基于集成支持向量机的故障诊断方法研究

为了提高支持向量机的泛化能力,研究了Bagging集成学习方法对于支持向量机的提升作用,试验结果表明提升作用不明显。通过模拟数据扰动的方法,在标准数据集上通过试验定量比较了支持向量机和神经网络的稳定性,结果表明支持向量机相对于神经网络来说是一种稳定的分类器。在此基础上,提出了双重扰动法,即通过子空间法扰动数据特征,通过Bagging算法扰动数据分布,来达到提高基分类器之间差异性的目的,在标准数据集和故障诊断数据上进行了试验,试验结果表明,双重扰动法较好地提升了支持向量机的正确识别率。

A Fault Diagnosis Method Based on Ensemble Support Vector Machines

In order to enhance the generalization ability of Support Vector Machine(SVM),Bagging ensemble learning algorithm was studied.The experimental results of Bagging SVM in the standard data set showed that the Bagging method couldn't enhance the generalization ability of SVM markedly.In order to find reason of this,the stability of SVM and neural network was studied.The results showed that SVM is a relative stable classifier in comparison with neural network.Then,a double disturbance algorithm was proposed,in which the subspace method was used for data characteristics disturbance,and Bagging method for data distribution disturbance.Experiments were made by using double disturbance algorithm for the standard data sets and fault diagnosis data set,and the results showed that the recognition rate of SVM is obviously enhanced by this method.