论文部分内容阅读
由于旋转设备故障数据样本存在不平衡性,导致传统的LSSVM无法对异常值样本正确分类,为了解决这一问题,首先采用LSSVM从训练集中提取错分样本及其分类的支持向量,再根据各类故障样本数量对惩罚因子进行加权,以减少样本数量不平衡对分类结果的影响;然后根据错分样本到本类边界支持向量的距离,对松弛系数设置不同的权值,使错分的异常值样本分类得以修正。通过煤矿风机故障数据集验证了该算法分类效果明显优于传统的LSSVM方法。它有效地消除了因故障样本数据不平衡、样本分布异常对分类器造成的影响,提高了设备故障诊断的正确率。
In order to solve this problem, traditional LSSVM can not correctly classify outliers due to the imbalance of data samples of rotating equipment failures. First, the LSSVM is used to extract the support vector of the misclassified samples and their classification from the training set. The number of fault samples is weighted by the penalty factors to reduce the influence of sample number imbalance on the classification results. Then, according to the distance from the misclassified sample to the boundary support vector in this class, different weights are set for the relaxation coefficients so that the outliers Sample classification can be amended. The coal mine fan failure data set shows that the proposed algorithm has better classification performance than the traditional LSSVM method. It effectively eliminates the influence of the abnormal sample distribution due to unbalanced sample data and the classifier, and improves the accuracy of equipment fault diagnosis.