AUFLS III - Appendix B - Transpower New Zealand · System Operator Report: AUFLS RoCoF Relay Performance Requirements Page 1 of 46 AUFLS III - Appendix B Rate-of-Change-of-Frequency

System Operator Report: AUFLS RoCoF Relay Performance Requirements Page 1 of 46

AUFLS III - Appendix B

Rate-of-Change-of-Frequency (RoCoF) Relays Bench Testing Methodology and Results

Ferranti Consulting Ltd on behalf of the System Operator June 2012


Ferranti Consulting Ltd on behalf of Transpower Ltd

TABLE OF CONTENTS 1 Summary 4

2 Background 6

3 Purpose of Testing 6

4 Testing Process 7

4.1 Testing Arrangement 7

4.2 Major Components 7

4.3 Test Process 8

4.4 Test Phases 8

4.5 Settings 9

4.6 Relays under test 10

5 Findings 11

5.1 Frequency Response 11

5.2 Response Time 14

5.3 Trip logic 16

5.4 Noise rejection 16

5.5 Stability 16

6 Recommendations 17

6.1 Frequency Response 17

6.2 Response Time 18

6.3 Logic 18

6.4 Stability 19

6.5 Uniformity 19

6.6 Recommendation Summary 20

7 Test Case Description and Results 21

7.1 Test case 1 – Trip point determination – 13 December 2011 event 22

7.2 Test case 2 – 13 December 2011 event with actual relay settings 24

7.3 Test Case 3 – Trip logic test 26

7.4 Test Case 4 – Glenbrook Steel Mill 28

7.5 Test Case 5 – Actual HVDC Commutation Failure 29

7.6 Test Case 6 – Accuracy at a steady frequency decline 31

7.7 Test Case 7 – Stability at abnormal frequencies 32

7.8 Test Case 8 – No Volts (auto-recloser) 33

7.9 Test Case 9 – Loss of a phase (auto-recloser) 35

7.10 Test Case 10 – Arapuni islanding and re-connection event 36

7.11 Test Case 11 – Power system oscillations 38

7.12 Test Case 12 – df/dt calculation delay (varying settings) 40

7.13 Test Case 13 – df/dt calculation delay (varying decay rates) 41



7.14 Test Case 14 – df/dt frequency response 42



Version Tracking

Reviewer Version Changes Date

R.Derks Ver 1 Revision 12 X Jun 2012

J.Blass Ver 2 Appendix format change – comments 25 Jun 2012

J.Blass Ver 3 Incorporated R.Derks Comments 26 Jun 2012

J. Blass Ver 4 Incorporated V.Lo Comments 27 Jun 2012



1 Summary

Findings

The purpose of the testing work described in this document is to provide greater visibility into the strengths and weakness of Rate of Change of Frequency (RoCOF) relays in order to determine the feasibility of and requirements for utilising RoCoF elements to trigger AUFLS load on the New Zealand power system. Seven numerical relays fitted with RoCoF elements were tested to help determine the suitability for this purpose. The tests aimed to analyse the performance of each relay in terms of uniformity, response time, stability, logic elements, and accuracy.

The tests demonstrated that RoCoF relays do not behave uniformly when subjected to identical inputs. Each relay uses slightly different filtering and algorithms to calculate the RoCoF and results in a range of results and response times. This lack of uniformity highlights a risk when considering allowing a range of different RoCoF relays to be implemented on the system. This is because there will be a range of different responses during frequency excursions, which may result in the tripping of AUFLS load at different times even though relays may be set identically. The findings focus the need to ensure the specific requirements of the df/dt calculation are included in the Code which the relays may be tested against to ensure the required level of uniformity is obtained.

The very nature of the algorithms used in these relays means that it will always take a finite amount of time to detect a change in frequency because it involves calculating over a set number of cycles. The response time of each relay is different and the tests show a range in response times of between 130ms and 460ms. This represents the time taken for the relay to detect that the frequency has changed and is independent of any discrete time delays that the user may choose to apply (i.e. it cannot be adjusted on any of the relays that were tested). This is not necessarily a problem provided that relays comply within a maximum allowed response time.

The sampled RoCoF relays were stable under all of the disturbances (including DC commutation failures) that they were subjected to with the settings available at the time of testing. It is noted that the ‘guard’ frequency used in these tests played a major role in preventing the undesired operation of RoCoF triggered AUFLS load.

Each manufacturer has a different way of arranging and setting their logic elements and care is required to ensure the desired result is achieved when programming and setting the various relays.

The tests used to quantify the relay’s response when detecting the underlying rates of frequency change in noisy conditions demonstrated that the sampled relays are susceptible to calculation inaccuracy. These tests highlighted a short-coming in the RoCoF algorithms and presented an associated risk with using in RoCoF relays. The likelihood of system oscillations could not be quantified. Mitigation steps are recommended.

In light of the risk associated it is not recommended to install RoCoF relays prior to Pole 3 commission until wider industry consultation on the risks and recommendations have been completed.



Recommendations

To ensure the reliable use of the RoCoF relays the following items are recommended to be specified in the Code in addition to the previously stated performance requirements:

A. Industry reviews the included tests and results and identifies if there is sufficient confidence in the stability of the relays.

B. Specify the required df/dt calculation filter response to mitigate the inaccuracy caused by the identified level of oscillations to be designed against during under-frequency events in the Code.1

C. Specify the maximum response time allowed for RoCoF relay operation in the Code.

D. Specify a testing outcome to verify the logic elements of the installed relays meet the performance requirements in the Code.

E. The recommended calculation filter response are discussed with current relay manufactures to ensure capable relays to meet Code requirements

1 See Appendix C for a draft example of the relay filtering response



2 Background

Transpower has recently undertaken an investigation into the efficacy and value of its AUFLS system and has concluded that the use of rate-of-change-of frequency, also known as df/dt, triggering provides the most technical and economic improvement. At present all AUFLS is triggered (or tripped) on a combination of frequency and time set points. The System Operator is interested in verifying whether a rate of change of frequency (df/dt) trigger will be suitable on the New Zealand power system to improve the response of AUFLS to system disturbances.

