Análise Comparativa de Métodos de Aprendizagem de Máquina para a Classificação de Faltas em Linhas Conectadas a Geradores Eólicos Interfaceados por Inversores

Abstract

The increasing insertion of Inverter-Based Wind Resources (IBWRs) into the primary network has generated new challenges in fault classification due to the strong influence of inverter control methods on the contributions of IBWRs to faults. In the present study, six classic and well-validated Machine Learning (ML) methods are evaluated for fault classification:Decision Tree,Support Vector Machines,Random Forest,K-Nearest Neighbors,Gaussian Naive Bayes andArtificial Neural Network. These methodologies will be evaluated from different aspects, such as the selection and adjustment of learning models, analysis of the influence of noise on signals, and generalizability to different measurement points. The methods were implemented in the Python programming language with the help of the open source scientific librarySklearn, which offers programming freedom ranging from data pre-processing to subsequent validation and improvement of models. The study shows promising results that validate the use of ML for fault classification in transmission systems in the presence of IBWRs.