半监督增量式SVM在故障诊断中的应用研究

Application Research of Semi-supervised Incremental SVM on Fault Diagnosis

基于半监督学习能够有效降低人工标注成本,以及增量学习可以加快训练速度,避免数据量大时训练时间过长等特性,本文提出了一种半监督增量式SVM算法。在算法中,首先对已标记样本进行训练得到初始分类器,然后利用此分类器对新增样本进行标记,最后结合KKT条件选择合适的样本对分类器进行更新。每当有新样本加入便执行以上过程,以保证分类器得到及时更新。将该算法运用于6135D型柴油机的故障诊断中,并与传统SVM算法和增量式SVM算法进行了对比,证实了本文所提算法的可行性与有效性。

For that the employment of the semi-supervised learning based approach decreases the cost of manually-marking,and the employ- ment of the incremental learning method gain the learning speed ,which avoid the long-time learning when the number of the data is too large. a semi-supervised incremental learning SVM （Support Vector Machine） algorithm is proposed. In this algorithm, the initial classifier is ob- tained by training the pre-labeled samples. Then the new samples are labeled by this classifier. At last, KKT constraint is used for the selec- tion of the proper samples which are employed for the updating of classifier. The classifier is renewed by new samples in this process. In the last, this algorithm is applied into the fault diagnosis of the 6135D diesel engine. A performance comparison between the conventional SVM al- gorithm and semi-supervised incremental learning SVM algorithm are also made. The experiment results demonstrate the effectiveness of the algorithm.