A novel ensemble classifier for fault diagnosis of thermal system under different operational conditions

Abstract

Proper and prompt fault diagnosis of a thermal system is important for the safe and reliable operation of a large-scale power plant. It is a difficult task due to the structural complexity of the thermal system and the variable operating points of the system. The present paper proposes an approach based on a novel ensemble classifier to identify faults in a power plant thermal system under different operational conditions. In this new ensemble classifier, the base classifier is an optimized nearest prototype classifier and the fusion classifier is a three-layer neural network. It is developed in such a way that the dataset is divided into several smaller and easier-to-learn partitions, and each base classifier learns only one of the partitions, while the fusion classifier combines the outputs of the different base classifiers to approximate the original complex decision boundary. Typical faults of the high-pressure feedwater heater system are simulated under several different operating points on a full-scope simulator of a 600-MW coal-fired power unit and the results demonstrate the validity of the proposed ensemble classifier.