Author: Charles Gathiani
Gathiani, Charles, 2024 A comparative analysis of machine learning algorithms for fault detection and classification in microgrids , Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
Optimal operation and control of microgrids depends on the accuracy of the fault detection and classification capabilities, which allows for quick fault identification, isolation, and recovery. Due to a reliance on large fault currents and the dynamic nature of microgrids, there is a need for the
development of new fault detection techniques.
This study investigates and proposes a machine learning-based microgrid fault detection scheme for high precision using Bayesian regularization algorithm. The proposed machine learning method extracts its learning features from the point of common coupling of the distributed energy resources and the main grid using the discrete wavelet transform. Under different fault and microgrid operating conditions, the learning features extracted were the three-phase measurements of the voltage magnitude, three-phase measurements of the current magnitude, fault impedance, zero sequence voltage values, zero sequence current values, and frequency. The Discrete Wavelet Transform was used to extract the learning features and then decompose them into the time-frequency characteristics. The eight extracted features were then applied as the input variables for purposes of machine learning.
To investigate and validate the performance and effectiveness of the detection and classification model, the results were compared to other training algorithms for accuracy, selectivity, and sensitivity. The results of the simulations were compared to the Levenberg Marquardt training
algorithm. The simulation results clearly indicate that the Bayesian Regularization algorithm provides more accurate detection and classification of faults while guaranteeing better response to changes in input variables resulting from microgrid operating conditions. The Bayesian Regularization algorithm did not experience overfitting and provided accurate results even with an introduction of new variables. Although the Bayesian Regularization algorithm provided accurate results and the best response, it had a longer processing time which may not be suitable for use in time-constrained operations.
Keywords: Microgrids, Faults, Fault Detection, Machine Learning
Subject: Engineering thesis
Thesis type: Masters
Completed: 2024
School: College of Science and Engineering
Supervisor: Dr Amin Mahmoudi