Probabilistic Weighted NPE-SVDD for chemical process monitoring

• WNPE-SVDD is proposed to improve NPE-based process monitoring performance.
• A probabilistic weighting strategy is proposed.
• The useful information in process monitoring has been highlighted.
• The irrelevant information in process monitoring has been suppressed.
• Monitoring performance has been significantly improved.

Probabilistic Weighted Neighborhood Preserving Embedding and Support Vector Data Description (WNPE-SVDD) is proposed to improve chemical process monitoring performance. First, the NPE that is a linear approximation of Locally Linear Embedding is used for dimensionality reduction. Then, a probabilistic weighting strategy is proposed to solve the suppression of useful information by highlighting useful information and removing irrelevant information. Finally, SVDD is used to assess the process status. Case studies on a numerical system, continuously simulated stirred tank reactor process, and Tennessee Eastman process are provided to demonstrate the efficiency of the WNPE-SVDD method.