Short Term Forecasting of Power Quality Distortions in Electrical Energy Systems with LSTM and GRU Networks

Abstract

Technological development has led to a diversification of loads in transmission and distribution systems. The rise of non-linear loads in the system is one of the biggest effects of this variation as semiconductor technology develops. Nonlinear loads are characterized by current and voltage characteristics that are not purely sinusoidal, also known as harmonics. Harmonics cause the system insulation to degrade and increase energy loss. Therefore, it's crucial to get rid of harmonics before they occur. This study intends to lower the risk of distribution system damage by employing complex harmonic forecasting methods. An RNN-based forecasting algorithm has been created by using actual system power quality data obtained from the Organized Industrial Zone in Bandırma, Turkey. Parameters that are most likely to be neglected in simulation studies are also taken into account in the calculation by using actual data. Active power data, current harmonic data and calendar data were used together to design harmonic forecasting model. Graphs and calculations were used to discuss the results. The obtained minimum values of the RMSE, MAE, and MAPE are 2,116, 0,666 and 11,619, respectively. The convergence as a result of these calculations has allowed high forecasting performance of power quality distortions.