摘要
针对二叉树支持向量机多分类算法在故障诊断中存在误差累积的问题,提出了一种新的算法优化方案;该方案旨在通过对分类顺序的优化来降低误差在二叉树架构层次间的传递和累积;算法充分考虑了小样本分布特点,首先从空间聚类分析的角度结合类间距离和类内密集性建立可分性测度作为主要的分类依据,其次从实际训练角度出发提出用预验证的方法作为对上述依据的补充;最后,利用UCI标准数据集,通过与不同多类算法进行比较,证明了该优化方案运用于小样本故障诊断中具有更高的推广性和鲁棒性。
An optimization aiming at binary tree support vector machine (BT--SVM) is proposed to address the error accumulation prob- lem existing in fault diagnosis. The algorithm is intended to optimize the classification order to reduce error propagation between the levels of the binary tree structure. The proposed algorithm gives full consideration to the characteristics of the small sample distribution. Firstly it combines class distance with data intrinsic density as the major basis for sorting order in the perspective of space clustering analysis. Then, it takes the pre-validated based on the practical training process as the supplement. Finally, the validation of the standard data sets proved that the optimization scheme used in fault diagnosis of small samples has higher promotion and robustness.
出处
《计算机测量与控制》
2015年第3期689-692,共4页
Computer Measurement &Control
关键词
支持向量机
二叉树
小样本
聚类分析
故障诊断
support vector machines
binary tree
small sample
clustering analysis
pre-validated