增量学习直推式支持向量机及其在旋转机械状态判别中的应用

直推式支持向量机(support vector machine, SVM)是基于已知样本建立对特定的未知样本进行有效识别的理论框架，与归纳式支持向量机相比，前者更经济、分类效果更佳。然而，直推式支持向量机的致命缺点是需要占用大量的训练时间，为此，提出了基于增量学习的支推式支持向量机训练算法，即把当前迭代训练得到的支持向量样本与新赋予类别标签的部分测试样本作为训练样本集参与下一次的迭代训目的是通过减少训练样本的数量以节约训练时间。同时，为确保算法的收敛性及分类准确率，在训练过程中引入了成对标注及错误回溯处理。实际的状态判别结果证明了该方法的有效性。

Transductive SVM Based on Incremental Learning and Its Application to State Judgement of Rotation Machinery

Transductive support vector machine (SVM) is to construct theoretical frame for recognizing specifically unlabelling sample based on labelling sample effectively. Compared to traditional inductive SVM, transductive SVM is more economical, and has more perfect classification ability. However, fatal shortcoming of transductive SVM consumes a great deal of training time.Therefore,training algorithm was proposed for transductive SVM based on incremental learning. That was,the set of sample for next interative training was made of two part: sample of support vector got from currently interative training and some testing sample enduced with label newly.The aim was to reduce training sample and save training time.At the same time, the method of pair labelling and back searching operation according to mistakenly classified testing sample were applied in the course of training to guarantee that the traning algorithm was convergent and had high precision of classification. Practical result of state judgement proves the method is valid.