Study on Identification of Power Quality Disturbances
Based on Compressive Sensing
*
Yue Shen, Hongxuan Wu, Guohai Liu, Hui Liu, Hanwen Zhang, Wei Xia
School of Electrical and Information Engineering, Jiangsu University, Zhenjiang 212013, China
shen@ujs.edu.cn
Abstract - Identification of power quality events is one of key
tasks in power system protection. This paper presents a new
approach based on compressive sensing (CS) for classifying
multiple power quality disturbances (PQD). First, every test
event sample of PQD is represented as a sparse linear
combination of training event samples using sparse
representation. A lower-dimensional random matrix is then
applied to both test sample of PQD and a CS-guided sensing
matrix derived from training samples to reduce dimensionality of
the linear combination expression. A L1-minimization solution
method is used to solve the sparse representation of every test
sample of PQD. Finally, the object class of the PQD event is
determined by the minimum of the residual error between test
sample and its sparse representation. Simulation and experiment
results show that the proposed CS-based method can effectively
extract features of PQD and has a high classification accuracy
rate with an average value larger than 95% under noise
circumstance for 10 types of PQD.
Index Terms - Power quality, disturbance classification,
compressive sensing, random matrix, dimensionality reduction
projection.
*
I. INTRODUCTION
The pollution problem of power quality is becoming
more and more serious. Feature extraction of various power
quality disturbances (PQD) is the precondition of taking
appropriate measures to improve power quality. Wavelet
*
This work is supported by the National Natural Science Foundation of China
(Grant #61301138), by the Professional Research Foundation for Advanced
Talents of Jiangsu University (Grant #10JDG136, #12JDG107) and by the
Priority Academic Program Development of Jiangsu Higher Education
Institutions (PAPD).
transform (WT) has good multi-scale analysis characteristics
and features of PQD can be effectively extracted by WT [1-2].
As a development of Wavelet transform, S-transform has
intuitional time-domain characteristics and well robustness
against noise[3-4]. However, both Wavelet transform and
S-transform exist many drawbacks such as huge calculation
complexity and excessive dependence to power quality signals.
Neural network [4-5], support vector machine (SVM) and
expert system methods are used in the classification of PQD
[6-8]. Among them, the SVM classification method can
effectively deal with small samples and has strong
generalization ability. But it needs to combine two-class
classifiers for multiclass ones and penalty factor parameters
are difficulty to be determined. In recent years, sparse
representation has been a popular research topic with the
emergence of compressive sensing (CS) theory [9-16]. In this
paper, the CS theory is introduced into the pattern recognition
of PQD. This paper studied the feasibility of random
dimensionality reduction projection (RDRP) and sparse
representation that applied to multiple classification of PQD
based on CS theory. Simulation and experiment results show
the effectiveness of the proposed CS-based PQD classification
method.
II.
Sparse representation of power quality disturbances
The key of sparse representation classification (SRC) lies
in selection of the sparse transform basis and foundation of the
sparse representation relationship. In this paper, PQD are
represented as a sparse linear combination of training samples
in an over complete dictionary[17-19]. The sparse relationship
between PQD and target samples is also built.
978-1-4799-5825-2/14/$31.00 ©2014 IEEE
Proceeding of the 11th World Congress on Intelligent Control and Automation
Shenyang, China, June 29 - July 4 2014