多分类SVDD混叠域识别的模拟电路故障诊断

An Approach to Discriminate Overlap Region of Multi-class Classification SVDD for Analog Circuits Fault Diagnosis

摘要:

JP+1]针对多分类支持向量域数据描述(SVDD)方法中混叠样本诊断精度差的问题,提出了一种带异类样本的多分类SVDD算法；该方法在普通SVDD超球模型基础上,对于存在混叠区域的类别,以该类所有样本为目标类,其他类与之混叠的样本为异类,利用带异类样本的SVDD算法重新训练,直至所有超球优化完毕；仿真实验验证了文章算法消除混叠和提高精度的能力,并将该算法应用于模拟电路故障诊断中；相较与SVDD多分类算法、一对一和一对多SVM算法,文章方法在模拟电路故障诊断中具有更高的诊断精度。

Abstract:

JP+1]To improve the discrimination accuracy of conventional multi-class classification support vector data description (SVDD) methods, a multiple classification Support vector data description algorithm with Negative Samples is proposed. Based on the general model of SVDD, the proposed algorithm treats the samples in the class as the target class, while the other classes of overlap and sample is heterogeneous for the overlap region. By using SVDD algorithm with Negative samples, the hypersphere model is trained again until all hypersphere models optimized. Simulated experimental results show that the proposed algorithm can eliminate overlap and improve the discrimination accuracy. The algorithm is applied in the implemention of analog circuits fault diagnosis, comparing with SVDD classification algorithm, one-to-one and one-to-many SVM algorithm, results show that the algorithm is more effective and higher accuracy in fault diagnosis.