging,the internal law of battery life attenuation is revealed.Grey correlation analysis is introduced to preprocess the data set and compare and verify it with other original data sets.Furthermore,a health index based on the fusion of isobaric drop time series and isobaric rise time series of discharge battery was proposed as the simplified data for the remaining useful life prediction.Three error evaluation methods were used to verify the effectiveness and superiority of the health index.The results show that the proposed data preprocessing method greatly improves the accuracy and computational efficiency of subsequent battery life prediction.(2)Research on the method of predicting the remaining useful life of lithium-ion batteries based on Convolutional Neural Networks(CNN).Based on the method proposed in Chapter 2,the pre-processed data was divided into training set and test set,and the battery capacity was selected as the target output,which was put into the CNN algorithm model to predict the remaining useful life of lithium-ion batteries.CNN uses three-layer convolution to learn and predict battery parameters,introduces BP neural network for comparison and verification,and designs experimental schemes under different conditions.The effectiveness and superiority of CNN are verified by example test and comparison with other algorithms.(3)Prediction Performance of remaining useful life of lithium-ion battery based on Bidirectional Gated Recurrent Unit(BiGRU).Based on the method proposed in Chapter 2,the preprocessed data is brought into the BiGRU algorithm model,and the model hyperparameters are determined by designing experimental schemes under different conditions.The historical data and future data are learned and predicted simultaneously.The superiority of BiGRU method in predicting the remaining useful life of lithium-ion battery is verified by comparing with other algorithms.(4)Prediction method of remaining useful life of lithium-ion battery based on CNN-BiGRU fusion.Aiming at the problem that the single prediction model of lithium-ion battery is easy to produce gradient disappearance in the iteration process,an optimization method based on CNN-BiGRU fusion was proposed to predict the remaining useful life of lithium-ion battery.Aiming at the problem of selecting algorithm model parameters,three experimental schemes under different conditions are designed.The CNN-BiGRU fusion model is formed through series of convolutional neural network and two oneway Gated Recurrent Unit(GRU)algorithm models in opposite directions,which further improves the accuracy and computational efficiency of battery life prediction.The prediction accuracy of CNN-BiGRU model is the highest under different conditions and compared with other algorithms,which verifies the superiority of CNN-BiGRU model in predicting the remaining useful life of lithium-ion batteries.In this paper,the prediction method of remaining useful life of lithium-ion battery is studied.The experimental data were preprocessed and the characteristic parameters were mined.A lifetime prediction method based on CNN-BiGRU fusion is proposed.This study provides an important method support for the prediction of the remaining useful life of lithium-ion batteries.