Study on Early Warning of Small Leakage in Primary Loop Based on Data Mining

Abstract

An excessive small leakage in the primary loop may cause serious accidents. In order to prevent such accidents, this study proposes an fault early-warning method of the improved Gaussian mixture model (GMM)-grey relational analysis (GRA)-entropy weighting method (EWM) based on the integration of multiple characteristic parameters. In this method, firstly, the mechanism of the dynamic operating characteristics of the small leakage in the primary loop is analyzed, and the early-warning characteristic parameters are thus determined. Then, based on these characteristic parameters, in combination with the EWM and GRA, a multi-parameter integrated early-warning model is established. Finally, the relational analysis and improved GMM algorithm are used to enable effective learning of the statistical features of massive data so that the early-warning thresholds can be self-adapted to different conditions. As this study shows, effective early warning can be realized in this method under variable operating conditions. Compared with single parameter and fixed threshold warning, this method is more stable and provides more accurate, effective and timely warning, and thus, it can provide a reference for the condition monitoring of the primary system.