Machine learning for fault detection in distributed networks

This research includes applications of machine learning algorithms for the detection, classification, and analysis of disturbances, emphasizing symmetric and asymmetric short-circuit faults in electrical networks. The required triphasic voltage and current values were obtained by simulation with DIgSILENT software according to different short-circuit states, and the analyzes were performed through Python software. First, the preliminary data processing is done by applying the discrete wavelet transform, where a model is developed to select the mother wavelet and level of decomposition by applying the minimum entropy decomposition and Support Vector Machine algorithm. Additionally, unsupervised dimensionality reduction techniques were applied to improve machine learning models' performance during the training step. Finally, features considered are minimized through feature extraction and, by considering fewer features, prevent model data sets from being overfitting and underfitting; hence, the performance of the algorithms can be enhanced. The algorithms and approaches developed can also be applied to different fault problems to obtain more reliable protection methods.