Modelo de predicción de fallas en líneas de transmisión aplicado al sistema IEEE de 39 barras usando redes neuronales artificiales

Abstract

In this degree work, a computational program is developed using Python programming language, to perform the prediction of electrical faults in transmission lines (L / T) using artificial neural networks (ANN) from the keras and Tensorflow library.

Due to the lack of data on L / T failures, it begins by creating a database taking names and distances of the L / T from the IEEE 39 bar test system. Generating random numbers, the L / T in failure is selected, the failure rates, probability of failure and repair, distance and type of failure are estimated. To find the failure and repair times, the inverse transformation method of probability distributions is used.

With the fault events created, your information enters the ANN, which is multilayered with forward feeding, and supervised learning. The optimal configuration of the RNA is done through trial and error, thus comparing different configurations (number of neurons and hidden layers), finding the RNA that best fits.

The comparison between RNA configurations uses the calculation of the root mean square error (MSE). The database divided into training and validation sets will be used in the training and validation stage respectively.