Concurrent monitoring of global-local performance indicators for large-scale process

Abstract

Abstract Multiblock methods with decision fusion are common schemes for performance indicator monitoring in large-scale process. However, single sub-block is insufficient for interpreting global performance indicator, thus the final fusion may cause inaccurate result. Besides, it is imperative to consider safety related local performance indicators (LPI) in each sub-block and important to model the correlation between each sub-block. In this paper, concurrent global-local performance indicator monitoring method is proposed. For more purposeful monitoring, this study constructs two feature subspaces, the local performance indicator related subspace (LPIRS) and the global performance indicator related subspace (GPIRS), with different significances. In LPIRS, LPI related variables in each sub-block are monitored. In GPIRS, considering the dynamic interactions between each sub-block, improved dynamic Canonical Correlation Analysis method is proposed for feature extraction. Moreover, the features are furtherly selected based on a novel selection criterion and the orthogonal decomposition on regression coefficients is employed to construct GPIRS. Finally, the effectiveness of the proposed method is validated via two cases.