Evaluation of Harmonic Trends Methods - IEEEewh.ieee.org/conf/tdc/Evaluation_of_Harmonic_Trends...Harmonic Trend Analysis {A trend is normal if 95% of the times the data points in

$Page 1: Evaluation of Harmonic Trends Methods - IEEEewh.ieee.org/conf/tdc/Evaluation_of_Harmonic_Trends...Harmonic Trend Analysis {A trend is normal if 95% of the times the data points in$
Evaluation of Harmonic Trends using Statistical Process Control Methods

S. Santoso, University of Texas at Austin D. D Sabin, EPRIM. F McGranagham, EPRI

IEEE PESTransmission and Distribution CONFERENCE and EXPOSITION, April 21st – 24th, 2008

McCormick Place

C H I C A G O

Practical Applications of intelligent data mining in power distribution systems

Introduction

Power quality data collected are generally voluminous and have to be analyzed in an efficient manner.

Variations in voltage and current harmonic distortions can be normal or abnormal.

Paper demonstrates a statistical process control method to determine if the statistical variations in harmonic trend can be considered a normal variation.

Why Monitor Power Quality? (Role of Advanced PQ Monitoring)

• Benchmark system performance levels— Understand power quality that can be expected— Evaluate performance with respect to standards— Provide baseline for premium services

• Reliability reporting — (Reliability defined based on customer impacts)— Service Quality Index

• Improve system operations and reliability• e.g. Fault Location

• Identify and solve problems • Equipment diagnostics

Example Power Quality Monitoring System

Example of SystemMonitoring Concept

Transmission

DistributionSubstation

SubstationMonitoring

System

CustomerMonitoring

System

CustomerMonitoring

System

MonitoringDatabase

Local Network

Data Collection Data Collection

Database Management/Local Data Analysis

CorporateIntranet

Internet/World Wide Web

Power Quality/Energy

Information Service

Power Quality/Reliability

Performance andData Analysis

Important Characteristics of a Monitoring System

Open architectureSystems should allow integration of different technologies within utility and customer networks

Standard Data Formats for exchanging data

Power Quality Data Interchange Format (PQDIF)COMTRADE

Web-based access to the informationIntelligent applications (timeliness of information becomes important)Automated reporting functions

OperationInformation from

SCADA LogsDataCharacterizer

Site & MeterInformation

EventViewer

ProtectionAnalysis

DataTrending

Bench-Marking

StatisticalAnalysis

PQView Applications

DataTranslator

DataTranslator

DataTranslator

Power Quality& Flicker Meters

Digital FaultRecorders & Relays

RevenueMeters

PQViewDatabase

ReportWriting

PQView® Architecture

Example of Typical Configuration that allows integration and advanced applications

Steady State Concepts

Steady State Concepts

Compatibility level

Assessed level

Disturbance magnitude

time

Equipment immunity test levels

Utility planning levels

0

0.5

1

1.5

2

2.5

00h00 24h00

95% or 99%of 10-minute values

10 minute values

Pst

Measurement period: 7 days

Highest phase

Using monitoring system to assess steady state PQ levels

10-Minute Min-Avg-Max VTHD Values for One Year

Statistical Process Control

There are two types of statistics

• Descriptive statistics

• Inferential statistics

Descriptive statistical analysis can be used to detect equipment problems. However failure or changes in the system condition must be known in advance or when the failure has already taken place

Inferential statistics is the appropriate for analyzing steady state power quality data

Run-chart and control chart analyses form a part of inferential statistics

Control Chart Analysis

Control chart analysis method uses control limits namely the upper control limit (UCL) and lower control limit (LCL).

Control limits act as boundary to distinguish between common causes and special causes for variation in trend.

The control limits are generally 0.001 probability limits.

Harmonic Trend Analysis What is normal variation?

To use the control chart method, one must specify “normal trend” or “normal variation” of the data

Question: How does a normal variation be defined?

Users to provide “reference data.” Based on this reference data, statistical characteristics or limits are derived. Users can simply give dates (e.g., 1/1/06 – 2/15/06) to indicate the reference data. All other data behavior will be compared to this reference data using the control chart method.

Harmonic Trend Analysis General Procedures – Users to specify reference data and compute UCL

Example: the first 4½ weeks of data as reference data.

Control analysis method is performed to determine upper control limits (UCL):

μx and σx are the mean and standard deviation of the variations, respectively

Control limits are generally based on 0.001 probability limits.

