基于支持向量机集成的故障诊断

为提高故障诊断的准确性,提出了一种基于遗传算法的支持向量机集成学习方法,定义了相应的遗传操作算子,并探讨了集成下的分类器的构造策略。对汽轮机转子不平衡故障诊断的仿真实验结果表明,集成学习方法的性能通常优于单个支持向量机,而所提方法性能则优于Bagging与Boosting等传统集成学习方法,获得的集成所包括的分类器数目更少,而且结合多种分类器构造策略可提高分类器的多样性。该方法能容易地推广到神经网络、决策树等其他学习算法。

Fault Diagnosis Based on SVM Ensemble

A novel support vector machine (SVM)ensemble method which searches for the optimal combination of the base SVM classifiers by genetic algorithms is presented. Genetic operators are defined and strategies of constructing diverse base classifiers are discussed. Experiments on a steam turbine rotor unbalance fault diagnosis dataset show that ensemble learning methods are usually better than single SVMs while the presented method can gain even better performance with less base classifiers than traditional ensemble learning methods such as Bagging and Boosting. In addition, combination of multiple strategies is helpful for improving the diversity of classifiers. The presented method can be generalized easily to using algorithms such as neural networks, decision tree, and so on.