CAJ | 학술논문

提出了基于非对称AdaBoost算法与邻近支持向量机的模拟电路集成故障诊断方法.该方法采用非对称AdaBoost解决邻近支持向量机处理多分类问题出现的数据不平衡问题, 提高邻近支持向量机的分类准确率;利用高斯变异策略自适应选择核函数及正则化因子, 提高集成邻近支持向量机的个体间差异性;最后获得分类精度高和泛化性能好的分类器.通过对模拟电路的故障诊断实验, 验证了该方法的鲁棒性和可行性.诊断结果表明本文方法具有故障定位准确和泛化性好的特点.
제출료기우비대칭AdaBoost산법여린근지지향량궤적모의전로집성고장진단방법.해방법채용비대칭AdaBoost해결린근지지향량궤처리다분류문제출현적수거불평형문제, 제고린근지지향량궤적분류준학솔;이용고사변이책략자괄응선택핵함수급정칙화인자, 제고집성린근지지향량궤적개체간차이성;최후획득분류정도고화범화성능호적분류기.통과대모의전로적고장진단실험, 험증료해방법적로봉성화가행성.진단결과표명본문방법구유고장정위준학화범화성호적특점.
Based on asymmetric AdaBoost algorithm and proximal support vector machine (PSVM), a new ensemble fault diagnosis method for analog circuit is proposed. The proposed approach deals with the case of PSVM multi-classifier, which has unbalanced data in classification by asymmetric AdaBoost algorithm to improve the rate of accuracy. The kernel function and regulation parameters of individual SVMs can be automatically selected with evolutionary strategy, so the diversity is emphasized among individual SVMs in ensemble. The realization of the proposed strategy is expounded by using two practical circuits.The experiment results show that the proposed method has the capability to diagnose faults in tolerance circuits and gains satisfactory performance.