1. Hidden Failure Detection in Protection Systems via Centralized Substation Protection
Sakis Meliopoulos*, George Cokkinides*, Hussain Albinali**, Paul Myrda*** and E. Farantatos***
* Georgia Institute of Technology, Atlanta, Georgia USA
**Aramco, SA
*** Electric Power Research Institute, USA

2. Presentation Outline Objectives and Driving Forces
Error Correction within Merging Units
Substation Centralized Protection Scheme
Detection of Hidden Failures
Real Time Correction (self healing)
Numerical Examples
Conclusion

3. Introduction Objectives: Driving Forces:
Core issues in Centralized Substation Protection: (a) ensure accuracy of input data to protection functions and (b) Monitor P&C systems for detecting problems that affect zone protection such as hidden failures, incorrect data, etc.
Driving Forces:
Minimize Protection system Mis-Operations
About 10% relay mis-operations (65% due to hidden failures)
Reduce Protection and Control System Complexity
Uncertainties in input data require safety margins in protection settings
Detection of Hidden Failures and Recovery
Limited Detection Capability Exists  Block Relay Operation
Need Ability to Self-Correct
Technological Advancements
Separation of data acquisition from logical functions
Advances in software technology
Communication infrastructure

4. The Setting-Less
Protection Method

5. DSE Motivation: In Search of Secure Protection
Setting-less Protection can be viewed as Generalized Differential protection
Analytics: Dynamic State Estimation (systematic way to determine observance of physical laws)

6. The Setting-Less Protection Method
(Dynamic State Estimation Based Protection)
Protection Zone State x(t): Minimum Number of Variables that Completely Define the Operating Condition of the Protection Zone, i.e. Voltages, Flux, Speed, etc.
Measurements z(t): Any Measurement Can be Expressed as a Function of the State x(t), i.e. z(t)=h(x(t)).
State Estimation: Compute x(t) that will make the difference: z(t)-h(x(t)) as small as possible. Note z(t) are Sample Values

7. The Setting-Less Protection Method
(Dynamic State Estimation Based Protection)
If Protection Zone is Free of Faults, the distance (z(t)-h(x(t))) is very small – comparable to Measurement Error.
If Protection Zone is Free of Faults and there are faults outside the protection zone, the distance (z(t)-h(x(t))) is very small – comparable to Measurement Error.
If Protection Zone experiencing an internal Fault, the distance (z(t)-h(x(t))) becomes large.

8. Protection of Saturable Core Transformers
Event:
14.4/2.2kV, 1000 kVA single- phase saturable-core transformer
A 800kW load connected to the system, generating transients and inrush currents at 0.72 seconds
Followed by a 5% turn-ground fault near neutral terminal of the transformer at 1.52 seconds
List of Measurements:
Single-phase voltages at two sides
Single-phase currents at two sides
Temperature measurements at selected points
Internal Turn-Ground Fault

9. Differential Protection for Turn-ground Fault 5% from Neutral
Differential Current (SV)
Operating Current
Restraining Current
Differential Ratio Index (%)
Trip Decision
Legacy Differential Protection MISSES the Fault

10. Setting-less Protection for Turn-ground Fault 5% from Neutral
Voltage Measurement
CurrentMeasurement
DSE Residuals
Chi Square Test
Confidence Level
Trip Decision
DSE Execution Time
Setting-Less protection detects fault in sub-msec time
and trips within a fraction of a cycle.

11. Effects of Input Data Accuracy
Quality of Data is Affected from (a) Instrumentation Channel Errors and (b) Hidden Failures. Both Affect Performance of Setting-less relays as any other legacy relay.
Relays and merging units are becoming more accurate by using higher resolution in data acquisition and higher sampling rates.
Errors from instrumentation channels remain practically the same. Instrumentation channel errors have been much higher than the errors introduced by the data acquisition even in earlier generations of sensor less systems.
Merging Units offer a unique opportunity to perform error correction within a merging unit  MU provides corrected data in primary quantities.