If random causes alone were responsible for the variation in the trend, the probability of a point falling beyond the control limits would be one out of a thousand.

xxUCL σμ ⋅+= 3

1 5 9 13 17 21 25 29 33 37 41 45 490

100

200

300

400

500

600

700

Week

Vol

ts

Voltage 5th Harmonic Order for the Entire Observation Period

UCL = 156.95 V

Harmonic Trend AnalysisA trend is normal if 95% of the times the data points in the harmonic distortion time-series are below the UCL

Segment violation is triggered if less than 95% of the data points are below the reference UCL

For example shown, variation is stable between the 5th and the 12th week

From week 13 and forward, many data points are above UCL and hence trend could be abnormal

1 5 9 13 17 21 25 29 33 37 41 45 490

100

200

300

400

500

600

700

Week

Vol

ts

Voltage 5th Harmonic Order for the Entire Observation Period

UCL = 156.95 V

Fig. 3. The overall fifth order harmonic voltage distortion data with a UCL computed based on the first 4 ½ weeks of

reference data

Harmonic Trend Analysis

Parameters used to analyze the trend are

Time Interval

Acceptable Probability

Permissible Consecutive Segment Variations

Upper Control Limit (UCL)

Example #1: using VTHD Data

Selected time span:

From: 01-Jul-1994, 00:13:00 To: 31-Dec-1994 , 18:40:00

0.4

0.6

0.8

1.0

1.2

1.4

Jul1994

Aug Sep Oct Nov Dec Jan 95

SITE1 - V THD Afrom 6/30/1994 4:06:30 PM to 1/3/1995 10:32:23 PM

Electrotek/EPRI PQView®

V T

HD

A (%

)

Time

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

Jul1994


SITE1 - V THD Bfrom 6/30/1994 4:06:30 PM to 1/3/1995 10:32:23 PM


V T

HD

B (%

)

Time

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

Jul1994


SITE1 - V THD Cfrom 6/30/1994 4:06:30 PM to 1/3/1995 10:32:23 PM


V T

HD

C (%

)

Time

Example Procedure (1) Control Chart Analysis of Reference VTHD Data

There are 4 week of data points in the reference. Each point corresponds to VTHD taken every 30 minutes. Total number of data points for the reference amounts to

(4 * 7 * 24 * 60 ) / 30 = 1344 data points

Phase Average of%VTHD

UCLµx + 3(σx)

Data points above UCL

A 0.65% 0.93% 0

B 0.78% 1.09% 1

C 0.60% 0.87% 0

Concatenated(all phases linked together)

0.68% 1.04% 1

Use this number as the upper control limit for VTHD for all phases

Example Procedure (2) 4-Week Reference VTHD Data and Control Chart Analysis

Dat

a ra

nge:

1 J

uly

1994

–28

Jul

y 19

94

12345671234567123456712345670

0.2

0.4

0.6

0.8

1

1.2

day

%TH

D

Reference THD phase A

Avg: 0.65%UCL: 1.04%

12345671234567123456712345670

0.2

0.4

0.6

0.8

1

1.2

day

%TH

D

Reference THD phase B

Avg: 0.78%UCL: 1.04%

12345671234567123456712345670

0.2

0.4

0.6

0.8

1

1.2

day

%TH

D

Reference THD phase C

Avg: 0.60%UCL: 1.04%

THDUCLAvg

THDUCLAvg

THDUCLAvg

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 260

0.2

0.4

0.6

0.8

1

1.2

THD For Phase A

Week

%TH

D

UCL: 1.04%

THDUCL

Example Procedure (3) Trend analysis of VTHD A (for entire data record)

Data to trend:1 July 1994 – 31 Dec 1994

Missing data

Reference data

UCL = 1.04% (computed based on the reference data)

RESULTS: Number of data points above the UCL is 1.30%

Conclusion for VTHD A:

“98.70% of the time, VTHD A is below 1.04%.”

Note: Missing data are not included in the trend analysis

NORMAL TREND

Example Procedure (4) Trend analysis of VTHD B: (for entire data record)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 260

0.2

0.4

0.6

0.8

1

1.2

THD For Phase B

Week

%TH

D

UCL: 1.04%

THDUCL



Conclusion 1 for VTHD B:

“96.07% of the time, VTHD B is below 1.04%.”


Missing data

Reference data


NORMAL TREND

Example Procedure (5) Trend analysis of VTHD C: (for entire data record)



Conclusion #1 for VTHDC:

“94.73% of the time, VTHD C is below 1.04%.”

