Fault Diagnosis for Power System Transmission Line Based on PCA and SVMs - Intelligent Computing for Sustainable Energy and Environment - page 525

The wide use of controlchartsfor multivariatequality controlsituations is based
onthecorrelationsbetweenallthe processvariables.Oneapproachthathasproved
particularly powerful is the use of Principal Component Analysis (PCA) in com-
bination with T 2 and Q charts. PCA divides data information into the significant
patterns, such as linear tendencies or directions in model subspace, and the uncer-
tainties, such as noises or outliers located in residual subspace [3].
For the use of PCA techniques to fault detection and identification, Venkat has
summarized a comprehensive and systematic review about the process diagnosis
[4]. Recently, Zheng utilized fault diagnosis method in power system using multi-
class least square support vector machines classifiers and proved the effectiveness
on the basis of experiments.
Despite the reported progress in power system fault detection over the past
few years, two major problems still exist in the fast detection of the transmission
lines. One problem is that there are a lot of data collected from online monitoring
systems, which are multivariate and correlated. The other one is that there are
various types of faults in transmission lines. For real-time fault diagnosis in
transmission lines, the two problems should be considered simultaneously. This
is a problem to balance the real time implementation and the accuracy.
In this paper, a fault diagnosis approach based on PCA and SVM is proposed
to tackle the problem. In the proposed approach, a feature extraction algorithm
based on PCA is used to reduce the dimensionality. The PCA monitoring scheme
and the SVM are utilized to pinpoint the fault inception point and implement
fault recognition, respectively. The experimental results show that the proposed
approach is capable of detecting and recognizing the faults effectively.
This paper is organized as follows. Section 2 presents the preliminaries for
the proposed approach, followed by the description of the monitoring scheme in
Section 3. The application study for various fault situations are given in Section
4, as well as the discussion of the implementation of the algorithms. Finally, the
conclusion is given in Section 5.

2 Preliminaries

2.1 Principal Component Analysis

Let the data in the training set, consisting of m observation variables and N
observations for each variable, be stacked into a matrix X

N×m ,givenby

∈ R

⎡ ⎤

x 11 x 12 ··· x 1 m

⎣ ⎦

x 21 x 22 ··· x 2 m

X = (1)

. . .

···

x N 1 x N 2 ··· x Nm

N × k and a loading

PCA decompose the data matrix into a score matrix T ∈ R

m × k ( k

matrix P m is the number of retained principal components) [3].

Then the sample covariance matrix of the training set is equal to

∈ R ≤

S = X T X
N

(2)

−

1