Page 490 - Decision Making Applications in Modern Power Systems

450 Decision Making Applications in Modern Power Systems

17.2.3.1 Classification
In a transmission line, when the input signals are sampled at the relay loca-
tion, some preprocessing is made to construct the input feature vector, and
the vector is fed into the classifier. This pattern-recognition-based relay is
allowed to determine that the input signals are belong to which output class.
The classification procedure contains two main sequential steps: training and
testing steps. The matrix of input data set including feature vector and target
vector can be divided in two parts: use one for training and the remaining is
for testing. Therefore the testing data set is unseen during the training step.
In the training step, the classification model is learned through training data
set, and hence, a learned classifier will be built. The testing date set is
employed to evaluate the classification accuracy of a learned classifier.
Using the classification function, the class labels for the testing data set are
predicted. The classification accuracy is computed as percentage of correct
predictions. Moreover, a confusion-matrix can be developed when more
details about the performance of the classifier are required in order to dis-
cover where the classifier is failing [21]. To obtain higher accuracy of classi-
fication, the one of the best ways is to test out different classification
algorithm and also examining different parameters corresponding to each
algorithm. The best selection can be determined through K-fold cross-
validation [1].
Different classification functions can be applied as a classifier in pattern
recognition. Since the final decision of a protection scheme is made through a
classifier, the selection of classifier among various classifiers is a challenging
issue. On the other hand, there is no single predictor superior to its rivals for
solving all the problems. However, evidence suggests and confirms that the
extraction of right features is the most important factor for proper designing a
pattern-recognition method. The classification algorithms can be categorized
into binary and multiclass algorithms. Binary classification is the task of clas-
sifying the instances of an input data set into two classes. Multiclass classifi-
cation is the task of classifying the instances of input data set into three or
more classes. It is worth mentioning that the binary classification can be used
in multiclass problems based on class-versus-class and one class-versus-rest
techniques. Some known classifiers that are used to decide in pattern-
recognition-based relays are artificial neural networks (ANN), support vector
machines (SVMs), probabilistic neural network (PNN), decision tree (DT),
k-nearest neighbor (k-NN), random forest (RF), etc.

17.2.3.2 Prediction
In some functions of protection relays of transmission line, the real-value
estimation is required. For example, one of the solutions to find the location
of the fault is pattern-recognition-based methods [22]. In classification pro-
blems, the pattern-recognition model predicts discrete outcomes, for