中文摘要：

核聚类是一类有效的水力发电机组振动故障诊断方法，为了解决核聚类有效性评价和核参数选择的问题，提出了一种引力搜索核聚类算法。首先建立以核Xie-Beni指标为目标的聚类模型；然后引入引力搜索框架，以聚类中心和核函数参数为优化变量，通过引力搜索求解核聚类模型；最后定义了基于核空间样本相似度的故障诊断模型。利用国际标准样本集对该方法进行分类测试，并将该方法应用于水电机组振动故障诊断。试验结果表明：与传统聚类方法相比，文中方法具有更高分类精度，且能对故障样本准确聚类并提取诊断模型参数，实现故障的准确诊断。

英文摘要：

Kernel clustering is a kind of valid methods for vibration fault diagnosis of hydro-turbine generating unit （HGU）. In order to solve the problem of evaluating clustering results and selecting parameter of kernel function, a novel gravitational search based kernel clustering （OSKC） was proposed. At first, the kernel clustering objective function was built based on kernel Xie-Beni clustering index, then the gravitational search method was introduced and applied to solve the objective function, while the clustering center and parameter of kernel function were encoded as optimization variables together; in this end the fault diagnosis model based on similarity was defined. UCI testing data sets were used to check the classification accuracy, and then CSKC was applied in fault diagnosis of HGU. Experimental results show that GSKC was more accurate in classification than traditional methods, meanwhile GSKC was able to cluster the fault samples of HGU effectively, and diagnosis different kinds of fault accurately.