摘要
在基于支持向量数据描述(SVDD)的故障诊断中,往往随着故障数据的不断增加而不断地进行再训练以调整诊断模型,浪费了大量时间。为了解决这一问题,提出一种新的SVDD增量学习算法。该方法在深入分析训练结果与数据样本的关系,多次利用KKT条件,对样本进行筛选,最终选择出影响最终结果的少量训练样本。通过实际电路故障提取采集数据并诊断,所得结果表明该算法可以选择出所有影响结果的相关样本,保证了准确率并避免了大量样本训练,节省了时间。
In fault diagnosis based on support vector data description （ SVDD）, usually the retraining is constantly carried out to adjust the diagnostic model along with the constant fault data increase, which waste lots of time. To solve this problem, we propose a new SVDD incre- mental learning algorithm. This method analyses in depth the relation between the training result and the data samples, and screens the sam- ples by using KKT conditions a couple of times, and finally selects out those few training samples affecting the eventual result. We extract and diagnose the collected data through practical electrical circuit fault, the derived result shows that the algorithm can select all the related sam- ples affecting the outcome, this ensures the accuracy and avoids huge sample training at the same time, saves the time.
出处
《计算机应用与软件》
CSCD
2015年第2期163-166,共4页
Computer Applications and Software
基金
国防预研基金重点资助项目(9140A27020211JB3402)
关键词
SVDD
增量学习
故障诊断
SVDD
Incremental learning
Fault diagnosis
