Density-Based Structure Preserving Projections Process Monitoring Model for Fused Magnesia Smelting Process

Abstract

In this work, we address the monitoring problem in fused magnesia smelting process (FMSP). Our main goal is to accurately detect anomalies occurring in FMSP and isolate as few abnormal responsibility variables as possible. To this end, we propose anomaly detection method with density-based structure preserving projections (DSPP) and abnormal variable isolation method. DSPP first measures the degree of dispersion in the data set by calculating the sample distance entropy, and obtains the weight coefficients under the density constraint of the samples in the neighborhood and the non-neighbors, thereby establishing the objective function of the global-local structure preserving to obtain projections. On this basis, negative compensation statistics (NCS) is constructed to calculate contribution of abnormal variables by setting a unified standard index. Finally, the DSPP-based anomaly detection method and the abnormal variable isolation method are applied to a benchmark dataset and the practical FMSP. The experimental results confirm the effectiveness of the proposed method.