兼顾后续分类器的小波特征选择及在模拟电路故障诊断中的应用

Wavelet based feature selection technique considering subsequent classifier and its application in analog circuit diagnosis

针对模拟电路故障诊断中的特征选择问题,提出了一种兼顾后续支持向量机分类器的小波特征选择技术。新方法利用核函数把特征的评估标准映射至高维空间进行计算,从而在特征的评估和后续支持向量机分类器之间建立联系,并在特征选择阶段解决了核函数的参数设计问题。另外,新方法考虑了所需故障分类器的结构信息,因此选择的小波特征更适合后续分类器使用,从而能够提高模拟电路故障的诊断精度。仿真和实际电路的实验结果均验证了所提出技术的有效性和正确性。

Focusing on the feature selection problem in analog circuit diagnosis, a new wavelet based feature selection method considering subsequent support vector machines classifier （SVC） is presented. This proposed method maps the feature evaluation rule into a high-dimensional space via kernel function, where relevant computations are implemented. Hence, the relationship between the evaluation of individual feature and subsequent classifier is established, and the selection problem of kernel function parameter can be solved in this process. The novel method considers the structure information of the needed SVC, and therefore, the selected features are suitable for the subsequent classifier and thus, subsequent fault detection and diagnosis efficiency is improved. Both simulation and actual circuit experiment prove the effectiveness and correctness of the proposed technique.