Process Monitoring Model for Fused Magnesia Smelting Process Based on Feature Extraction With Reduced Inter-Variable Correlation

This article proposes a novel solution to resolve the process monitoring problem in the fused magnesia smelting process (FMSP). The proposed solution consists of the following main steps. First, we introduce feature extraction with reduced intervariable correlation (FERIVC) method, which reduces correlation between variables to capture more representative low-dimensional features. Second, on this basis, we establish the FERIVC abnormal detection model and use the reconstruction-based contribution method to isolate abnormal variables and construct an abnormal variable set. Third, we use transfer entropy (TE) to causally analyze the abnormal variable set and locate the root cause variable, determining the variable propagation path. Fourth, we calculate the compensation magnitude of the root variable to return to the critical normal operating condition, to facilitate the accommodation of the root variable. Finally, we apply the proposed solution to actual FMSP data, and experimental results confirm the effectiveness of the FERIVC process monitoring method.