12. Basic Question
Given a measurement at the secondary of an instrumentation channel, can we extract the correct value of the primary quantity?
Can it be done on a sample by sample basis?

13. Basic Approach to Error Correction
Construct the mathematical model of the instrumentation channel: Example CT Channel
Use actual measurements and augment with:
(a) virtual measurements,
(b) derived measurements, and
(c) pseudo measurements
Perform dynamic state estimation  Best Estimate Of Primary Current
Redundant Measurements.

14. Dynamic State Estimation
Measurements
Actual: - Voltage or Current at Burden
Add: -Virtual,
-Derived, and
-Pseudo-Measurements
CT Channel math Model
Example Virtual Measurement:
Example Derived Measurement:
Example Pseudo Measurement:
States: 15 Actual Measurements: 1
Virtual Measurements: 14
Derived Measurements: 5
Pseudo measurements: 1
21 Equations in 15 unknowns.
One of the unknowns is the Primary Current

15. Dynamic State Estimation
Subject to:

16. Example System Example System Parameters 115-kV transmission system.
Current transformer on phase A
CT ratio is 800:5A
Error class: 10C100.
CT instrumentation cable: #10 copper
Instr. cable length: 96 meters
Burden resistance: 0.1 Ω
Current Transformer Parameters

17. Example System Results

18. Phase A to ground fault at “MIDBUS”
Example System Results – Mild CT Saturation Simulated CT Secondary Current Waveform
CT Secondary Current
Phase A to ground fault at “MIDBUS”

19. Example System Results – Mild CT Saturation Error Correction Performance
Secondary Current x CT Ratio
Estimated Primary Current
Actual Primary Current
Estimated results match actual primary current with less than 1% error

20. Phase A-to-C fault at “MIDBUS”
Example System Results – Deep CT Saturation Simulated CT Secondary Current Waveform
Phase A-to-C fault at “MIDBUS”

21. Example System Results – Deep CT Saturation Error Correction Performance
Secondary Current x CT Ratio
Estimated Primary Current
Actual Primary Current
Estimated results match actual primary current with less than 2.5% error

22. Limitations in Securing Accurate Data
Setting-less relays work well when input data are accurate. Accuracy of Input data is improved with error correction methods. Error Correction methods work well when instrumentation is healthy.
In the presence of hidden failures input data are laced with huge errors and error correction methods do not have enough information to correct the condition. Data are compromised.

23. Impact of Hidden Failures
Hidden failures corrupt the data “seen” by a relay, legacy relay or setting-less protective relay. Hidden failures will cause relay mis-operation whether it is a legacy protective relay or a setting-less relay. Need to identify hidden failures and avert relay mis- operations. Present State of Art: Some legacy relaying schemes can identify some hidden failures and inhibit relay operation. No capability to take corrective action.
The setting less relay process all the measurements and preform DSE to compute the best estimate of the zone state.
For every time sample the relay measure the consistency between the measurements and the zone model through the Chi-Square Test which computes the probability that the measurement is consistent with model. This probability is known as confidence level. High confidence level (90%)indicates healthy zone otherwise, it is faulty one.
The advantage of this technique is that its independent about the system and does not require coordination's.
However, the system is vulnerable to hidden failures especially in the instrumentation channels which might generate inaccurate measurements and cause the relay to misoperate. In order to secure the setting-ess relay from misoperation we are proposing a centralized protection scheme.

24. Proposed Method for Securing Data

25. Proposed Method for Securing Data

26. Dynamic State Estimation Based Centralized Protection Scheme (DSEBCPS)
Overview
This slides shows that the architecture of the proposed DSEBCPS. It shows that process started by data streaming of the measurement as sampled values from the MU to settingless relays which process the data to perform zone level protection and compute phasor quantities. The phasor quantities are streamed to the DSEBCPS via the station bus at a rate of 1/cycle. The DSEBCPS processes the data to preform the DSE and hypothesis testing.
It is important to note that preforming the DSE requires the substation model. Therefore, we added a module to compute the substation model from the individual zone models. Substation model is used to derive the measurements model at the beginning of the process or during any changes in the substation topology.

