FPGA implementation of harmonic detection methods using Neural Networks

This work presents a Neural Networks based intelligent Active Power Filter control unit for harmonics current elimination. The paper is centered on an Improved Three-Monophase (ITM) method and a novel one called Two-Phase Flow (TPF) method for detecting harmonics from disturbed currents due to non linear loads in electric power systems. The TPF method introduces new currents decomposition in the DQ-space that results in separating AC from DC components. The DC terms are estimated by an Adaline Neural Network because of its learning capabilities with respect to real-time applications. Then the resulting AC terms from this approach are transformed to obtain the reference currents. After analyzing those two methods with respect to their performance and hardware resources consumption by means of Altera Dsp Builder^reg, a comparative study with the direct method is reported and some FPGA implementation results are also shown. From there, one can notice that the direct method is the simplest and offers the best performance whereas the TPF is the fastest and consumes less material resources.