A Practical Approach for Fault Location in Transmission Lines with Series Compensation Using Artificial Neural Networks: Results with Field Data

This paper presents a new method for fault location in transmission lines with series compensation, using data from voltage and current measurements at both terminals, applied to artificial neural networks. To determine the fault location, we present the proposal of using current phasors, obtained from the oscillography recorded during the short circuit, as the input to the neural network for training. However, the method does not rely on the internal voltage values of the sources or their respective equivalent Thevenin impedances to generate training files for the neural network in a transient simulator. The source data are not known exactly at the time of the short circuit in the transmission line, leading to greater errors when neural networks are applied to real electrical systems of utility companies, which reduces the dependency on electrical network parameters. To present the new method, a conventional fault location algorithm based on neural networks is initially described, highlighting how the dependency on source parameters can hinder the application of the artificial neural network in real cases encountered in utility electrical systems. Subsequently, the new algorithm is described and applied to simulated and real fault cases. Low errors are obtained in both situations, demonstrating its effectiveness and practical applicability. It is noted that the neural networks used for real cases are trained using simulated faults but without any data from the terminal sources. Although we expect the findings of this paper to have relevance in transmission lines with series compensation, the new method can also be applied to conventional transmission lines, i.e., without series compensation, as evidenced by the results presented.