Cluster Recognition and Early Warning Modeling for Rotating Stall of Gas Turbine (iSPEC 2020)

Abstract

Aiming at the problem that it's difficult to accurately evaluate the deterioration stage of rotating stall in gas turbine power plant, an early warning model construction method based on the whole stage identification of rotating stall deterioration is proposed. The fractional norm improved k-means clustering algorithm is introduced to cluster the whole process of rotating stall development, and the optimal number of clusters is determined by the contour coefficient. The Tanimoto coefficient is used to screen out the key parameters, and the current operation state of the unit is reflected by the dispersion degree of the deviation from the normal value. The correlation between the discrete values of key parameters and each stage of rotating stall is calculated, and then the early warning model is established. Based on the actual operation data of the power station, this method can accurately judge the fault stage consistent with the actual operation and maintenance test results, and provide high reference value for operators.