Anomaly early warning method based on SVDD algorithm in industrial field and its effectiveness research

In industrial contexts, the importance of early warning systems lies in their capacity to identify and respond to potential anomalies promptly, thereby mitigating the risk of equipment failure, production interruptions, and safety incidents. This paper proposes two SVDD (Support Vector Data Description)-based early warning methods for industrial sites. Anomaly diagnosis algorithm based on correlation analysis-clustering-SVDD combs the historical dynamic characteristics of control loops of industrial production, and carries out dynamic anomaly diagnosis and early alarm online. The anomaly diagnosis method based on the recent online rolling data is to diagnose the fluctuation anomaly and early warning alerts for key measurement points and key bits online by combining the expert experience debugging with the feature extraction method of recent fluctuation characteristics of parameters (SVDD). The paper further explores the concept of early warning effectiveness, including the timeliness and sensitivity of early warning and the comprehensiveness of early warning mechanisms. Ways to ensure that early warning is effective in the long term are also explored-Periodic online learning.The reliability of the methodology is verified through simulation experiments and practical applications in industrial sites.