27. Dynamic State Estimation Based Centralized Protection Scheme (DSEBCPS)
Hypothesis Testing: Observations
At substation level redundancy is high (over 2000%)
System is continuously running.
Probability of simultaneous failure events is low
Hypothesis Testing: Mechanics
Identify suspect measurements from residuals
Group suspect data with certain criteria (see paper)
Determine “faulted devices” from setting-less relays output
The hypothesis testing is activated when the DSEBCPS detected abnormality. It is used to detect hidden failures or power faults. The principle is capitalizing on the fact that at substation level we have a higher degree of redundancy in the measurements which means more measurements than the substation states. In real substation this redundancy might reach 2000%. If the confidence level is low ,this redundancy results in higher value in measurement residual for the measurements that suffer from abnormality. Therefore, the measurements with higher values of normalized residual (We are using threshold of 2) classified as suspicious measurements. Then theses measurements are eliminated one at time and re-perform the DSE. If the elimination process reveals high confidence level, abnormality is detected in the eliminated measurements.
We have defined a common mode criteria to distinguish between hidden failures and power fault.
The three phase common mode criteria verifies the adjacent phases of the measurements with the highest normalized residua. If any of them exceeds the threshold of 2, it will be included in the elimination process. If this elimination process reveals high confidence level, hidden failure is detected in the instrumentation channel associated with the eliminated measurements.
The device common mode criteria capitalizes on the fact that most of measurements are modeled using the device model. Therefore, if the measurement with highest normalized residual is modeled by one device, the device common mode criterion verifies the normalized residual for all measurements modeled by the same device. If all of them exceed the threshold of 2, they will be included in the elimination process. If this elimination process reveals high confidence level, power fault is detected in the device.
If the confidence level is still low, the measurement with the second highest normalized residual that was not part of the previous elimination process is selected as suspicious measurement. The process is repeated until we get high confidence level.

28. Dynamic State Estimation Based Centralized Protection Scheme (DSEBCPS)
Hypothesis Type 1 (H1): Remove suspect measurements and rerun DSE. If probability high  removed measurements are bad  identify root cause  issue diagnostics  replace bad data with estimated values. End hypothesis testing. Otherwise go to H2.
Hypothesis Type 2 (H2): (determine if a fault decision is correct). For the reported faulted device, remove all internal device measurements and remove the faulted device model from the substation model. Then rerun DSE. If probability high  the device is truly experiencing an internal fault. Allow zone relay to trip the faulted device. End hypothesis testing.
Hypothesis Type 3 (H3): This test combines type 1 and type 2 hypothesis testing to cover the case of a simultaneous fault and a hidden failure.
The hypothesis testing is activated when the DSEBCPS detected abnormality. It is used to detect hidden failures or power faults. The principle is capitalizing on the fact that at substation level we have a higher degree of redundancy in the measurements which means more measurements than the substation states. In real substation this redundancy might reach 2000%. If the confidence level is low ,this redundancy results in higher value in measurement residual for the measurements that suffer from abnormality. Therefore, the measurements with higher values of normalized residual (We are using threshold of 2) classified as suspicious measurements. Then theses measurements are eliminated one at time and re-perform the DSE. If the elimination process reveals high confidence level, abnormality is detected in the eliminated measurements.
We have defined a common mode criteria to distinguish between hidden failures and power fault.
The three phase common mode criteria verifies the adjacent phases of the measurements with the highest normalized residua. If any of them exceeds the threshold of 2, it will be included in the elimination process. If this elimination process reveals high confidence level, hidden failure is detected in the instrumentation channel associated with the eliminated measurements.
The device common mode criteria capitalizes on the fact that most of measurements are modeled using the device model. Therefore, if the measurement with highest normalized residual is modeled by one device, the device common mode criterion verifies the normalized residual for all measurements modeled by the same device. If all of them exceed the threshold of 2, they will be included in the elimination process. If this elimination process reveals high confidence level, power fault is detected in the device.
If the confidence level is still low, the measurement with the second highest normalized residual that was not part of the previous elimination process is selected as suspicious measurement. The process is repeated until we get high confidence level.

