中文摘要：

水电机组故障大多以振动的形式表现出来，其成因非常复杂，机组故障和振源缺乏非此即彼和一一对应的显明关系，故障样本存在交叠区域。常规水电机组故障诊断分类器没有考虑分类问题的交叠区域，而处在交叠区域的样本的分类难度大，且含有不确定信息。针对这一问题，提出一种新的粗糙集与一对一多类支持向量机结合的诊断方法，该方法利用粗糙集基本理论对支持向量机的分类结果进行描述，对于可能存在样本交叠的分类间隔区域进行进一步处理。诊断分两阶段进行，第一阶段采用粗糙集理论将故障样本划分为某类的上近似集、下近似集，属于下近似集的样本认为被确切分类。第二阶段，对于属于上近似集而不属于下近似集的样本，采用一种基于距离的可信度来判断其属于某一类的程度。此分类方法考虑了水电机组故障模式中可能存在样本交叠的间隔区域，经数值试验和工程应用检验，更符合实际情况，且对于其他电力设备的故障诊断具有借鉴意义。

英文摘要：

Faults of hydroelectric generator unit （HGU） mostly manifest in the form of vibration, and its genesis is very complex. The map between faults and their symptoms is not one-to-one. There exists overlapping patterns in fault diagnosis for HGU. It is difficult to classify the overlapping patterns which maybe take on some interesting characteristics. However, the traditional classifiers of fault diagnosis for HGU do not consider overlapping patterns. For the above reason, a new hybrid method of rough set and 1-v-I multi-class SVM is proposed. The proposed method conducts in two phases. Firstly, it adopts rough set theory to describe the classification result of support vector machine. The fault samples are categorized as the lower approximation and upper approximation of some class. The samples belonging to the lower approximation are regarded as definitely classified. Secondly, for the samples belonging to the upper approximation and not belonging to the lower approximation, a reliability based on some distances is applied to describe to what extent the sample belonging to one class. At last the proposed method is successfully applied in diagnosing the vibrant faults of an HGU. The results show that the proposed classifier is suitable for fault diagnosis of HGU. And it offers references for fault diagnosis of other power equipments.