基于数据特性分析的多变量过程监测

以PCA,ICA为代表的多元统计监测方法总是基于各种各样的前提假设,如果不考虑它们的适用条件盲目选择监测算法,则可能给出错误结论,增加故障误报漏报的概率.针对理论方法在应用时面临的条件限制问题,在无先验知识的情况下,提出一种数据特性的分析方法,通过参数寻优并逐步剔除线性相关变量组的方法,实现多变量过程线性非线性的自动判别.仿真分析表明所提方法可以根据数据特点及各算法的适用条件自动选择适当的监测算法,具有一定的实用价值.

Multivariate Process Monitoring Based on the Characteristic Analysis of the Data

Multivariate statistical process monitoring methods such as PCA and ICA are always based on a variety of assumptions. If the constraints of these methods have not been considered when selecting these methods， wrong conclusions will be obtained and the rate of high leaking and false alarm will be increased. To solve the constraints problem of these methods in application， a data characteristic analysis method was proposed to test the correlation of the variables automatically， in which parameter optimization was conducted and the set of linear variables was eliminated sequentially. The simulation illustrated that the proposed method can select appropriate modelling method automatically according to data characteristics and applicable conditions of the methods， which has considerable practical value.