基于CNN-BiLSTM双通道特征融合的PEMFC水淹故障识别方法  

PEMFC FLOODING FAULT IDENTIFICATION METHOD BASED ON CNN-BILSTM DUAL-CHANNEL FEATURE FUSION

作　　者：赵旭阳 袁裕鹏[1,3] 童亮[1,3] 朱小芳 李骁 Zhao Xuyang;Yuan Yupeng;Tong Liang;Zhu Xiaofang;Li Xiao(State Key Laboratory of Maritime Technology and Safety,Wuhan University of Technology,Wuhan 430063,China;School of Naval Architecture,Ocean and Energy Power Engineering,Wuhan University of Technology,Wuhan 430063,China;Reliability Engineering Institute,School of Transportation and Logistics Engineering,Wuhan University of Technology,Wuhan 430063,China;Wuhan Troowin Power System Technology Co.,Ltd.,Wuhan 430056,China)

机构地区：[1]武汉理工大学水路交通控制全国重点实验室,武汉430063 [2]武汉理工大学船海与能源动力工程学院,武汉430063 [3]武汉理工大学交通与物流工程学院可靠性与新能源研究所,武汉430063 [4]武汉众宇动力系统科技有限公司,武汉430056

出　　处：《太阳能学报》2025年第4期30-38,共9页Acta Energiae Solaris Sinica

基　　金：国家重点研发计划(2021YFB2601601)。

摘　　要：为及时准确地识别质子交换膜燃料电池(PEMFC)水淹故障,提出基于卷积神经网络(CNN)和双向长短期记忆网络(BiLSTM)双通道特征融合的PEMFC水淹故障识别方法。首先,采用归一化消除原始特征之间的量纲;在此基础上利用随机森林(RF)评估数据的特征重要性进行特征筛选;采用并联式结构将CNN与BiLSTM结合分别提取空间特征和时间特征并进行串联融合;最后利用支持向量机(SVM)进行水淹故障识别。实例分析表明,所提方法可快速准确地识别PEMFC的正常状态和水淹故障,总体分类准确率为99.08%,测试用时为0.0929 s,可有效提高故障分类的准确率。In order to identify the flooding fault of proton exchange membrane fuel cell(PEMFC)in a timely and accurate manner,a PEMFC flooding fault identification method based on the dual-channel feature fusion of convolutional neural network(CNN)and bidirectional long short-term memory network(BiLSTM)is proposed.Firstly,normalization is used to eliminate the effect of dimension of the original features.On this basis,random forest(RF)is used to evaluate the feature importance of the data.The CNN and the BiLSTM are used to extract spatial features and temporal features,respectively.Then,these features are fused in series.Finally,the support vector machine(SVM)is used to identify the flooding fault.The case analysis shows that the proposed method can quickly and accurately identify the normal state and the flooding fault state of PEMFC,with an overall classification accuracy of 99.08%and a test time of 0.0929 s,which can effectively improve the accuracy of fault classification.

关 键 词：质子交换膜燃料电池 故障诊断 卷积神经网络 长短时记忆网络 随机森林 支持向量机 

分 类 号：TM911.4[电气工程—电力电子与电力传动]