In order to determine whether df/dt is suitable for use as an AULFS trigger, Transpower has undertaken a review of the performance required from df/dt relays and then conducted a review of relay hardware available on the market that meets such functionality. The review (refer to appendix A) concluded that there are a number of relays presently available with df/dt functionality.

The next stage of the investigation is to obtain a sample of these relays and run them through a series of bench tests to observe their response to a combination of actual recorded system disturbances and simulated events. The purpose of the bench testing is to observe the response of the relays to various input waveforms and to ascertain whether the responses align with the needs of the AUFLS system.

3 Purpose of Testing

There are a number of ways that numerical relays may calculate frequency, and each method may give a slightly different result for the same voltage waveform. The differences are small when the voltage waveform is ‘clean’ but when disturbances occur, the voltage waveform may become distorted and the different frequency calculating algorithms may give different results.

Likewise, mathematical constants used in the calculation of rate of change of frequency may vary between relays giving different rates of change for the same voltage input waveform. The result of all of this variability is that different relays with the same df/dt settings may respond differently to the same voltage input waveforms. This in itself is not necessarily a problem, provided the response of the relays is known and predictable. It is however important that if more than one type of relay is used in the power system to trigger AUFLS on df/dt that they all behave within a desired range and the necessary requirements can be specified.

One purpose of bench testing is therefore to observe how a range of df/dt relays respond to the same voltage input waveforms with the objective of ensuring that any relays used on the system behave in a similar manner and that there are no ‘outliers’ or relays that responded significantly differently from the others.

In addition to looking for ‘outliers’, the tests will also determine:

How the relays respond to distorted voltage input waveforms

Relay stability under extremes of frequency

Relay stability with loss of voltage



4 Testing Process

4.1 Testing Arrangement

Most of the relays being tested do not provide a readily accessible output of calculated frequency or rate of change of frequency, and therefore the only way to determine these values is to observe the point at which the relay trips. This can be done by developing a range of different test cases that include voltage waveforms selected on the basis that they will produce frequencies and rates of change of frequencies that will allow the trip points of the relays to be observed.

Relay bench testing was configured as shown below:

Figure 1 - Equipment arrangement used for bench testing of RoCoF elements

4.2 Major Components

The major components of the bench test arrangement are:

4.2.1 Omicron secondary injection test set

These test sets are commonly used to test relays on the Transpower network. For the purposes of the relay testing process, they will be used to convert data files into actual voltage waveforms. The test set will also be configured to record the operation of the relay trip output and output the recorded file in COMTRADE format for each test.



4.2.2 Relay under test

Each ‘relay under test’ will be injected with a voltage waveform from the Omicron. All relay test voltages will be configured with a steady 50Hz voltage of at least 10 seconds duration prior to the test case voltage being applied to the relay. This is necessary to ‘initialise’ the relay. The relay will be programmed (using proprietary software supplied by the relay manufacturer) to operate an output contact when the desired test set points are reached. This output contact is the only measured output from the relay under test as internal measurements of frequency and df/dt are not usually available.

4.2.3 Matlab

A key component of the test arrangement is the use of Matlab to perform an independent calculation of frequency and df/dt from the various input voltage waveforms. This is necessary because frequency and df/dt is not commonly available from the test relays but it is required in order to provide a reference against which the operation of the relays can be compared.

The input voltage waveforms, calculated frequency, calculate df/dt and the relay trip points will be combined in Matlab on the same time axis for the purposes of observing when the relay tripped (i.e. at what calculated rate of change).

4.3 Test Process

A typical test will involve:

Selecting a voltage waveform to inject into the relay under test and ‘running it’ through Matlab to calculate frequency and df/dt (refer to following sections). This will be done before the day of testing.

Loading the selected voltage waveform into the Omicron

Programming the relay under test with the desired settings

Playing the selected voltage waveform into the relay under test

Recording when the relay under test operates

Comparing the operating point with the voltage, frequency and df/dt all on the same time axis in Matlab. This will be done post-test.

4.4 Test Phases

The tests described in this document were carried out in three ‘phases’ or ‘rounds’ as follows:

Phase I: April 10th to12th, 2012

Phase II: April 30th to May 2nd, 2012

Phase III: June 11th to 13th, 2012

The tests were phased to allow for analysis of the results which permitted subsequent tests to be refined in order to further explore findings of the previous round. The testing was an iterative process in which the relevance of the test completed changed as the knowledge of the relays behaviour increased. Phasing was also necessary to accommodate the availability of resources and personnel.

All of the tests were carried out at Transpower House, 96 The Terrace, Wellington.



4.5 Settings

There are two basic types of tests each with different types of settings:

Determining the df/dt that each relay ‘sees’ for a particular waveform. This type of test is necessary because none of the relays provide an output of the rate of change frequency that they are calculating and therefore the only way to find out what a relays calculated df/dt is, is to move the relays settings up and down until the trip level is found.

Determining if a relay would trip using the ‘actual’ settings. Two sets of ‘actual’ settings were used, namely those proposed for Pole 3 commissioning and those proposed as part of the long-term implementation of df/dt. The long-term settings differed due to dependence on changes to the interruptible load (IL) products or the Reserve Management Tool changes. These tests aimed to check that the relays will not operate for non-AUFLS type system disturbances when the recommended settings provided below are programmed into the relay.

