中文摘要：

针对传统的K均值聚类方法无法解决故障特征数据维数高，在故障样本交叠严重时多分类性能较差的问题，在电路故障特征数据预处理阶段，提出了一种类互均衡核聚类预处理方法，不仅克服了传统方法中分配不均或漏分问题，而且解决了特征数据维数高带来的奇异性问题，有效地提高了故障样本交叠时的多分类聚类性能．在此基础上，设计了一种用于模糊支持向量机的核密度函数，实现多故障的分类．将该方法应用于国际标准电路中的CTSV（continuous-time state-variable）滤波器电路故障诊断中．结果表明，该方法能突出不同故障的特性，具有很好的故障识别率．

英文摘要：

Traditional K-means clustering methods cannot deal with the high dimensional fault characteristics data and their classification performance is rather poor when fault samples overlap heavily. To overcome this disadvantage, a cross- balanced kernel cluster preprocessing method is proposed for preprocessing stage of circuit fault characteristic data, which not only overcomes the issues of uneven distribution or data leakage points in traditional methods, but also solves the singularity problem introduced by high dimensional feature data and improves the performance of the multi-classification when the overlapping of fault samples happens. Finally, based on the above work, a kernel density function for fuzzy support vector machine is designed to clarify multiple faults. In addition, the method is applied to the fault diagnosis of CTSV （continuous- time state-variable）, which is an international standard filter circuit. The results show that the proposed method enables a good recognition rate of different failures.