Page 191 - Decision Making Applications in Modern Power Systems

154 Decision Making Applications in Modern Power Systems

An adaptive filter is a system with a linear filter that has a transfer func-
tion controlled by variable parameters and a means to adjust those para-
meters according to an optimization algorithm. The property of an adaptive
filter is self-modifying its frequency response to change the behavior in time,
allowing the filter to adapt the response to the input signal characteristics
change. The adaptive filters have various applications such as echo cancella-
tion in the telephones, signal processing in the radars, navigation systems,
biometric signal processing, navigation signals, and communications channel
equalization. The main purpose of an adaptive filter in noise cancellation is
to eliminate the noise from a signal adaptively to improve the signal-to-noise
ratio (SNR). For accurate assessment of PQ events in the presence of noise,
tremendous efforts are being made for a long period of time to design effec-
tive and robust algorithms. The occurrence of harmonics frequently causes
communication interference, resonance of mechanical devices, and melting
of magnetic parts of electrical appliances, etc. Nowadays, the detection and
removal of harmonics using appropriate harmonic filter and forecasting of
PQ disturbances are key research aspects of power engineers. The methods
for detection of PQ events have been classified into two types; those are
parametric and nonparametric methods. Fourier transform (FT), wavelet
transform (WT), S-transform (ST), H-transform, etc., all come under the non-
parametric methods and are restricted by the length of the data. Adaptive fil-
tering is popular parametric estimation technique to track and estimate the
PQ events. A common adaptive filter design is based on transversal filter
with adaptive weight update mechanism such as least mean square (LMS)
adaptive filtering algorithm that has been widely used due to its simplicity
and numerical robustness. On the other hand, normalized LMS (NLMS) and
recursive least square (RLS) give better convergence properties than LMS.
To estimate amplitude, phase, and frequency, an extended Kalman filter
(EKF) has been implemented. Further the EKF approach has the advantage
that the estimates are computed recursively using one-step prediction [3 5].
In general, several researchers in this area have applied one of the well-
known signal-processing techniques to extract the features and complete the
classification process by using an artificial intelligence technique as a classi-
fier. The signal-processing techniques give some redundant features that
affect the efficiency of the classifiers. Moreover, there is no discussion on
how to set the best parameters for the classifiers. Only few researchers have
attempted optimization techniques for selecting the suitable feature subset
and selection of parameter. In this view, signal-processing techniques for fea-
ture extraction and artificial intelligent techniques for the classification are
the most important parts of the pattern recognition of PQ disturbances. The
feature extraction stage provides a set of statistical data to make the analysis
more effective. The set of feature extraction is then used as input for the
classification system. In spite of technical advancement in signal-processing
techniques, the proper selection of feature extraction is still a challenge.