摘要
提出了一种适用于变工况过程的动态多主元模型故障检测方法。首先对过程数据进行分类得到各稳态工况下的数据;然后根据分类数据建立主元模型组来描述整个过程的统计特性;最后在故障检测中根据检测样本对分类数据的隶属度和主元模型组计算出与当前工况相适应的主元模型并进行检测。以火电厂锅炉过程为例对比研究了传统方法和新方法的应用情况。试验结果表明新方法能适应工况变化,减少误检并提高了检测灵敏度。
Traditional PCA fault detection methods are designed for simple steady working conditions and may occasion erroneous conclusions if used for frequently changing working conditions. A new detection method based on dynamic multi-principal component models is being proposed for this purpose. First process data for various steady-states are obtained by data classification, then principal component model groups are established to describe the statistical features of the whole working process. Finally during fault monitoring a PCM fitting current working conditions are calculated according to membership of the classified detection data samples and the principal component model group. With a fossilfired power plant boiler as an example, comparisons between traditional methods and the new one show that the new method has good adaptability to working condition variations, reduces chances of detection misktakes and has a higher varying detection sensitivity. Figs 6, table 1 and refs 10.
出处
《动力工程》
EI
CSCD
北大核心
2005年第4期554-558,598,共6页
Power Engineering
关键词
自动控制技术
变工况过程
主元分析
故障检测
K均值聚类分析
automatic control techanique
varying working conditions
principal component analysis
fault detection
Kmean cluster analysis