基于改进混合概率主元分析模型的过程监控

Process Monitoring Based on Improved Mixture Probabilistic Principal Component Analysis Model

基于混合概率主元分析(M PPCA)的监控方法,存在要求各子模型中主元个数相同、监控指标不一致、监控表格过多等缺陷.为此对M PPCA算法进行改进,分两步建立模型:首先求出混合高斯模型(GMM),然后利用概率主元分析(PPCA)建立每个子模型的主元模型.改进方法中各子模型主元的选取兼顾了主元的解释率及其变化趋势,并引进基于PPCA的监控方法,保证了监控指标的一致性,减少了过程监控图.

Mixture probabilistic principal component MPPCA-based methods which the MPPCA model analysis （MPPCA） model is dealt with. To the disadvantages in needs all the sub-models have the same number of principal components, its two monitoring indices often can not consist with each other, and the monitoring charts are too many, an improved MPPCA model is presented in two steps. Firstly, a mixture Gaussian model is built. Then, by using probabilistic principle component analysis （PPCA）, the principal component model of each sub-model is developed. Compared with MPPCA, the improved MPPCA selects the principal components based on not only the explanation of process variables in different sub-models but also its trend of changing. By introducing PPCA, the consistence of monitoring indices is guaranteed and the number of charts used by monitor is reduced.