Anomaly detection in power generation plants using similarity-based modeling and multivariate analysis

This paper introduces an anomaly detection method based on a combination of nonparametric models of the process and multivariate analysis of residuals. This method basically intends to recognize abnormal conditions in the operation of a monitored system, considering for this purpose the definition of "baseline" operation through historical datasets. In particular, 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. The methodology not only helps to detect changes in the operation of the system, but also provides a structured algorithm for the inclusion of representative samples in the data set that is used to define the baseline of the system. The method is validated using data from a power generation plant.