基于动态独立成分的单类支持向量机方法及其在故障诊断中的应用

针对工业过程故障诊断中数据的动态性、非高斯性和非线性特点,提出一种基于动态独立成分的单类支持向量机(OCSVM)方法。为了分析数据的动态特性和非高斯性,应用动态独立成分分析(DICA)方法提取数据变量中的动态独立成分作为特征信息,基于特征信息建立OCSVM模型并构造非线性监控统计量。检测到故障后,计算故障数据与故障模式数据决策超平面的相似度,通过相似度分析识别故障模式。在Tennessee Eastman基准过程上的仿真结果表明,提出的方法能够比单类支持向量机更有效地检测过程故障,并且能够正确识别故障模式。

One-class support vector machine based on dynamic independent component and its application to fault diagnosis

In order to analyze dynamic,non-Gaussian and nonlinear property of data in industrial process fault diagnosis,one-class support vector machine based on dynamic independent component was presented.Dynamic independent component analysis was firstly applied to deal with dynamic and non-Gaussian data to obtain dynamic independent components as feature information.Then one-class support vector machine was used to build nonlinear monitoring statistics based on feature information.After fault was detected,the similarity between new fault data and fault pattern data was computed for fault pattern identification according to their decision hyper planes.The simulation results on Tennessee Eastman benchmark process show that the proposed method can detect fault more effectively than one-class support vector machine and detect diagnosis fault pattern correctly.