中文摘要：

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。

英文摘要：

Since imbalance exists in sample data of rotating equipment failure and traditional LSSVM fails to correctly classify abnormal samples,an imporved method is proposed in this thesis Double Weight LSSVM.First of all LSSVM to extract support vector of wrongly classified samples and their classification in training set,then use regularization factor is weighted on basis of different number of various fault samples to reduce effect of imbalance in sample size on classification.The wronly lassified abnormal samples will be amended into right classification by setting different estimation error variables according to their differences.The datas on fan failure database prove it to be a better way of classification than traditional LSSVM method.The double weight LSSVM is effective in eliminating imbalance in sample data of failure and effect of abnormal distribution of sample points on classifier,as well as improving power of failure diagnosis.