一种基于支持向量机的增量学习算法

An increasing learning algorithm based on SVM

具有增量学习功能的数据分类技术与普通的数据处理技术相比较,增量学习分类技术具有明显的优越性。在新的训练过程中充分利用了历史的训练结果,从而显著减少了后继训练的时间。介绍了支持向量机的基本理论和一般的支持向量机增量学习算法,针对有些渐变问题(如机械设备的早期故障期和损耗期),新样本所提供的信息量与历史样本所提供的信息量是不同的,给出一种新息加权的支持向量机的增量学习算法,通过循环来获得最优分类面。仿真实验表明,采用加权的增量算法更能反映新样本点的特征。

Compare to normal technology of data classifies, technology of data classify which have increasing learning ability have distinct advantage: it can take full advantage of history training result; it can reduce subsequence training time marked ness. This article introduces basic principle of SVM and normal increasing learning arithmetic. To some gradual change problem (such as forepart malfunction period and spoilage period of machine facility), new sample providers different information from history sample. This article puts forward a new kind of weight increasing learning algorithm, and gain better classify face through cycle. Simulate experiment shows that weight increasing learning algorithm can reflect character of new sample better.