4.5.1 Proposed Pole 3 commissioning settings2:

Block 1:

df/dt = -0.4Hz/s

Guard frequency = 48.0Hz

Time delay = 0.2s

Block 2:

df/dt = -0.8Hz/s


Time delay = 0.2s

4.5.2 Proposed long-term (‘normal’) settings:

Block 1:

df/dt = -0.8Hz/s


Time delay = 0.2s

Block 2:

df/dt = -1.5Hz/s


Time delay = 0.2s

2 Note the under-frequency back-up elements are not specified for either set of settings.



4.6 Relays under test

The following relays were tested3:

Manufac-

turer

Relay

model

Description Target markets*

SEL 751 Multifunctional feeder protection Transmission / Distribution

SEL 351-7 PS Multifunctional feeder protection Transmission / Distribution

GE SR760 Multifunctional feeder protection Transmission / Distribution

Siemens 7SJ64 Multifunctional feeder protection Transmission

Siemens 7SJ80 Feeder protection Distribution

ABB RED670 Line differential protection Transmission

ABB REU615 Voltage protection Distribution

Figure 2- Relays included in the testing programme

* Transmission relays are usually provided with more functionality and are usually more expensive than distribution products

These are all multifunctional numerical relays that they can be programmed to provide a range of different protection functions in addition to frequency protection.

Two additional relays were also provided for testing, namely a RMS 2H34-S frequency relay and an Alstom P145 feeder protection relay. The RMS relay was found to have a bug that prevented settings to be changed reliably, and the Alstom relay arrived half way through the second round of testing. Neither of these relays was tested as a result however they will be retained on loan for the manufacturers should it be decided to include them in a further round of tests.

Further details on why these relays were chosen for testing, as well as the basis for their selection is provided in appendix A.

3 The individual relay results are not disclosed in the report and will remain until agreed with by relay

manufacturers.



5 Findings

Rate of change of frequency is derived (by differentiation) from frequency, which is in turn derived from voltage waveforms. The differentiation operation is sensitive to higher frequency noise, and therefore heavy low-pass filtering must be applied for it to work reliably. The frequency derivation itself is not a simple calculation, and various algorithms with different characteristics are used. The result is that the calculation of df/dt varies widely between relays, and each of these relays can behave very differently, especially for the unusual waveforms encountered during large grid events.

The test results illustrate the differences between the ways that each relay calculates df/dt. The fact that relays were observed to trip at a different rate of change and at different times for the same input waveforms shows that each relay uses different algorithms and filters. This even appears to apply to relays from the same manufacturer. The variation in results prompted a closer look into the possible reason for the differences including consideration of how df/dt algorithms are implemented in the relays under test.

Tests 11, 14 and 15 demonstrate the filtering that is used to assist with removing the effects of high frequencies. In addition to illustrating the frequency response of the tested relays, the results of these tests also indicate that each relay uses different filtering at different set points. Tests 1 and 2 use the system frequency recorded during an actual event to demonstrate the undesirable effect that the frequency response characteristics of the relays can have under certain conditions.

Tests 12 and 13 show that the response times of the relays varies depending on the magnitude of the difference between the frequency decline and the relay setpoint.

The flexibility of the way that the internal elements of the relays can be arranged is exercised in test 3, while relay stability and their ability to reject noise is looked at in tests 4 to 10 and test 16.

5.1 Frequency Response

As testing of the relays progressed through phases I and II, it became clear that ‘low pass’ filters were being used within the df/dt algorithms to filter out ‘high’ frequencies depending on the relays df/dt set point.

In order to determine the response of these filters on two of the relays under test (the Relay F and the Relay A4), waveforms with sinusoidal frequency-oscillations at different sinusoidal-frequencies were input into the relay. The oscillation frequencies were chosen irregularly to minimize sampling with periodic frequency responses such as those from a rectangular averaging filters. For each sinusoidal frequency, the threshold magnitude of the sinusoid required for relay operation was found. The peak

df/dt for an oscillation of frequency and magnitude (2M pk-pk) is given by:

4 This set of tests was carried out on Relay F and Relay A relays due to time constraints and the need to

understand the response of these relays as part of the Pole 3 Ready project.



Using this equation, we can compute the attenuation in the sinusoid, and hence the frequency response of the filters. The results are best summarised on the graph below:

Figure 3 – Attenuation of oscillations by RoCoF relay filters

The graph shows that at low frequencies, oscillations will be ‘passed through’ the df/dt filters and will therefore affect the df/dt as measured by the relay.

Using the plot of the Relay F with a set point of 0.4Hz/s as an example, it can be seen that if an oscillation of 1.6Hz were to occur, it would be attenuated to 67% of its original size when passing through the df/dt low pass filter and then measured as an actual rate of change of frequency.

Should there be a steady decline in the underlying 50Hz frequency, and if an oscillation of 1.6 Hz were occurring at the same time, then the relay would ‘see’ a df/dt greater than the underlying frequency decay, i.e. the oscillation will cause an error in the measured rate of change of frequency. However, if the oscillation frequency were greater than about 4Hz, the relay will filter it out and it will not impact significantly on the calculated df/dt value. Refer to test 15 for a demonstration of this phenomenon. The amount of pass though for unfiltered frequencies changes depending on then relay type (i.e. the df/dt algorithm) and the df/dt set point that is selected, as shown on the graph above.

For example, assume that there is a phase oscillation of 2 degrees magnitude (4 degrees peak to peak) at 1.6Hz superimposed on a 0.1Hz/s steady frequency decline and measured by Relay F set at 0.4Hz/s. Then using the equation:

with , we find .

Multiplying this by the gain of the 351 at 1.6Hz of 0.67 gives 0.38Hz/s. So the relay sees the 0.1Hz/s frequency decline as a 0.1+0.38=0.48Hz/s decline and therefore trips.

It must also be noted that the test used to derive the frequency response curves shown above was carried out using sinusoidal waveforms. The response of the df/dt algorithms to non-sinusoidal oscillations will be different from those shown in the

F F A A



graphs. In general for “noisier” signals the peak df/dt (and therefore errors) seen will be larger.

The frequency response plots produced by this test can be used to approximate the response of the relay. The frequency response is resampled and then inverse Fourier transformed to get the time domain response, which is then time-shifted and truncated to ensure causality and then convolved with any given input signal to get the simulated output signal. This method was used to create the red “relay df/dt” plot shown above.

