Fault detection and classification based on semi-supervised machine learning using KNN

Abstract

The fault analysis in electrical power system has become more critical problem in the present days, because of the incorporation of renewable energy sources with existing system. The Machine Learning is the latest technique which is taking attention for various power system applications. The study presents a Semi-Supervised Learning algorithm for the protection of transmission line. The semi-supervised learning method can handle both the labeled and unlabeled data effectively. An advanced approach K-Nearest Neighbor (KNN) model is introduced to detect and categorize the transmission line faults. The fault current values are measured at bus B1 and used as inputs to the KNN classifier to perform the detection and classification. In this work, by calculating distance between each labeled sample and the new or untagged sample, the fault class is determined corresponding to the kth minimum distance labeled points. The detailed methodology has been implemented in MATLAB environment. All possible fault situations with different fault resistances at different fault locations are simulated. It has been noticed that the present approach can effectively detects and classify the transmission line faults with the accuracy of 97 percent.