Prediction of the true harmonic current contribution of nonlinear loads using NARX neural network

This paper presents a method to predict the load current harmonics injected into micro grid power systems using Nonlinear Auto Regressive neural networks with eXogenous input (NARX neural network). The proposed NARX network will be used to model the nonlinearity of electric loads. The network will be trained using data obtained from field measurements. After training the proposed network, it will be injected with pure sinusoidal voltage waveform to identify and isolate the current harmonics caused by nonlinear loads. The measurements of the nonlinear load are taken from Khalda - Main Razzak (MRZK) power station 1.2 MW capacity. The station consists of four Distributed Generators (DG) supply various linear and nonlinear loads, so it can be considered as an isolated micro grid. The nonlinear load under test is an Electrical Submersible Pump (ESP) which is driven by induction motor and controlled by Variable Speed Drive (VSD). The proposed method has proven a good performance for ESP current harmonic prediction and thus justifies accuracy and reliability. (C) 2017 Faculty of Engineering, Alexandria University. Production and hosting by Elsevier B.V.