WIDESPREAD ARTIFICIAL NEURAL NETWORK (ANN) STRUCTURES FOR CURRENT HARMONICS EXTRACTION

Abstract

The extraction of current harmonics produced by
nonlinear loads in the power system is significant issue for
compensating them fast and accurately. In this paper, the
main contribution is that widespread artificial neural
network (ANN) structures are used to acquire harmonic
components generated by a six-pulse uncontrolled rectifier.
For this purpose, ANN including computational algorithms
operate according to the functionality and structural
scheme of biologic neurons. Among ANN network
structures, radial basis function (RBF) and multilayer
feedforward (MLF) networks are two widespread used
architectures for harmonic extraction in literature. Thus,
this paper examines harmonic extraction performances of
these common used ANN structures by using
MATLAB/Simulink environment. In this way, the networks
are firstly trained by using sample input data and output
data with respect of training algorithms. In order to train
the networks, some training algorithms are applied, and the
optimal network and algorithm are emphasized