铁道科学与工程学报2024，Vol.21Issue(12)：4899-4909,11.DOI:10.19713/j.cnki.43-1423/u.T20240291
基于TimeGAN-GRU的镍镉蓄电池RUL预测
TimeGAN-GRU method for RUL prediction of Ni-Cd battery
摘要
Abstract
Nickel-cadmium batteries are widely used as auxiliary power supply for high-speed trains in China,and their performance reliability is directly related to the safety of high-speed trains.The remaining service life of the battery refers to its performance from the current state of degradation to the failure of the length of time or the number of times it can be charged and discharged.It is an important indicator to characterize the performance of the battery.The current Ni-Cd battery life model for high-speed trains is limited by small sample data and has the problems of poor accuracy and generalization.Therefore,the battery degradation characteristics were extracted from the new Ni-Cd battery life experimental data,and the time-series adversarial generative network was used to enhance it so as to improve the scale and quality of the data.The enhancement effect was evaluated based on the classification scores,prediction scores,principal component analysis,and t-distributed stochastic neighborhood embedding analysis methods.Second,the prediction model of the remaining service life of Ni-Cd batteries for high-speed trains was established by using the enhanced data with the gated recurrent unit method.Finally,different prediction starting points were verified by the experimental data of cycle life of four-stage repair Ni-Cd batteries,and the prediction effects of the time-sequence adversarial generative network-gated cyclic unit fusion model,the gated cyclic unit model,and the long-and-short-term memory model were compared.The research results show that:for the data enhancement effect of time-series generation adversarial network,the distribution of real data and simulated data is similar,the average absolute error is small,and the simulated data quality is high;and the fusion model of time-series generation adversarial network with gated recurrent unit verified by the data of Ni-Cd battery repaired at the fourth level has higher generalization performance and prediction accuracy compared with the model of gated recurrent unit and the model of long-and-short-term memory.The research results for high-speed train Ni-Cd battery in the small sample data limitations could establish a better accuracy and generalization of the remaining life prediction model for high-speed trains to ensure the safety of high-speed trains and optimize the development of maintenance programs to provide a reference.关键词
蓄电池/镍镉蓄电池/剩余寿命预测/时序对抗生成网络/门控循环单元网络Key words
battery/nickel-cadmium battery/remaining useful life prediction/time-series generative adversarial network/gated recurrent unit分类
信息技术与安全科学引用本文复制引用
于天剑,杨雨萌,刘海涛,伍珣,代毅,向超群..基于TimeGAN-GRU的镍镉蓄电池RUL预测[J].铁道科学与工程学报,2024,21(12):4899-4909,11.基金项目
湖南省自然科学基金资助项目(2020JJ5757) （2020JJ5757）
