基于GA-ELM算法的燃料电池性能预测模型  

Fuel cell performance prediction model based on GA-ELM algorithm

作　　者：刘智宇 郝冬 张妍懿 侯永平[1] LIU Zhi-yu;HAO Dong;ZHANG Yan-yi;HOU Yong-ping(School of Automotive Studies,Tongji University,Shanghai 201804,China;CATARC New Energy Vehicle Test Center(Tianjin)Co.,Ltd.,Tianjin 300300,China)

机构地区：[1]同济大学汽车学院,上海201804 [2]中汽研新能源汽车检验中心(天津)有限公司,天津300300

出　　处：《电池》2023年第3期243-247,共5页Battery Bimonthly

基　　金：国家重点研发计划(2021YFB4001005)。

摘　　要：为减少燃料电池堆耐久性能试验物料成本,提升耐久性能预测效率,利用燃料电池堆稳态耐久性试验数据,基于遗传算法(GA)-极限学习机(ELM)算法结合神经网络与遗传算法,搭建燃料电池稳态耐久性能预测模型。该模型为双输入[时间和电流(或电流密度)]单输出(电压)。利用试验数据对建立的模型进行训练与验证,发现该模型具有较高的预测精度。将GA-ELM模型与长短记忆网络(LSTM)模型对比,在预测精度(误差2%左右)相当的情况下,GA-ELM模型计算时间仅为LSTM模型的1/5。搭建的预测模型具有较好的通用性、较高的稳定性和精度。In order to reduce the material cost of fuel cell reactor durability test and improve the efficiency of durability prediction,a fuel cell steady-state durability prediction model was built based on genetic algorithm(GA)-extreme learning machine(ELM)algorithm combined with neural network and genetic algorithm using the steady-state durability test data of fuel cell stack.The model had two inputs[time and current(or current density)]and one output(voltage).The model was trained and verified by the experimental data,it was found that the model had high prediction accuracy.Comparing GA-ELM model with long and short memory network(LSTM)model,the calculation time of GA-ELM model was only 1/5 of that of LSTM model when the prediction accuracy(error was about 2%)was equivalent.The prediction model built had good universality,high stability and accuracy.

关 键 词：燃料电池堆 稳态性能预测 极限学习机(ELM)神经网络 

分 类 号：TM911.4[电气工程—电力电子与电力传动]