Test 15 was carried out to demonstrate the impact of the low pass filter on the behaviour of the relay. It shows that low frequencies are ‘let through’ and therefore have an impact on the trip point of the relay. For the relay and settings used in this test, it can be seen that errors caused by low frequency oscillations will in some cases cause the relay to trip even though the underlying frequency decline is lower than the set-point (i.e. it will trip incorrectly). The magnitude of the error and hence the degree of incorrect tripping is dependent on the frequency and amplitude of the oscillation, the df/dt set point, and the relay’s filtering.

It is difficult to quantify the likely amount of power oscillation expected, especially during an AUFLS event. The impact of oscillations may be scaled down by the relay’s low-pass filter however the oscillations can still have an impact on the df/dt calculation and at worst cause the relay to trip incorrectly.

In addition to the potential problems caused by simple power oscillations, another failure mode is seen in Test 2, where the filter removes the initial ~6Hz oscillations, but when the oscillation frequency falls to ~2Hz it is let through, causing the relay to operate even though the underlying frequency is recovering. This is shown in the plot below.

Figure 4 - Frequency and trip plot for test 2 on the Relay G



5.2 Response Time

SEL literature for the df/dt element includes a table listing different ‘time windows’ for different settings:

Figure 5 – Excerpt from SEL 751 instruction manual showing df/dt calculation ‘time window’ vs relay df/dt setting (Table 4.39 Time Window Versus 81RnTP Setting)

It is suspected that the ‘time windows’ quoted for the SEL relays above will affect the response time of the relay’s df/dt element. Tests 12 and 13 were devised to observe how the different time windows may affect the response time of the relays. The results of test 12 can be illustrated by considering the behaviour of Relay F at two different settings as illustrated in the diagrams below:

Figure 6a – Response times of Relay F at 0.1 Hz/s setting



Figure 6b – Response times of Relay F at 0.95 Hz/s setting

For the first case (setting = 0.1 Hz/s) the relay was recorded to trip 230ms after the rate of frequency changed. As the 1000ms ‘window’ starts to ‘cover’ the frequency elbow, the relay begins to register that the frequency is changing. After 130ms the algorithm has detected that the df/dt is higher than the set point and an inbuilt 100ms timer is started. After the timer has expired, and the frequency drop-off is still higher than the set point, the relay trips. Because the difference between the set point and the actual drop-off is relatively large, the time taken for the relay to calculate that df/dt is greater than the set point is relatively short, even though the time window is long at 1000ms (i.e. the relay will not wait for the duration of the window to operate the df/dt element). It is noted that the width of the calculation period cannot be adjusted by the user for any of the relays that were tested.5

The same thing is occurring in the second case, but now the set point is much closer to the actual rate of change of frequency. In this case, the relay requires the duration of the entire time window to register that the frequency is falling faster than the set point. The inbuilt delay time of 100ms then results in the relay taking 300ms to trip.

The observation that can be made is that the relay will respond quicker to frequency drop-offs that are significantly greater than the set point compared to drop-offs that are close to the set point.

Test 13 looks at the corollary of test 12, i.e. how calculation times may vary as the rate of frequency decline changes for a fixed df/dt set point. The test demonstrated the same response time characteristics as seen for test 12, namely that the relays take longer to detect rates of change close to the set point than they do for rates of change markedly different from the set point.

It is noted that the above illustrations use the results of Relay F however all of the relays subjected to these tests gave similar results, albeit with different tripping times.

In addition to the effect that the ‘time windows’ have on the overall response time of the relays, it can be seen that for lower df/dt set points, the calculation periods are longer, providing more accuracy and limiting the error relative to the setpoint. For higher df/dt setpoints, the calculation periods are shorter, providing greater speed since at high df/dt rates there is less time for AUFLS to react before the system reaches a dangerously low frequency. Therefore for this test as the df/dt setpoint increases, the calculation window decreases, which increases the post-filtered df/dt and hence the df/dt setpoint required for it to not trip.

5 The Alstom P145 relay is provided with an adjustable averaging window, but it arrived too late to be

included in the tests. See discussion below.



Since the system frequency takes at least 1 second (likely much more) to reach the guard frequency, the response times provided by the relays are far faster than necessary. A robust system-wide AUFLS implementation would tradeoff a slower response time for better accuracy and stability.

5.3 Trip logic

With the recommended df/dt AUFLS settings, there are pre-defined conditions that must be met before a trip occurs. For block one, these are that the frequency must be less than 48.5Hz, it must be falling at a rate greater than 0.8 Hz/s, both of these conditions must persist for 0.2 seconds or more.

Test 3 simulates the combination of one, two and all three of these conditions to observe if the relay logic can be set to trip only when all three conditions are met. The relays should only trip for the last excursion below 48.5Hz and this is the only time where df/dt < 0.8Hz/s and f <48.5Hz for more than 0.2 seconds. If the relays trip at any other time, then further investigation will be required to determine the reason.

Four of the six relay subjected to this test performed as expected and only tripped for the last of the three excursions below 48.5Hz. Two of the relays tripped for all of the excursions. Although this result was unexpected, it is assumed that fine tuning the logic elements and delay settings will be able to prevent this. It was not investigated further as was not required for the remaining tests.

The variability of trip time delays indicates that careful attention needs to be paid to the way that the tripping logic is implemented in the relay. It also reflects the different calculation times that different relays use for their df/dt algorithms.

5.4 Noise rejection

Voltage waveforms are often distorted by loads. The arc-melter at the Glenbrook Steel plant produces voltage waveform distortions whenever it operates. Even though these are among the most distorted voltage waveforms in New Zealand, the melter does not have a significant effect on frequency and the df/dt function of the relay should not be triggered. Likewise HVDC commutation failures can cause very short lived but severe distortions of the voltage waveform. The distortions can include multiple sub-cycle zero crossings which could have an effect of the frequency calculation algorithms of some relays. HVDC commutation faults will affect system frequency but generally not to the degree that AUFLS operates.

Even though the frequency excursions are significant and the rate of change of frequency is high for some of the ‘noisy’ waveforms used, the durations of the events are extremely short and the frequencies of the noise are high and therefore the df/dt algorithms will attenuate them and none of the relays trip, even when the ‘sensitive’ settings are used.

