Power harmonic identification and compensation with an artificial neural network method

This paper introduces a new neural method for harmonic identification and compensation. Based on Adaline networks, the proposed method is called the diphase currents method. The architecture and the learning are formulated based on an original decomposition of the disturbed currents. These currents are converted in the alpha beta-or DQ-spaces to separate each harmonic component in a linear expression. In this harmonic compensation method, the harmonic components may be individually selected and the reactive power may be compensated. The proposed method is robust and has been efficiently compared to other conventional and neural harmonic compensation methods. In order to validate the performance of the diphase currents method, simulation studies are carried out in the presence of plant variations. Experiments are also presented to show the performance of the proposed neural method under many practical industrial conditions.