Investigating Machine Learning Capabilities for Fault Classification and Fault Region Identification in Onshore Wind Farm Collectors

This paper explores a Machine Learning (ML)-based methodology for fault classification and fault region identification in onshore wind farm collector systems. For the studies, a system with a realistic wind farm topology is modeled in the PSCAD software. The main contributions of this paper are i) recommendations about ML methods that, when combined with the proposed feature extraction process, are most suitable for fault diagnosis in onshore wind farm collectors; ii) analyses of ML methods with a practical implementation bias, by assessing the impacts of the training set size, signal noise levels, and signal sampling frequency in the methods’ performance; and iii) investigations on the ML methods’ generalization capability.