Note that multiple restarts of the HVDC after it trips was not tested because voltage waveforms were not available at the time of testing. This is a possible failure mode and should be tested at the next available opportunity.

5.5 Stability

Some of these relays may be installed at locations where they are exposed to frequency excursions outside of the range normally expected. Auto-reclosers and faults including the loss of phases will switch voltage inputs in rapid succession.

The relays did not trip for the tests that simulated these conditions which indicates that the frequency algorithms remained stable.



6 Recommendations

This section will highlight the observations made from the test results and put forward recommendations to ensure the observed results deliver the desired outcomes from any changes to the AULFS scheme.

6.1 Frequency Response

The ability to rely on RoCoF relays for load shedding requires the relay to be able to accurately calculate the system RoCoF when subjected to variety of grid conditions. The consequences of inaccuracy could result in the relay mal-operating by tripping inappropriately or not tripping when needed.

The relays performed within the specified accuracy margins (±0.05 Hz/s) for clean frequency decays. However, the testing results did verify that RoCoF relays are susceptible to calculation inaccuracy due to system oscillations.

The tests used to quantify the relay’s accuracy of detecting the underlying rates of frequency change in noisy conditions (test 1, 2, 14 & 15) demonstrated the calculation inaccuracy due to system oscillations.

Test 14 and 15 highlighted the importance of filtering on the relays ability to accurately calculate the system rate of change of frequency during a system oscillation. The test demonstrated with sufficient filtering and calculation time the df/dt relays may suitably minimize the impact of the oscillations and maintain the required levels of accuracy.

It is therefore recommended that the following further action is taken:

Specify the required df/dt calculation filter response to mitigate the inaccuracy caused by the identified level of oscillations to be designed against during under-frequency events in the Code.

It is noted that the complexity of the relays frequency response was demonstrated through the 13 December 2011 event. The varied performance of the relays during the event depended on the settings and the associated filtering used.

Relay Availability

A review of the technical manual for the Alstom P145 relay shows that it includes a setting that allows the user to adjust the length of the time window used by the df/dt algorithm. This is the only df/dt relay that was found to have such a feature as all of the others had non-adjustable time windows. The benefit of being able to control the length of the window is that it could be fixed at a relatively long time (say 800ms or more) which would be likely to give better immunity to high frequency noise and possibly improve the response of the relay to oscillations such as those found in test 1.

A downside of long averaging windows is the slow response to changes in system frequency. This should be seen in the light of the intended application of df/dt relays for AUFLS in that the time taken for the system frequency to fall to below the guard frequency should allow enough time for a relay with a long averaging window to have calculated that the underlying frequency is declining, so that when the guard frequency is reached, the relay can operate as intended.

Using the test 1 waveform as an example, it can be seen that the frequency takes about 2 seconds to fall from 50Hz to 48.5Hz. The relay would have detected that the frequency is falling at about 0.5Hz after 800ms and when the frequency drops below



48.5Hz (the guard frequency) the logic elements would initiate the pre-set time delay and after it had expired would trip. In other words, the relay will not wait until after the frequency drops below the guard level to start calculating df/dt, it does it continuously.

It is therefore recommended that the following further action is taken:

The recommended calculation filter response and calculation ‘time window’ are discussed with current relay manufactures to ensure capable relays to meet Code requirements

6.2 Response Time

The very nature of the algorithms used in these relays means that it will always take a finite amount of time to detect a change in frequency because it involves measuring frequency at two different points in time and calculating the difference between them. The tests demonstrate that the closer the rate of change is to the set point, the longer the response time and conversely, the greater the difference, the faster the relay operates (see tests 12 and 13).

As each relay has a unique df/dt algorithm, the response time of each relay is different and the tests show a range in response times of between 130ms and 460ms. This represents the time taken for the relay to detect that the frequency has changed and is independent of any discrete time delays that the user may choose to apply (i.e. it cannot be adjusted). Therefore the following action is recommended:

Specify the maximum response time allowed for RoCoF relay operation in the Code.

6.3 Logic

All of the relays tested had some form of logic programming that allowed users to implement various combinations of df/dt, frequency and time delay elements to achieve a desired tripping response.

Each manufacturer has a different way of arranging and setting their logic elements and care is required to ensure the desired result is achieved when programming and setting the various relays. This was illustrated in test 3 where Relay G and Relay A relays exhibited an unexpected response to a simple logic test. It is suspected that this could be addressed by adjusting the logic arrangements used in these relays. Therefore the following action is recommended:

Specify a testing outcome to verify the logic elements of the installed relays meet the performance requirements in the Code.



6.4 Stability

The tests indicate that the df/dt relays tested were stable under all of the disturbances that they were subjected to. This included:

HVDC commutation failures and multiple restarts

Distorted voltages (Glenbrook steel melter)

Loss of input voltages

Wide frequency excursions

This indicates that the df/dt algorithms are designed to reject noise and fast transient events which will help in avoiding undesired operation of df/dt triggered AUFLS load. It is noted that the guard frequency used in these tests will also play a major role in preventing the undesired operation of df/dt triggered AUFLS load. In light of the findings the following action is recommended:

Industry reviews the included tests and results and identifies if there is sufficient confidence in the stability of the relays.

6.5 Uniformity

A level of uniformity is required to ensure the installed AUFLS system achieves the desired levels of operation should a range of different relays be used. The tests demonstrated that the sampled df/dt relays do not behave uniformly when subjected to identical inputs. Tests 1 and 2 give a vivid example of this where a waveform for an actual event was used to find the rate of change of frequency that each relay calculates. No two relays calculated the same rate of change, with the range of calculated values being from 0.5 to 6.6 Hz/s.

This effect was further explored in tests 14 and 15 which illustrated the effects of filters used by the df/dt algorithms. This lack of uniformity was seen across all of the tests, and it indicates that different relay manufacturers use different algorithms for calculating df/dt.

