Classification of Power Transmission Line Faults Using an Ensemble Feature Extraction and Classifier Method

Abstract

This paper proposes an ensemble of feature extraction techniques for extracting features and an ensemble machine learning algorithm for classification of transmission line faults from voltage, voltage angle, and frequency signals. Transmission line protection is an important facet of a reliable power system. Many measures have been adopted by utilities worldwide for transmission line protection. Phasor Measurement Units (PMU) have been deployed in power grids throughout the world. PMU data can be used for detection, classification, and localization of faults in transmission lines. The characteristics of voltage, current, and frequency signals changes on different kinds of faults. The PMU data can be analyzed for signal characteristics, and these can be extracted as features. Maximum overlap discrete wavelet packet transform (MODWPT), Autoregressive coefficients, and Wavelet variance methods are used for feature extraction from power system signal data. A Machine learning algorithm for classification, ensemble bagging with a tree as a weak learner is used for classification of transmission lines faults. The performance of the ensemble classification algorithm is compared with other widely used machine learning classification algorithms.