Estimation of feeding composition content based on novel variable sliding window method and layered data reconciliation with multiple modes

Abstract

The real-time detection of feeding composition content holds significant importance in process monitoring and control optimization in industrial systems. However, the current feeding composition content is obtained by manual assay, with low detection frequency and significant hysteresis. Moreover, due to adjustment in production plans and adaptation to market demand, multiple operation modes frequently exist in real processes, making it difficult for a single mode to adapt to the actual site. To address the above problems, firstly, a novel variable sliding window method incorporating the improved longest common subsequence (LCSS) is proposed to achieve steady-state detection and multimode recognition in industrial processes. Secondly, according to the characteristics of different modes, the corresponding layered data reconciliation and parameter estimation (LDRPE) models are built to reduce the errors of the measured data and estimate the unmeasurable data. Finally, an online estimation model of feeding composition content based on multimode LDRPE is designed for estimating the coke content in the mixtures. The effectiveness of the proposed method is verified by the application to a real industrial system.