This lack of uniformity highlights a weakness when considering allowing a range of different df/dt relays to be implemented on the system. This is because there will be a range of different responses during frequency excursions, which may result in the tripping of AUFLS load at different times even though all relays are set identically.

It is therefore recommended that Code recommendations include specific requirements of the df/dt calculation which the relays may be tested against to ensure the required level of uniformity is obtained.

Further testing of any df/dt relays selected for use on the AUFLS system would be prudent to determine their response to a wider range of events than provided in this round of tests. This can be done in parallel to the analysis work recommended above.



6.6 Recommendation Summary

In light of the findings from the testing work the following actions are recommended (summarised from above):

A. Industry reviews the included tests and results and identifies if there is sufficient confidence in the stability of the relays.

B. Specify the required df/dt calculation filter response to mitigate the inaccuracy caused by the identified level of oscillations to be designed against during under-frequency events in the Code.6

C. Specify the maximum response time allowed for RoCoF relay operation in the Code.

D. Specify a testing outcome to verify the logic elements of the installed relays meet the performance requirements in the Code.

E. The recommended calculation filter responses are discussed with current relay manufactures to ensure capable relays to meet Code requirements.

6 See Appendix C for a draft example of the relay filtering response


7 Test Case Description and Results

Each test case is described in terms of:

Objective Why this test is being carried out

Test Waveform A description of the input waveform being applied to the relays for this test case

Relay settings The settings that will be programmed into each relay for the test case

Results Empirical results from the testing process



7.1 Test case 1 – Trip point determination – 13 December 2011 event

Objective

To observe whether the df/dt algorithms operate at the same time for identical input waveforms and with identical settings given a complex frequency input.

Test Waveform

50Hz steady voltage for 10 seconds

13 December 2011 event (approximately 20 second duration)

The ‘event’ voltage waveform will be back-calculated from the frequency trace captured by an ION meter that was recording frequency at Henderson on 13 December 2011.

The underlying rate of change of frequency can be visually seen to be around 0.5Hz/s. The noise that is superimposed on the frequency decline may however cause the relays to detect a different df/dt.

Figure 1- Frequency trace at Henderson for 13 December 2011 event

Relay Settings

df/dt time delay = 0 seconds

df/dt averaging period (if available) = 0 seconds

df/dt = 10 Hz/s initially, then decrease by 1.0 Hz/s until relay trips then increase by 0.1 Hz/s until relay does not trip. Continue in this fashion until the operating point of the relay is found.



Results

The results for this test are summarised in the table and graph below.

Table 1 – Summarised results for test 1

Relay Trip Trip point

(Hz/s) Trip point

(ms)* Test date

A Y 2.45 310 12-Apr

B Y 0.5 116 2-May

C Y 6.6 120 1-May

D Y 2.95 290 2-May

E Y 2.5 585 1-May

F Y 2.4 145 30-Apr

G Y 2.34 200 11-Apr

* This is the elapsed time between the time that the frequency begins to change and the time that the relay issues a trip signal from its df/dt element.

Figure 2 – Test 1 relay trip points

A F G D D E C



7.2 Test case 2 – 13 December 2011 event with actual relay settings

Objective

To observe if any relays would have operated on 13 December 2011 with the recommended (‘actual’) df/dt settings.

Test Waveform

The input waveform from Case 1 will be used along with a trace of the frequency as seen at Upper Hutt GXP will also be used:

50Hz steady voltage for 10 seconds

13 December 2011 event (approximately 20 second duration) – at Henderson GXP

Figure 3 – Frequency trace at Henderson for 13 December 2011 event



Relay settings

The recommended relay settings for ‘normal’ operation will be used:

Block 1:

df/dt = -0.8Hz/s


Time delay = 0.2s

Block 2:

df/dt = -1.5Hz/s


Time delay = 0.2s

Results

The results for this test are summarised in the table below.

Table 2 – Summarised Results for test 2


(Hz/s) Trip point

(ms) Test date

A N N.A. No trip 12-Apr

B N N.A. No trip 2-May

C Not tested*

D N N.A. No trip 2-May

E Y N.A. 1550 1-May

F Y N.A. 6600 30-Apr

G Y N.A. 7000 11-Apr

* Relay C was not put through this test due to the variance observed in its tripping point compared to the other relays in test 1. Due to this it was considered unlikely that this relay would be compatible with an AUFLS scheme that was triggered on df/dt however as the filtering was not understood until later tests there is potential with minor changes that this relay could be made compliant with future requirements.



7.3 Test Case 3 – Trip logic test

Objective

To observe how the combination of trip elements logic operate, in particular the time delay element.

Test Waveform

A simulated waveform will be used for this test. The frequency trace used will be as per that shown below, and the voltages required to produce this trace will be derived using MatLab.

Figure 4- Test Case 3 frequency trace

Relay settings

The recommended ‘block 1’ relay settings for normal will be used:

df/dt = -0.8Hz/s


Time delay = 0.2s



Results

The results for this test are summarised in the table below. Detailed results including scanned copies of result sheets and graphs of relay responses are provided in the appendices.

Table 3 – Summarised results for test case 3


(Hz/s) Trip point

(ms) Test date

A Y* N.A. 175 12-Apr

B Y N.A. 280 2-May

C Not tested**

D Y N.A. 300 2-May

E Y N.A. 260 1-May

F Y N.A. 230 30-Apr

G Y* N.A. 180 11-Apr

* These relays tripped for every frequency excursion below 48.5Hz. The operating time for the last frequency excursion is shown for these relays.

** Refer to results for ‘Test 1’.



7.4 Test Case 4 – Glenbrook Steel Mill

Objective

The objective of this test is to observe the stability of the relays with distorted voltage waveforms.

Test Waveform

An actual voltage waveform taken from a recorder located at Glenbrook Steel plants ‘dirty’ bus will be used as an input to the relays. The waveform will be preceded by a steady 50Hz voltage of 10 seconds to initialise the relay.

