摘要
为了更准确地预测燃料电池性能的衰减情况,提出Wavelet-Elman-LSTM算法,通过小波分解实现基准电压信号的分解,利用Elman神经网络实现高频分量的预测,采用长短期记忆(LSTM)神经网络实现低频分量的趋势预测,从而实现更精准的性能衰减预测。通过稳定工况(恒流工况)和极端工况(启停加速老化、怠速加速老化)下3组不同的衰减数据集,从短期预测和长期预测2个维度进行Wavelet-Elman-LSTM神经网络与Elman、LSTM神经网络的性能衰减预测效果的横向比较,结果证明,所提出的Wavelet-Elman-LSTM神经网络在稳定工况与极端工况下均具有显著优越性,特别是在恒流工况下,短期预测、长期预测的均方根误差均大幅降低。
In order to more accurately predict the degradation of proton exchange membrane fuel cell performance,this paper proposed the Wavelet-Elman-LSTM prediction algorithm,which decomposed the reference voltage signal through the wavelet decomposition,used the Elman neural network to predict the high-frequency component,and used the LSTM to predict the trend of the low-frequency component,so as to make more accurate performance degradation prediction.Through 3 sets of different degradation data sets under stable constant current condition and extreme condition(startupshutdown accelerated aging test and idle speed accelerated aging test),the performance degradation forecast accuracy of Wavelet-Elman-LSTM neural network,Elman neural network and LSTM neural network was compared from two dimensions of short-term prediction and long-term prediction.The results show that the proposed Wavelet-Elman-LSTM neural network has significant advantages under both stable condition and extreme condition.Especially under stable constant current condition,the Root Mean Square Error(RMSE)of short-term and long-term prediction can be reduced significantly through the new neural network.
作者
丛铭
马天才
王凯
姚乃元
Cong Ming;Ma Tiancai;Wang Kai;Yao Naiyuan(Tongji University,Shanghai 201804)
出处
《汽车技术》
CSCD
北大核心
2022年第9期23-29,共7页
Automobile Technology
关键词
质子交换膜燃料电池
耐久性
小波分解
ELMAN神经网络
长短期记忆
Proton Exchange Membrane Fuel Cell(PEMFC)
Durability
Wavelet decomposition
Elman neural network
Long Short-Term Memory(LSTM)