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 260

0.2

0.4

0.6

0.8

1

1.2

THD For Phase C

Week

%TH

D

UCL: 1.04%

THDUCL


Missing data

Reference data


ABNORMAL TREND

Conclusion for Example #1Based on the analysis, THD variation in Example #1 is considered abnormal. This is due to the persistence increase in V THD C .

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

Jul1994


SITE1 - V THD Cfrom 6/30/1994 4:06:30 PM to 1/3/1995 10:32:23 PM


V TH

D C

(%)

Time

Example 2: Database (VTHD)

0.8

1.0

1.2

1.4

1.6

1.8

2.0

2.2

2.4

2.6

Apr2001

May Jun Jul Aug Sep Oct Nov DecJan 2002Feb Mar

GCV4604M - V THD Afrom 3/31/2001 12:03:30 PM to 2/28/2002 9:31:08 PM


V T

HD

A (%

)

Time

1

2

3

4

5

6

Apr2001


GCV4604M - V THD Bfrom 3/31/2001 12:03:30 PM to 2/28/2002 9:31:08 PM


V T

HD

B (%

)

Time

1.0

1.5

2.0

2.5

Apr2001


GCV4604M - V THD Cfrom 3/31/2001 12:03:30 PM to 2/28/2002 9:31:08 PM


V T

HD

C (%

)

Time

Selected time span:From: 31-Mar-2001, 19:25:15 To: 22-Feb-2002, 14:02:07

Example 2 Database. (2) Control Chart Analysis of Reference VTHD Data

There are 4 week of data points in the reference. Each point corresponds to VTHD taken every 30 minutes. Total number of data points for the reference should amount to

(4 * 7 * 24 * 60 ) / 30 = 1344 data points

but there are 2 missing data between 1 Apr 2001 01:55:28 and 1 Apr 2001 03:25:20

Phase Average of%VTHD

UCLµx + 3(σx)

Data points above UCL

A 1.33% 2.18% 1

B 1.17% 2.07% 12

C 1.37% 2.15% 13

Concatenated 1.29% 2.17% 9

Example 2 Database (3) 4-Week Reference VTHD Data and Control Chart Analysis

Dat

a ra

nge:

31

Mar

200

1 –

22 F

eb 2

002

12345671234567123456712345670

0.5

1

1.5

2

day

%TH

D

Reference THD phase A

Avg: 1.33%UCL: 2.17%

THDUCLAvg

12345671234567123456712345670

0.5

1

1.5

2

day

%TH

D

Reference THD phase B

Avg: 1.17%UCL: 2.17%

THDUCLAvg

12345671234567123456712345670

0.5

1

1.5

2

day

%TH

D

Reference THD phase C

Avg: 1.37%UCL: 2.17%

THDUCLAvg

Example 2 Database (4) Trend analysis of VTHD A: (for entire data record)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 460

1

2

3

4

5

6

THD For Phase A

Week

%TH

D

UCL: 2.17%

THDUCL


Missing data

Reference data


RESULTS: Number of data points above the UCL 0.60%

Conclusion for VTHD A:

“99.40% of the time, VTHD A is below 2.17%.”

Missing data

NORMAL TREND

Example 2 Database (5) Trend analysis of VTHD B: (for entire data record)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 460

1

2

3

4

5

6

THD For Phase B

Week

%TH

D

UCL: 2.17%

THDUCL



Conclusion #1 for VTHD B:

“99.42% of the time, VTHD B is below 2.17%.”

Missing data

Reference data

Missing data

NORMAL TREND

Example 2 Database (6) Trend analysis of VTHD C: (for entire data record)



Conclusion #1 for VTHD C:

“98.68% of the time, VTHD C is below 2.17%.”

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 2324 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 460

1

2

3

4

5

6

THD For Phase C

Week

%TH

D

UCL: 2.17%

THDUCL

Missing data

Reference data

Missing data

NORMAL TREND

Conclusion for Example #2

Based on the analysis above, THD variation in Example #2 is considered normal.

Conclusion

Control chart analysis can be used to analyze the statistical behavior of steady-state power quality dataThis method can be used to determine if the harmonic trend is caused by normal variation or abnormal variation

Documents

Evaluation of Harmonic Trends Methods - IEEEewh.ieee.org/conf/tdc/Evaluation_of_Harmonic_Trends...Harmonic Trend Analysis {A trend is normal if 95% of the times the data points in