CAJ | 학술논문

针对支持向量机( SVM)硬判定输出分类结果缺乏定量评价的问题，提出了一种多分类SVM后验概率建模的改进方法。通过引入D-S证据理论，得到多分类SVM在D-S证据理论识别框架下的基本概率分配，使样本在分类时同时具有定性解释和定量评价。接着，将多源信息送入SVM之后在决策级对多个SVM分类输出进行证据融合，以提高诊断精度。最后，将该方法应用于轴承故障的诊断中。结果表明，该方法能正确分类采用单源信息时所错分样本，降低识别的整体误差，显著提高故障诊断的准确性。
침대지지향량궤( SVM)경판정수출분류결과결핍정량평개적문제，제출료일충다분류SVM후험개솔건모적개진방법。통과인입D-S증거이론，득도다분류SVM재D-S증거이론식별광가하적기본개솔분배，사양본재분류시동시구유정성해석화정량평개。접착，장다원신식송입SVM지후재결책급대다개SVM분류수출진행증거융합，이제고진단정도。최후，장해방법응용우축승고장적진단중。결과표명，해방법능정학분류채용단원신식시소착분양본，강저식별적정체오차，현저제고고장진단적준학성。
[ Abstract] In view of the problem that the classification results of hard decision output of support vector ma-chine ( SVM) lack of quantitative evaluation, an improved modeling method for the posterior probability of multi-class SVM is proposed. Through the introduction of D-S evidence theory, the basic probability assignment ( BPA) of multi-class SVM under the recognition frame of evidence theory is obtained to enable the samples have both qualita-tive explanation and quantitative evaluation. And then the multi-source information is delivered to SVM to conduct the evidence fusion of several SVM classification outputs for improving diagnostic accuracy. Finally the method is applied to the fault diagnosis of bearings with a result showing that the method proposed can correctly classify the samples being classified wrongly using single-source information, reduce the overall error of recognition frame, and enhance the correctness of fault diagnosis remarkably.