Hamonic Detection in Active Filter by Using Neural Network

Abstract

Recently, active power filter has been studied as it enables the harmonic compensation according as the fluctuation of harmonic patterns. The compensation of active filter is conducted by detecting harmonic component from load current, then inject the equal-but-opposite harmonic current into power line by inverter. Therefore the accuracy and the speed of harmonic detection is important for the performance of active filter. In case that it has to specify the frequency of harmonic for high performance compensation, it needs to analyze load current into each kind of harmonic by using such the Fourier transformation. Fast Fourier transformation(FFT) enables high speed analysis, however, over one cycle of wave form is required, then the compensation is delayed more than two cycles of current wave.
The purpose of this paper is to improve the harmonic detection based on the frcquency decomposition by using artificial neural network. A half cycle of current wave form is used for harmonic analysis, the amplitudes at time series is the input signals of neural network. Each node in output-layer is corresponded for each kind of harmonic, the content of each harmonic is detected as the signals of the corresponding node. As neural network enables parallel processing, the content of each harmonic component is speedy detected within a half cycle. That means it is possible to compensate the harmonic in each half cycle of current wave. Therefore, the response time of active filter is improved more than 4 times comparing with the usual method. As the simulation results, we conclude that neural network is available to be used as the harmonic detector. By the proposed algorithm, it is possible to improve the performance of active power filter.