Anomaly detection in gas turbine compressor of a power generation plant using similarity-based modeling and multivariate analysis

Abstract

This paper introduces advances on the implementation of anomaly detection modules based on a combination of nonparametric models and multivariate analysis of residuals. The proposed anomaly detector utilizes similarity-based modeling (SBM) techniques to represent the process behavior and principal component analysis (PCA) for the study of model residuals; while partial least squares (PLS) is used to select an optimal subset of process variables to be included in the design of the detection module. In addition, the method considers a structured algorithm for the optimal inclusion of representative samples from the data set that is used to define the normal operation of the system. The method is validated using data that characterizes the operation of a compressor in a power generation plant.