A Novel Control Strategy For Shunt Active Power Filter Using NARX Neural Network

摘要

This paper presents a novel control method for shunt Active Power Filter (APF) based on the Nonlinear Auto Regressive with eXogenous neural network (NARX). The proposed series parallel NARX network is trained by using learning algorithm called static backpropagation algorithm for electric loads nonlinearity modelling. Then the well-trained series-parallel architecture NARX network is converted to parallel architecture NARX network and tested with different patterns. The algorithm of the instantaneous reactive power is integrated with the proposed NARX networks to analyze the predominant harmonics. Finally, the control reference signal of the shunt APF is designed. The proposed method is applied on an Electrical Submersible Pump (ESP) rotated by three-phase induction motor which is one of most widely used loads in petroleum industry field. The results show very good behavior of the NARX neural network in harmonic detection and mitigation.