摘要
提出采用考虑到精度/差异权衡的SVM作为弱分类器的一种新的组合分类诊断方法——DiverseAdaBoost-SVM。该方法通过在一组具有适当精度的弱分类器中进一步选择具有较大差异性的弱分类器,对这些具有较大差异性的弱分类器进行组合,从而较好解决AdaBoost算法中存在的精度/差异权衡的难题;同时该方法也较好地解决了现有的AdaBoost方法存在的弱分类器本身参数选取困难问题及训练轮数T的合理选取问题。通过对基准数据库的测试及航空发动机故障样本的诊断,结果表明和其他方法相比,DiverseAdaBoost-SVM方法具有更好的泛化性能,更适合对分散程度较大、聚类性较差的航空发动机故障样本进行分类,也更适合对非对称故障样本集进行分类。
A novel approach of fault diagnosis named Diverse AdaBoost-SVM is presented, which uses SVM considering the accuracy/diversity dilemma as weak learner for AdaBoost. The proposed method successfully solves the dilemma in AdaBoost algorithm by selecting more diverse weak learners in those moderately accurate ones, meanwhile overcomes the difficulty of selection of weak learner parameter and learning cycles T in the existing AdaBoost methods. The practical applications to UCI Repository and aeorengine faulty samples show that the proposed method has better generalization performance, and is more fitting to classify the faulty sam pies scattered greatly and also more fitting to classify the unbalanced faulty samples.
出处
《航空学报》
EI
CAS
CSCD
北大核心
2007年第5期1085-1090,共6页
Acta Aeronautica et Astronautica Sinica
基金
军队重点科研基金(2003KJ01795)
