基于KPCA和KFCM集成的汽轮机故障诊断

Steam turbine fault diagnosis based on KPCA and KFCM ensemble

为了提高汽轮机故障诊断的正确率和鲁棒性,提出一种基于核主元分析(KPCA)和模糊核聚类(KFCM)集成的汽轮机故障诊断方法。该方法针对汽轮机故障数据高维非线性的特点,采用核主元分析方法对故障数据进行特征提取,提高了神经网络的学习效率和精确度。然后由训练样本集相互独立地训练出多个神经网络,对其编号并按顺序将网络输出构成输出矩阵,再采用模糊核聚类算法对输出矩阵进行分析并对所有个体网络归类,计算所有类别中每个网络在独立验证样本集上的泛化误差,最后分别选取每个类别中泛化误差最小的个体网络作为这一类的代表进行相对多数投票法集成。实验结果表明,即使在学习样本较少的情况下,该方法也能取得较好的故障诊断效果。

To improve the accuracy and robustness,the fault diagnosis based on KPCA（Kernel Principal Components Analysis） and KFCM（Kernel Fuzzy C-Means Clustering） ensemble is proposed for steam turbine.In order to improve the learning efficiency and accuracy of neural network,KPCA is used to extract the main features of steam turbine fault data,which has high-dimensional and non-linear characteristics.With the same training sample set,several neural networks are trained independently and then numbered.According to their number,the network outputs form an output matrix,which is then analyzed by KFCM algorithm to classify all neural networks.The generalization error of individual neural network is calculated with the independent validation set and the neural network with the smallest error in each category is selected as the representative of that category,which is combined together by the majority voting.The experiments show that the proposed approach obtains a better diagnosis result even with small learning sample set.