The main objective of the electric utilities is to supply its customers with continuous
and sinusoidal voltage of constant magnitude. However, this is becoming more difficult as
the sizes and numbers of nonlinear loads are increasing rapidly. The injection of harmonics
by these nonlinear loads in power system leads to transformer heating,
communication interference, and solid-state device malfunctioning. Therefore, it is important to
reduce the dominant harmonics below 5%, as specified in IEEE 519-1992 harmonic
standard. There are two types of harmonic filters: passive and active. Passive filters are easy
to implement, but they depend significantly on load and source characteristics. Active
Power Filters (APFs) such as Shunt APF are complex, but can be used to effectively
remove harmonics. Hybrid topology consists of both active filters and passive tuned filters. In
this topology, the active filter improves compensation characteristics of passive tuned
filters relatively at lower cost than shunt active filter (Grady et al., 1990; Peng et al., 1990; and Akagi,
1994). However, their filtering characteristic strongly depends on the accuracy
of reference signal and its speed of computation. As such, numerous schemes have
been developed and studied for the control of hybrid active filters such as Fast Fourier
Transform (FFT), Kalman filter, and Artificial Neural Network (ANN)
(Pukhraj, 2001). The extraction by FFT leads to incorrect results if the signal is contaminated by noise and/or the
DC component is of decaying nature. The Kalman filter technique estimates the
harmonic components by utilizing a smaller number of samples and in relatively shorter time
as compared to FFT (Ramadan et al., 2001). However, Kalman filter technique
suffers from being computationally demanding due to transcendental function evaluations,
which makes it unfit for online applications such as active power filtering. The ANNs, based
on back propagation learning rule, are trained to estimate the harmonic components
(Mori, 1997; and John and Tim, 2002). This approach requires too much data for
training of ANN and leads to inaccurate results in the presence of random noise. ADALINE,
an adaptive Neural Network (NN) technique, has also been investigated for application
in APF. Its main advantages are speed and noise rejection (Lu, 1998; Rukonuzzaman
and Nakaoka, 2002; Rukonuzzaman et al., 2003; Vazquez and Salmeron, 2003; Karabag et al., 2004; Abdeslam et al., 2007; George, 2007; and Singh et al., 2007). A combination of NN and fuzzy logic-based control scheme has also been investigated with shunt
active filters. Most of these control schemes are complex and difficult to apply under
non-ideal conditions. In this paper, a three-phase three-wired NN-controlled parallel hybrid
active filter is proposed. The proposed controller is self-adapting, fast, and simple in
architecture and it can be successfully applied for harmonic filtering under various power
system operating conditions. The proposed controller's performance has been evaluated
under different non-sinusoidal and unbalanced source and load conditions. The details of
the controller have also been presented.
|
