Forecasting of voltage disturbances at the beginning of ferroresonance phenomena in power systems using by a new approach of long-short term memory (LSTM)

LSTM for Ferroresonance in power siystems

Abstract

Ferroresonance is an unexpected, continuous, and undesirable event that causes excessive growth and distortion in the voltage and current waveform, whose cause is unknown and suddenly develops in power systems. To take precautions against the ferroresonance phenomenon, it is necessary to be able to predict how the distortions in the ferroresonance voltage will continue. Today, the ferroresonance problem has not been fully solved yet and more scientific studies are still needed. In this study, a new approach of an LSTM network has been developed that predicts the irregular and excessively large-amplitude continuing behavioral disturbances of the phase voltage in a real electrical power network exposed to ferroresonance. As a result of the study, the ferroresonance-voltage continuing in distorted waveform was estimated with an error of 0.0346 according to the Mean Absolute Percent Error (MAPE). The training data used in the study is only about 5% of all predicted ferroresonance voltage data. The successful estimation of 95% of the Ferreroresonance voltage using only about 5%, proves the success of the LSTM model applied with a new approach for the study.