Analysing Non-Ensemble Machine Learning Methods for Solving the Transient Classification Problem

This article analyzes machine learning methods for solving the problem of classifying transients occurring in the power system, in particular, the classification of short circuits and engine start. The article is part of a study of the possibility of using machine learning algorithms in the operation of relay protection. The general purpose of the study is to increase the sensitivity of relay protection by changing the adjustment coefficient of the engine start protection setting by one. The transient data is a time series with instantaneous current and voltage values obtained from the relay protection terminal recorder. Three non-binary machine learning methods for solving the classification problem are considered: Decision tree, K-nearest neighbors, and logistic regression. The article describes the methodology of data preparation and implementation of machine learning methods. According to the results of the study, the most effective of the simplest machine learning algorithms is a “Decision Tree”, but it is not enough to reliably classify transients.