Page 484 - Decision Making Applications in Modern Power Systems

444 Decision Making Applications in Modern Power Systems

Identification of main or backup protection: The time operation of a
pattern-recognition method can be defined based on the identifying of
main or backup protection schemes. Backup protection schemes have
more time to operate compared to the main schemes.
Identification of classification or estimation task: Some functions of a
distance relay are based on the classification purpose such as fault classi-
fication, and some functions are based on estimation purpose such as
fault location. An appropriate pattern-recognition model should be
designed based on classification or estimation task.
Identification of data which are available for protection schemes: The
power system network is an interconnected system, and hence, the mea-
sured signals, which are available through transducers at the relay loca-
tion, should be identified. Nowadays more data are available for
protection purposes due to developing some applications such as phase
measurement units (PMUs).
Preprocessing of data and construction of feature vector: Providing the
appropriate input patterns for pattern-recognitions methods has a vital
role to achieve desired performance. The aim of preprocessing is to
extract useful features from the original measuring that are fast to calcu-
late and also preserve suitable information.
Decision-making of the pattern-recognition-based protection scheme:
Finally, after the providing input patterns, the trained classifier or estima-
tor model makes a final decision. In supervised models, the classifier or
estimator model is trained using the training data and the desired output
can be a class label (in classification problems) or continues variable
(in estimation problems).

17.2.1 Feature extraction
Generally, a pattern-recognition method employs two basic subroutines,
including feature extraction and decision-making. The process of feature
extraction can be considered as an intermediary formulation to constrict a
large vector of data to a small vector of attributes. In other words, based on
definition in Ref. [1], “a feature of a given parameter set refers to an attri-
bute described by one or more elements of the original pattern vector.”
These features construct the input vector of a pattern-recognition method.
The feature vector constructed through different preprocessing techniques
improves the performance of decision-making. In pattern-recognition-based
schemes for protection of transmission lines, the features may include the
following:
Raw data: Since the voltage and current signals are available for distance
relays, the instantaneous amplitudes of these signals can be used as a vec-
tor of input feature. In this category of the feature, there is no need to