Concurrent analysis of variable correlation and data distribution for monitoring large-scale processes under varying operation conditions

Abstract Large-scale processes are commonly characterized with hybrid correlations including both linear and nonlinear relationships. Besides, variables may present different probability distributions with different modes in response to changing operation conditions. For these varying-condition large-scale processes, it may be very challenging to describe process characteristics comprehensively and accurately, which has been seldom reported before. To handle the problem, a distributed and hierarchical monitoring framework is developed in the present work. First, a variable division method called hierarchical information-theoretic decomposition algorithm is proposed, in which both variable-wise correlations and sample-wise distributions have been concurrently analyzed. Second, a distributed and hierarchical modeling strategy is developed to extract both local and global characteristics. The lower-level distributed Gaussian mixture models integrated with principal component analysis are constructed for different sub-blocks. And the upper-level monitoring model is established to analyze the relations over different sub-blocks. Finally, a hierarchical monitoring framework based on Bayesian fusion strategy is developed, in which the local statistics are combined into the global statistics. To illustrate the feasibility and effectiveness, the proposed algorithm is applied to a numerical example and a real industrial process of the large scaled thermal power plant.