基于在线序列超限学习机和主成分分析的蒸汽冷却型燃料电池系统快速故障诊断方法

Fast Fault Diagnosis Method of Evaporatively Cooled Fuel Cell System Based on Online Sequential Extreme Learning Machine and Principal Component Analysis

为解决蒸汽冷却型燃料电池系统的故障诊断问题,该文提出基于在线序列超限学习机和主成分分析的蒸汽冷却型燃料电池系统快速故障诊断新方法。新方法采用主成分分析过滤冗余信息,得到能反映蒸汽冷却型燃料电池系统状态的故障特征向量;使用在线序列超限学习机对故障特征向量进行分类,能有效提高模型诊断正确率并降低运算时间。实例分析表明:新方法可快速识别膜干故障、氢气泄漏故障和正常状态共三种健康状态。算法的诊断正确率为99.67%,运算时间为0.296 9s。新方法的诊断正确率比SVM和BPNN分别高出14.34%和9.34%,在线序列超限学习机的运算时间仅为SVM和BPNN的1/1 011和1/132。因此,该文所提方法适用于大数据量样本和多数据维度下的蒸汽冷却型燃料电池系统在线故障诊断。

In order to solve the fault diagnosis problem of evaporatively cooled fuel cell system, a novel fast fault diagnosis method for evaporatively cooled fuel cell system was proposed based on online sequential extreme learning machine and principal component analysis. The principal component analysis was used to filter the redundant information and obtain the fault eigenvectors that could reflect the states of evaporatively cooled fuel cell system. The fault feature vectors were classified by online sequential extreme learning machine, thereby effectively improving the diagnostic accuracy of the model and reducing the computation time. The case analysis shows that the novel method can quickly identify three healthy states of membrane drying failure, hydrogen leakage failure, and normal state. The diagnostic accuracy of the algorithm is 99.67% and the computation time is 0.296 9 seconds. The diagnostic accuracy of the novel method is 14.34% and 9.34% higher than those of SVM and BPNN respectively, and the computation time of online sequential extreme learning machine is only 1/1 011 of SVM and 1/132 of BPNN. Therefore, the proposed method is suitable for online fault diagnosis of evaporatively cooled fuel cell system with large data samples and multi-data dimensions.