一种SVM不平衡分类方法及在故障诊断的应用

针对支持向量机不平衡样本分类倾斜性问题,提出一种欠采样支持向量机分类器。构建包含少类样本的最小封闭超球体,计算各个多类样本到包含少类样本最小封闭超球体球心的距离,利用该距离对多类样本进行欠采样,产生新的训练集,实现训练集的平衡。该方法和其他不平衡分类方法在基准数据集的分类结果表明该方法在识别率和分类速度方面的有效性。将该方法应用于永磁同步电机驱动电路功率开关管开路故障诊断中,结果表明该方法缩短故障分类器的训练时间,提高了故障分类器的泛化能力和诊断速度。

Imbalanced pattern classification method based on support vector machine and its application on fault diagnosis

Aiming at support vector machine solving imbalanced pattern classification tip problem,an imbalanced pattern classification method for support vector machine was proposed,basing on under-sampling.Hyper-sphere that contained small class samples with minimum volume was constructed.The distances between each sample of big class to hyper-sphere center were computed.According to above distances,under-sampling was carried out on big class.The balanced training set was formed.Simulations on benchmark dataset for classification showed the effectiveness of the proposed approach on classification accuracy and classification speed,compared with other imbalanced pattern classification method.This approach was applied on fault diagnosis of permanent magnet synchronous motor driving circuit power device open fault.Experiment result shows that the proposed approach can reduce training time,and improve generalization ability and diagnosis speed.