Using the simplest neural network as a tool for fault location in power lines

Abstract

Due to the fast development of digital technologies, wave method of fault location in electric power networks finds wide application. In contrast to traditional methods of determining the location of damage in the electrical system, the wave method is based on the exact determination of the beginning moment. It is not based on assessment of the transition process parameters. In this research, it is proposed to determine fault location in power networks using algorithms based on Artificial Neural Network (ANN) apparatus. For this purpose, feed forward ANN, trained using error back propagation method, and minimum number of neurons has been constructed. This approach has allowed simplifying and speeding up the process of ANN training. For training the ANN, transients simulation results in power networks using the method of synthetic circuits (Dommel's algorithm) have been used. It has been shown that the fault location problem can be solved as follows. First, on the basis of recordings with a traditional digitization frequency of 0.6–2.4 kHz, the approximate area containing fault is determined. Then, fault location is performed using ANN with an accuracy of hundreds, even tens meters. However; this requires the presence of a “high-frequency” (from the order of 0.5 MHz) analog-to-digital converter with a cyclically updated buffer in the measuring apparatus. Such accuracy is associated with the fundamental possibility of obtaining temporal resolution in determining the “front” of the transition process in a time interval which duration is an order of magnitude shorter than the sampling period.