Figure 5 – Typical Glenbrook ‘dirty’ bus voltage and current waveforms

Relay settings

Two sets of settings will be applied for this test:

A. Actual settings (0.8Hz/s, 48.5Hz, 0.2s delay)

B. Sensitive settings (0.2Hz/s, no guard frequency, no time delay)

Results


Table 4 - Summarised results for test case 4


(Hz/s) Trip point

(ms) Test date

A N N.A. N.A. 12-Apr

B N N.A. N.A. 2-May

C N N.A. N.A. 1-May

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May

F N N.A. N.A. 30-Apr

G N N.A. N.A. 11-Apr



7.5 Test Case 5 – Actual HVDC Commutation Failure

Objective

Test the stability of the relays under an actual HVDC commutation failure event.

Test Waveform

The waveform of an actual HVDC commutation failure event recorded at Upper Hutt of 13-09-2010 will be used for this test. This waveform includes voltage distortions and multiple sub-cycle zero crossings.

Figure 6 - Observed voltage and current waveforms for a HVDC commutation failure

Relay settings

Two sets of settings will be applied for this test:

A. Actual settings (0.8Hz/s, 48.5Hz, 0.2s delay)

B. Sensitive settings (0.2Hz/s, no guard frequency, no time delay)



Results




(Hz/s) Trip point

(ms) Test date


B N N.A. N.A. 2-May

C Not tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May



* Refer to results for ‘Test 1’.



7.6 Test Case 6 – Accuracy at a steady frequency decline

Objective

To test that the relay does not trip for frequency drops close to but above the relay set points.

Test Waveform

A simulated waveform will be used that provides a steadily dropping frequency from 50Hz at a rate of 0.79Hz/s for 4 seconds. A steady 50Hz voltage waveform will precede the test waveform for 10 seconds to allow for initialisation of the relay.

Relay settings

Actual relay settings will be used:

Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

Results


Table 6 – Summarised results for tests 6


(Hz/s) Trip point

(ms) Test date

A Y 0.75 N.A. 12-Apr

B Y 0.83 N.A. 2-May

C Y 0.75* N.A. 1-May

D Y 0.81 N.A. 2-May

E Y 0.79 N.A. 1-May

F Y 0.8 N.A. 30-Apr

G Y 0.8 N.A. 11-Apr

* The df/dt element of this relay has a resolution of 0.25 Hz/s which is very coarse compared to the remainder of the relays.



7.7 Test Case 7 – Stability at abnormal frequencies

Objective

To observe the stability of the relays at frequencies well outside of the normal range.

Test Waveform

Simulated waveforms that provide a rising frequency from 50Hz to 60Hz at +0.7 Hz/s followed by a falling frequency from 60Hz to 30Hz at -0.7Hz/s will be used.

A steady 50Hz voltage waveform will precede the test waveform for 10seconds to allow for initialisation of the relay.

Relay settings


Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

Test B (Block 2):

df/dt = -1.5Hz/s


Time delay = 0.2s

Results




(Hz/s) Trip point

(ms) Test date


B N N.A. N.A. 2-May

C Not tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May






7.8 Test Case 8 – No Volts (auto-recloser)

Objective

To observe how the relay responds to loss of supply events such as those that may be caused by an auto-recloser operating.

Test Waveform

A simulated wave form will be used where

f = 50Hz for 10 seconds with a steady 3 phase voltage

Voltage is zero for 1 second

f = 50Hz for 1 second with a steady 3 phase voltage

Voltage is zero for 2 seconds

f = 50Hz for 1 second with a steady 3 phase voltage

Voltage is removed

Relay settings


Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

Test B (Block 2):

df/dt = -1.5Hz/s


Time delay = 0.2s



Results




(Hz/s) Trip point

(ms) Test date


B N N.A. N.A. 2-May

C Not tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May






7.9 Test Case 9 – Loss of a phase (auto-recloser)

Objective

This test is similar to the ‘No volts’ test above but this time one phase it at zero volts, simulating a single phase fault.

Test Waveform

A simulated wave form will be used where

f = 50Hz for 10 seconds with a steady 3 phase voltage

f = 50Hz for 1 second with 2 phases normal and one phase at zero volts

Voltage is zero for 1 second


Voltage is zero for 2 seconds


Voltage is removed

Relay settings


Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

Test B (Block 2):

df/dt = -1.5Hz/s


Time delay = 0.2s

Results




(Hz/s) Trip point

(ms) Test date


B N N.A. N.A. 2-May

C Not tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May






7.10 Test Case 10 – Arapuni islanding and re-connection event

Objective

To observe the response of the relays under the conditions experienced at Arapuni on 15 Jan 2012 when Arapuni and Kinleith were islanded from the rest of the North Island for a short time before being reconnected. When reconnection occurred, there was an angle difference of about 145 degrees and the islanded frequency was about 50.36Hz.

Test Waveform

The actual waveform of the event as recorded at Arapuni will be used, as illustrated below:

Figure 7 – Arapuni bus voltage and modelled phase angel after reconnection of island to main system

Relay settings


Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1x 10

5

Time (sec)

Voltage

-0.5 0 0.5 1 1.5 2 2.5 3-200

-150

-100

-50

0

50

100

150

200

X: 0.9907

Y: -145.1

Digs

ilent

Pha

se

Time



Test B (Block 2):

df/dt = -1.5Hz/s


Time delay = 0.2s

Results




(Hz/s) Trip point

(ms) Test date

A N N.A.** N.A. 12-Jun

B N N.A. N.A. 2-May

C Not tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May

F N N.A.** N.A. 12-Jun



** Relay F and Relay A were subject to these tests with df/dt set points of 0.1Hz/s and with no guard frequency. Neither of the relays tripped with these settings.



7.11 Test Case 11 – Power system oscillations

Objective

To determine the set point at which the df/dt algorithms operate for a stable system frequency with power system oscillations, sometimes referred to as intermodal oscillations, of the type observed on the New Zealand power system.

Test Waveform

The following modelled waveform will be used for these tests. This oscillation will be superimposed on the system frequency.

