改进的递归PCA方法对某型发射机的自适应监测

摘要

In order to avoid false alarms for time-varying process and missed alarms for weak faults, an improved recursive principal component analysis (RPCA) method based on block-wise updating was proposed to monitor time-varying process adaptively. The method updated loading matrix and eigenvalue matrix by applying low-rank singular value decomposition ( SVD) to the decomposition of correlation matrix, provided the selection strategy of data-block size and the update strategy of mean and variance, and updated control limits recursively using exponential weighted algorithm. The comparison was made between the proposed method and other methods for monitoring working process of a radar transmitter. The results show that the proposed method can track time-varying process adaptively to detect faults, and reduce false alarm rate. In addition, the adaptive control limits are more sensitive to weak faults and can avoid missed alarm effectively.%针对现有监测手段对时变过程易产生误警且对微弱故障监测力不足等问题,提出一种改进的递归主元分析方法,对时变过程实现自适应监测.以数据块为单位更新主元模型,采用低秩奇异值分解方法完成相关矩阵的递归分解,实现负荷矩阵和特征值矩阵的递归计算,并制定了数据块大小的选取策略和均值、方差的更新策略.同时,引入指数加权思想实现了控制限的递归更新.通过将该方法与其他监测方法应用于某型雷达发射机工作过程的监测并进行对比验证,结果表明该方法能自适应地跟踪过程时变并实时监测故障,降低了误警率；同时,自适应的控制限对微弱故障具有较高的灵敏度,有效地避免了漏报的发生.