采用核主元残差空间分析方法诊断传感器故障

RESIDUAL SPACE OF KERNEL PRINCIPAL COMPONENT ANALYSIS BASED SENSOR FAULT DIAGNOSIS

将核主元的高维特征空间分为主元空间和残差空间,提出了1种基于残差空间的故障分离算法,其约简了反映主元空间的分项,使计算量大幅降低。应用该方法诊断传感器故障具有不受残差污染,易于实现等优点。对某600MW机组回热系统测量数据进行分析结果表明,基于核主元残差空间分析传感器的故障诊断方法能够及时诊断出不同类型传感器的故障,即使在多个传感器同时出现故障时,也能够准确诊断出故障,并能够及时地分离出故障传感器。

High dimensional feature space of the kernel principle component was divided into principle component subspace and residual space in this paper.A residual space based fault isolation algorithm was proposed,which reduced the subitem reflecting the principle component subspace,made the computational complexity dramatically decreased.Fault isolation algorithm applying residual space of kernel principle component analysis has such advantages as restriction of residual contamination and easy to be realized,so it can be used for sensor fault diagnosis.Taking the regenerative system of a 600 MW unit as an example,analysis on the measured data shows that,this method can timely diagnose the sensor fault of different types,even when multiple sensors' failures occur simultaneously,it can accurately diagnose the fault and isolate the fault sensor in time.