Page 485 - Decision Making Applications in Modern Power Systems

Pattern-recognition methods for decision-making Chapter | 17 445

any preprocessing techniques. It has simple procedures, but it suffers
from serious disadvantages. The instantaneous amplitude of measured
voltage and current signals highly depends on the power system condi-
tions, loading level, fault parameters, etc. The wide variation of input pat-
terns decreases the generalization ability of trained model. For example,
in a pattern-recognition-based fault classification scheme, when the input
feature vector is raw signals, any sudden change of current or voltage can
be misclassified as a fault. Moreover, large input data need more memory
space and has more computation burden. The raw data can be useful for
some simple function such as ground detection during a fault. Finding a
trend and unique behavior in original signals during an events in power
systems is a challenging issue.
Preprocessed data: The signal transforms process the signals obtained
from transducers to extract useful features. Sometimes the filtering pro-
cess is also used before the signal processing. Three types of filters are
used in protection functions of transmission line, that is, high-pass filter
such as elliptic filters [2], low-pass filter such as Butterworth filters [3],
and DC offset removal filter such as mimic filter [4]. Uncovering higher
or lower frequency components by filtering of the measured voltage or
current transients can improve the extracted features. Moreover, using the
DC removal and low-pass filters are inevitable to extract fundamental
components of power signals. Signal transforms generate representations
that are useful in various protection schemes, including harmonic analy-
sis, phasor measurement, noise reduction, filtering, frequency tracking,
and feature extraction. Signal transforms are developed in three domains:
frequency, time, and time frequency. In protection schemes, the trans-
forms in frequency and time frequency domains are more popular com-
pared to the transforms in the time domain. This is because of the facts
that the fault signals in power transmission lines contain usefulness infor-
mation in the frequency domain. Accordingly, some signal transforms in
the time domain which obtain the frequency content of a signal can be
interested for researchers. In this section a brief overview of signal trans-
forms is represented from the protection viewpoint:
1. One of the most important transforms in the frequency domain is dis-
crete Fourier transform (DFT). The DFT is used to derive a
frequency-spectrum representation of a time-varying input signal.
Using the coefficients obtained by the DFT, the fundamental and non-
fundamental components of input signals can be applied to protection
schemes. The application of fundamental phasor components of input
signals of relays is known in literature. Since faults in a power trans-
mission line create signals having a wide frequency range, nonfunda-
mental frequency components of input signals, as well as fundamental
frequency components, are used for feature extraction in pattern-
recognition-based protection schemes [3,5 7]. The DFT is a proper