Shunt Active Power Control Based On Feed-Forward Neural Network

Abstract

The shunt active power filter (SAPF) represents a prevalent instrument for the mitigation of harmonic distortions in three-phase electrical power systems. Numerous control methodologies have been established, encompassing approaches grounded in Artificial Neural Networks. Nevertheless, the conventional feedforward neural network, which has demonstrated efficacy in addressing various nonlinear challenges, has yet to be utilized with optimal performance in the execution of the SAPF control aimed at producing the reference currents for three-phase inverters. To validate the potential of this straightforward neural architecture, this study delineates an appropriate application strategy, predicated on the precise estimation of the Fourier coefficients associated with the fundamental harmonic of any distorted voltage or current. The aims included harmonic reduction and enhancement of the power factor. By utilizing simulations and experimental design, I evaluated practical scenarios, thereby confirming the feasibility of the recommended strategy.