In the fault diagnosis of a hydro-turbine generating unit (HGU), kernel clustering is a valid non-supervised learning method. In order to solve the problems of kernel parameter selection and cluster center calculation, a novel electromagnetism-like artificial bee colony weighted kernel clustering (EAWKC) is proposed. First, after considering the influence of different symptoms, the data is weighted, and the clustering model is built based on the kernel Xie-Beni clustering index. Then, the electromagnetism-like artificial bee colony (ELABC) method is proposed and introduced in order to solve the objective function to realize the synchronized optimization of the clustering center, symptom weight and kernel parameter. The classification accuracy of EAWKC is checked by three of the UCI testing data sets and the HGU fault samples, and compared with the traditional method. The experimental results show that EAWKC has higher accuracy and can effectively complete the fault diagnosis.