Synchronous Reference Frame Based Active Filter Current Reference Generation Using Neural Networks

The increased use of nonlinear devices in industry has resulted in direct increase of harmonic distortion in the industrial power system in recent years. The significant harmonics are almost always 5th, 7th, 11th and the 13th with the 5th harmonic being the largest in most instances. Active filter systems have been proposed to mitigate harmonic currents of the industrial loads. The most important requirement for any active filter is the precise detection of the individual harmonic component's amplitude and phase. Fourier transform based techniques provide an excellent method for individual harmonic isolation, but it requires a minimum of two cycles of data for the analysis, does not perform well in the presence of subharmonics which are not integral multiples of the fundamental frequency and most importantly introduces phase shifts. To overcome these difficulties, this paper proposes a multilayer perceptron neural network trained with back-propagation training algorithm to identify the harmonic characteristics of the nonlinear load. The operation principle of the synchronous-reference-frame-based harmonic isolation is discussed. This proposed method is applied to a thyristor controlled DC drive to obtain the accurate amplitude and phase of the dominant harmonics. This technique can be integrated with any active filter control algorithm for reference generation