【摘要】 为解决质子交换膜燃料电池电堆的故障分类问题，提出一种基于模糊C均值聚类和概率神经网络的故障诊断新方法。首先基于修正后的燃料电池电堆Fouquet等效电路模型，并结合电堆阻抗谱实验数据，得到电堆的正常、水淹、膜干和氧饥饿4种工作状态与电路模型参数的对应关系，进而提取合适的故障特征量作为聚类算法的特征输入。然后，利用模糊C均值聚类算法对故障样本进行聚类，形成标准聚类中心，并在此基础上，采用概率神经网络算法对故障样本实现多故障分类，有效剔除奇异数据并提高模型分类的正确率。最后，对200组实验数据进行实例分析，并与支持向量机和K最邻近方法进行对比，结果表明所提方法能对4种电堆工作状态进行快速识别，分类准确率达98.33%，验证了所提算法的有效性。更多还原

【Abstract】 To solve the problem of fault classification of proton exchange membrane fuel cells stack,one new fault diagnosis method based on fuzzy C means clustering and probabilistic neural network is proposed in this paper. Firstly,this paper is based on the modified Fouquet equivalent circuit model of the fuel cells stack and combines the experimental data of the stack EIS. The corresponding relationship between the four working states of the stack,namely normal,flooding,membrane drying and oxygen starvation,and the circuit model parameters is obtained. The appropriate fault characteristic quantity is extracted as the feature input of the clustering algorithm. Then,the paper uses the fuzzy C means clustering algorithm to cluster the fault sample data to form a standard clustering center. The probabilistic neural network algorithm is used to achieve multi-fault classification for the fault samples on this basis,which can effectively eliminate the singular data and improve the accuracy of the fault classification. Finally,the paper analyzes 200 sets of experimental data,and compares it with the support vector machine and the K-nearest neighbor method. The analysis results show that the method proposed in the paper can quickly identify the four working states of the stack,and the classification accuracy rate reaches98.33%,which verifies the effectiveness of the proposed algorithm.更多还原