29. Numerical Example Case Study: 5 Protection Zones:
115 kV Transmission Line
115 kV Bus
115/13.8 kV , 36 MVA Transformer
13.8 kV Bus
13.8 kV Distribution Line (one of the two)
We have simulated the response of the DSE using a small substation consisting of five protection zones.
We have simulated two cases.

30. Numerical Example Case1: Primary Fuse Blown Y-Y, PT-4A
Sequence of Events
Time (seconds)
Hidden Failure 1 2 3 4 5
Fuse
Blown
6 MW
Load
Switched On
6MW
Load
Switched
Off
The first case is a hidden failure, which is a PT fuse blown. PT4, Phase A fuse was blown. This fuse supplies the settingless relay of transformer zone with voltage measurement of the secondary side of the transformer.
The sequence of events started at 2 s when the fuse blown event is initiated . Also we included 6 MW load switching at 3s and 4s to test the stability of our algorithm during load switching.

31. Numerical Example
Case1: Primary Fuse Blown Y-Y, PT-4A Setting-less Relay of Transformer Zone:
This slide depicts the response of the setting less relay

32. Case1: Primary Fuse Blown Y-Y, PT-4A Centralized Protection Scheme :
Numerical Example
Case1: Primary Fuse Blown Y-Y, PT-4A Centralized Protection Scheme :
This slide depicts the response of the DSEBCPS which shows the DSEBCPS activated the hypothesis testing. The figure shows that the measurements extracted from PT-4A experienced the highest normalized residual and it is considered for elimination. Verifying the common mode criteria reveals that it is the only suspicious measurements to be considered. Therefore, it is selected for elimination process

33. Case1: Primary Fuse Blown Y-Y, PT-4A Centralized Protection Scheme :
Numerical Example
Case1: Primary Fuse Blown Y-Y, PT-4A Centralized Protection Scheme :
This slide shows that the elimination process reveals high confidence level, which means hidden failure is detected within PT-4A that corresponds to the eliminated measurements.

34. Numerical Example Case1: Primary Fuse Blown Y-Y, PT-4A
Compromised Data Correction :
This slide shows the activation of the Data Correction Module. In this module the DSEBCPS streams the estimated phasor quantities of the compromised measurements. The setting less relay uses this information to compute the sampled value to replace its compromised measurements. To facilitate this process we introduced a delay of two cycle as following
1.25 cycle to detect abnormality considering that we preform the DSE at every 1 cycle
0.5 cycle for hypothesis analysis
0.25 cycle to send the compromised data and calculate the sampled value
The slide shows that the confidence level of the setting-less relay responded to the correction and recovered. Also there was no trip signal initiated because of the two cycles delay.

35. A Word on Architecture

36. Conclusions / Recommendations
Three basic innovations for Centralized Substation Protection have been presented: Innovation 1: Setting-less Protective Relay Innovation 2: Merging Units can provide corrected primary values via instrumentation error correction embedded in intelligent Merging Units Innovation 3: Supervision of protection functions (logical nodes) to determine that hidden failures do not exist. In case of hidden failures use existing redundancy to replace compromised data with estimated data and enable continuous reliable operation of logical nodes (protection functions)

37. “Find out why! To accept something the way it has always been done is simply not acceptable. Furthermore, it is boring.”   Walter A. Elmore, Τέλος