Harmonic Detection By Using Different Artificial Neural Network Topologies

At this paper the performance of different artificial neural networks (ANN) topologies has been analized for harmonic detection by distorted waveforms. With this information, it’s possible to obtain the reference signal for an active power filter (APF) control by nonlinear loads compensation. In particular, two ANN types, the static multilayer perceptron (MLP) and the dynamic MLP, stand out as the most suitable for distorsion identifying. Acceptable results were also obtained with recurrent networks, but with a lower performance than with the other topologies. Two different control strategies have been applied. One of them is based in the static MLP, neural network that has proved to be the most appropiate by measuring the rectangular components of the signal harmonics. The other strategy, based in the dynamic MLP, permits extracting the instantaneous value of the fundamental waveform. The three mentioned ANN topologies have been conveniently trained and simulated with waveforms distorted by several harmonics. Finally, the obtained results with practical cases of harmonic distorted waveforms are presented and discussed.