传统的多变量统计过程监控方法一般都假设过程只运行在一个稳定工况下,但很多实际工业过程往往具有多工况特征。针对这一问题,提出一种基于混合PCA模型的多工况过程监控方法。将混合高斯模型和PCA相结合,用改进的EM算法估计模型的工况数以及各工况的分布参数和主元数,并构建归一化的统计量实现对多工况过程的监控。TE过程的仿真研究表明,所提出的方法相对传统PCA方法能更精确地估计各工况的统计特性,从而更准确及时地检测出多工况过程的各种故障。 Traditional multivariate statistical process monitoring methods generally assume that the process runs in a stable condition, but many of the actual industrial processes often have multi-condition features. In response to this problem, a hybrid PCA model based multi-condition process monitoring method is proposed. Combining the mixed Gaussian model and PCA, the improved EM algorithm is used to estimate the number of working conditions, the distribution parameters and the number of principal components of each model, and the normalized statistics are constructed to monitor the multi-working process. Simulation results of TE process show that the proposed method can estimate the statistic characteristics of each working condition more accurately than the traditional PCA method, so as to detect various faults of the multi-working process more accurately and timely.