Detection of Faults on Power Line with Artificial Neural Networks

摘要

The problem of power lines fault recognition is relevant for various automation and relay protection devices. In this article the authors suggest considering the issue of determining the emergency mode on power line as a classification problem using power line operation parameters as classification features. It was proposed using an artificial neural network for this problem solution. For obtaining training and control samples, the authors made up a model of a power line, on which they simulated various types of short circuits and normal modes. The training sample contained 185,601 examples with various types of short circuit in different points of a power line. The control sample was formed in similar way. For the estimation of neural network performance the article proposed using such metrics as classification precision, recall, and f-measure. The influence of the configuration of the neural network on the accuracy of the classification of modes was investigated: it was noted that increasing the number of neural in the neural network improves the quality of fault recognition. It is noted that most of the classification errors occur in transient modes, which may be due to the high sampling rate of the training and control samples. It was also found that single-phase ground faults are the most difficult to recognize. The results of the experiments show that neural networks can be used to classify the modes of operation of power lines, for example, as part of control devices and power line protection.