基于近邻距离加权主元分析的故障定位

Weighted Principal Component Analysis Based on Neighbor Distance for Fault Positioning

随着现代工业过程越来越复杂,采集到的数据量也越来越大,数据集中各个变量之间存在相互影响,一些传统的故障定位方法不能达到满意的定位效果.因此,在主元分析(PCA)故障定位方法的基础上,采用正常工况样本近邻距离对样本进行加权的方法,将每个样本与其最近邻样本的距离作为加权因子,削弱变量之间的相互影响,降低拖尾效应,改进PCA的故障定位效果.首先,对样本进行预处理并加权;然后,利用加权后正常工况的样本建立主元监视模型,在线监视过程中,在发生故障时刻,根据之前建立的模型计算相应时刻加权样本的重构监视统计量和每个变量的重构贡献值,通过最大贡献率法对变量进行逐步定位;最后在数值案例和TE过程进行仿真研究,并与基于PCA故障定位方法进行比较,结果表明了所提方法的有效性.

As the modern industrial process becomes more and more complex, the amount of data collected becomes larger and larger, and the variables in the data set interact with each other. Some traditional fault location methods can not achieve satisfactory position results. On the basis of principal component a- nalysis （PCA） fault positioning method, the method of weighting samples by nearest neighbor distance of normal working condition is put forward. The distance between each sample and its nearest neighbor sample is used as a weighting factor, which weakens the interaction between variables and reduces the smearing effect, improving PCA fault positioning effect. Firstly, the sample is pre-processed and weighted. Then, the principal component monitoring model is established by weighted samples of normal working conditions. In the process of monitoring, the moment of breakdown, calculating the reconstruction monitoring statistics and the reconstruction rate of each variable corresponding this instant based on previous models,to gradually positioning variables. Finally, numerical example and TE process are simulated and compared with fault location method based on PCA. The results show the effectiveness of the proposed method.