Diapositiva 1

1
Automatic Power Quality Disturbances Detection
and Classification Based on Discrete Wavelet
Transform and Support Vector Machines
Authors MI (C). Valdomiro Vega García Dr.
Gabriel Ordóñez Plata MPE. César A. Duarte
Gualdrón
Universidad Industrial de Santander
2
CONTENT

• Introduction
• Wavelet transform
• Algorithms
• Power quality disturbances
• Detection strategy
• Identification strategy
• Automatic classification strategy
• Simulation results
• Conclusions

3
INTRODUCTION
To diagnose the quality of electric energy
service
Electromagnetic disturbances
-
Losses

• Electric sector
• Industry
• Trading
• Domestic

4
INTRODUCTION
OBJECTIVES

• To determine patterns based on discrete wavelet
  transform to identify voltage sags, swells,
  transients, harmonics and flicker.
• To establish a detection - time location
  strategy.
• To set up an automatic classification strategy

5
DISCRETE WAVELET TRANSFORM
6
DISCRETE WAVELET TRANSFORM
ORTHONORMALITY

• Signal decomposition

Orthogonal

• No energy cross

7
ALGORITHMS
DECOMPOSITION SCHEME
RECONSTRUCTION SCHEME
8
ALGORITHMS
Signal decomposition into approximation and
detail sequences
9
ALGORITHMS
FILTERS USING WAVELET DAUBECHIES 4
10
ALGORITHMS
MULTIRESOLUTION ANALYSIS
11
POWER QUALITY DISTURBANCES
12
DETECTION STRATEGY
DISTURBANCE DETAILS AT FIRST LEVEL
13
IDENTIFICATION STRATEGY
ENERGY DEVIATION OF WAVELET COEFFICIENTS
14
SHIFT NO-INVARIANT PROPERTY
15
AUTOMATIC CLASSIFICATION STRATEGY
BAYES DECISION TECHNIQUE
K classes wk X Input Vector P(wk/X) A
posteriori probability P(wk) wk class
probability P(X/wk) a priori probability
JARQUE-BERA TEST 89 of rejection
16
AUTOMATIC CLASSIFICATION STRATEGY
ARTIFICIAL NEURAL NETWORK (ANN)
17
AUTOMATIC CLASSIFICATION STRATEGY
ARTIFICIAL NEURAL NETWORK (ANN)
18
AUTOMATIC CLASSIFICATION STRATEGY
SUPPORT VECTOR MACHINES SVM
19
SIMULATION RESULTS
SUCCESS PERCENTAGES BAYES vs. KOHONEN LVQ
200 signals
20
SIMULATION RESULTS
SUCCESS PERCENTAGES ANN-PERCEPTRON vs. SVM
200 signals
21
SIMULATION RESULTS
SUCCESS PERCENTAGES ANN-PERCEPTRON vs. SVM
Other 200 signals
22
CONCLUSIONS

• A DWT SVM automatic classification strategy has
  been implemented.
• Based on the energy of DWT detail coefficients is
  possible to identify voltage disturbances.
• SVM could be the best classifier for patterns
  obtained in this work. Though, ANNs (supervised)
  display good performance.

23
CONCLUSIONS

• For most disturbances classes the success
  percentage was better than 90 in spite of
  pattern resemble.
• A disturbance database (17 489 signals) was
  generated for training, validating and evaluating
  each classification scheme.

24
QUESTIONS?
25
ELECTRIC POWER SYSTEMS RESEARCH GROUP UNIVERSIDAD
INDUSTRIAL DE SANTANDER Carrera 27, Calle 9.
UIS Bucaramanga Colombia. PO BOX. 678 PBX
(57 7) 6344000, Ext 2360 - 2361 2703 -
2472 TEL [phone] FAX
(57 7) 6359622 BUCARAMANGA
COLOMBIA http//www.uis.edu.co/investigacion/pagin
as/grupos/gisel.htm gaby_at_uis.edu.co,
cedagua_at_uis.edu.co, valdomirovega_at_ieee.org