CAJ | 학술논문

文章将向量投影法应用在变压器故障诊断中，解决了如何构建有效SVM 层次的问题。按照类与类样本集之间的相交情况，利用欧氏距离和径向基函数计算类与类的空间距离和类间可分性测度，根据可分性测度进行排序，设计比较合理的层次结构进行分类。这种方法建立的故障诊断模型，是一种一对多、多对多分类相结合的故障诊断模型，用于解决多分类问题效果较好；这种方法对于 N类分类问题，只需构造（N-1）个SVM分类器，并且不存在不可识别的区域，分类过程比较快速，具有较好的泛化能力。实验证明与传统的三比值法和神经网络方法相比，所提出的方法在故障诊断的正判率上都有了提高，具有较好的实用价值。
문장장향량투영법응용재변압기고장진단중，해결료여하구건유효SVM 층차적문제。안조류여류양본집지간적상교정황，이용구씨거리화경향기함수계산류여류적공간거리화류간가분성측도，근거가분성측도진행배서，설계비교합리적층차결구진행분류。저충방법건립적고장진단모형，시일충일대다、다대다분류상결합적고장진단모형，용우해결다분류문제효과교호；저충방법대우 N류분류문제，지수구조（N-1）개SVM분류기，병차불존재불가식별적구역，분류과정비교쾌속，구유교호적범화능력。실험증명여전통적삼비치법화신경망락방법상비，소제출적방법재고장진단적정판솔상도유료제고，구유교호적실용개치。
By applying vector projection method in fault diagnosis for transformer ,the problem that how to structure effective SVM hierarchy based on decision-tree-based support vector machines (DTBSVM ) is solved . According to the cross situation between classification and classification sample sets ,Euclidean distance and radial basis function are utilized to calculate spatial distance and divisibility measure between different classifi-cations ,and the sequence on the basis of divisibility measure is made to design more reasonable hierarchy structure for classification .The fault diagnosis model combining one-to-rest with rest-to-rest classification is established by using the method of vector projection on decision-tree-based support vector machines ,and it can solve the multi-classification problem better .The method of vector projection aiming at N classification problem just constructs (N-1) SVM classifiers and has no unrecognized sector ,so the classification process is faster and the generalization ability is better .The test results show that correct-sentence rate increases compa-ring with traditional three-ratio method and neural network method in fault diagnosis ,so the method has bet-ter utility value .