Figure 8- Modelled power system oscillation

Relay settings

df/dt time delay = no delay (i.e. time delay setting is disabled)

df/dt averaging period (if available) = disabled

df/dt = 5 Hz/s initially, then decrease by 50% increments until relay trips then increase by 50% increments until relay does not trip. Continue in this fashion until the operating point of the relay is found.

10 10.5 11 11.5 12 12.5 13 13.5 14 14.5 15-4

-3

-2

-1

0

1

2

3

4

Time (sec)

Phase (

degre

es)



Results




(Hz/s) Trip point

(ms) Test date


B N N.A. N.A. 2-May

C Not

tested*

D N N.A. N.A. 2-May

E N N.A. N.A. 1-May






7.12 Test Case 12 – df/dt calculation delay (varying settings)

Objective

To observe how the relay responds when different df/dt settings are applied to the same constant frequency decline. This is intended to show if the relay df/dt calculation times vary as the set point is changed.

Test Waveform

A simulated wave form will be used where the frequency is first kept constant at 50Hz and then set to a steady 1Hz/s decrease. This gives a step change in the df/dt and the time it takes for the relay to detect the change with different set points applied will provide an insight into how the algorithms respond to changes in frequency.

Relay settings


df/dt averaging period (if available) = standard

df/dt = 0.1, 0.5, 0.7, 0.8, 0.97Hz/s.

Results

The results for this test are summarised in the table below.


Relay

Set point

Test date

0.1 Hz/s

0.8 Hz/s

0.95 Hz/s

A Trip ** ** **

Trip time

(ms) ** ** **

B

Trip Y Y Y

2-May Trip time (ms)

138 322 352

C

Trip ** ** ** Trip time

(ms) ** ** **

D

Trip Y Y Y 2-May

Trip time (ms)

157 265 394

E Trip Y Y Y


186 230 267

F Trip Y Y Y

30-Apr Trip time (ms)

231 295 325

G Trip ** ** **

Trip time

(ms) ** ** **


** Test 12 has yet to be carried out for these relays



7.13 Test Case 13 – df/dt calculation delay (varying decay rates)

Objective

To observe how the relay responds when different frequency decay rates are applied with the same df/dt settings. This is intended to show if the relay df/dt calculation times vary for different frequency decline rates.

Test Waveform

A series of simulated wave forms will be used where the frequency is first kept constant at 50Hz and then set to steady declines of 0.3 Hz/s, 3.0 Hz/s and 14 Hz/s. These rates were chosen to provide extremes of the relays settings hence highlighting any differences in calculation times.

Relay settings



df/dt = 0.3 Hz/s (remains constant for each test)

Results



Relay Description

Set point

Test date 0.31 Hz/s

0.35 Hz/s

3.0 Hz/s

14.0 Hz/s

A

Trip ** ** ** Trip time

(ms) ** ** **

B

Trip Y Y Y Y


387 280 253 243

C Trip ** ** **

Trip time

(ms) ** ** **

D Trip Y Y Y Y


330 296 254 133

E Trip Y Y Y Y


412 369 338 234

F

Trip Y ** Y Y

30-Apr Trip time (ms)

461 ** 171 141

G

Trip ** ** ** ** Trip time

(ms) ** ** ** **


** Test 13 has yet to be carried out for these relays



7.14 Test Case 14 – df/dt frequency response

Objective

To investigate the frequency response of the low pass filters associated with df/dt calculation.

Test Waveform

Waveforms with sinusoidal frequency-oscillations at different sinusoidal-frequencies are input into the relay.

Relay settings

df/dt set point = 0.4Hz/s and 1.5Hz/s (i.e. two sets of tests carried out)



Results

NOTE: This set of tests was carried out on Relay F and Relay A only due to time constraints and the need to understand the response of these relays as part of the Pole 3 Ready project.

The results are best summarised on the graph below:

Figure 9 – Attenuation of oscillations by RoCoF relay filters

The graph shows that at low frequencies, oscillations will be ‘passed through’ the df/dt filters and will therefore affect the df/dt as measured by the relay.

The frequency response plots produced by this test can be used to approximate the response of the relay.

F F A A




Objective

This test sets out to demonstrate the response of the df/dt elements of the SR760 relay to oscillations.

Test Waveform

Steady 50Hz for 5 seconds then a steady frequency decline of 0.35 Hz/s with a

superimposed ‘noise’ of 1.5Hz at an amplitude of 2⁰ p-p.

Relay settings

The relay will be set to trip at 0.4 Hz/s with no guard frequency.

Results

The results show that the relay tripped even though the underlying frequency decline was less than the relay set point.

Figure 10 – Relay A with underlying frequency decline of .35 Hz/s, an oscillation of 1.5 Hz and a trip point of .4Hz/s

A further test was carried out to observe whether increasing the frequency of the oscillation to above the filters ‘cut-off’ point would impact on the response of the relay. This test used the same set-point and underlying frequency decline as the first test, but the frequency of the oscillation was increased to 2.5Hz, which is above the cut-off point of the low pass filter as shown in test 14. The result of this test is shown below.



Figure 11 – Relay A with underlying frequency decline of .35 Hz/s, an oscillation of 2.5 Hz and a trip point of .4 Hz/s

The relay did not trip in this instance which demonstrates that the low pass filter was removing the 2.5 Hz oscillation.




Objective

Observe relay response to a series of simulated ‘pole 3’ HVDC commutation failure events.

Test Waveform

Four waveforms were provided by Siemens from their simulations of commutation failure of pole 3 HVDC. All of the voltage waveforms (and the derived frequency plots provided below) are as modelled at the Haywards substation.

The df/dt relays should not trip for any of these waveforms as these are not considered to be AUFLS events.

Test 16a

Test 16b

Test 16c

Test 16d

Figure 12 – Frequency traces for the commutation failures



Relay settings


Test A (Block 1):

df/dt = -0.8Hz/s


Time delay = 0.2s

Test B (Block 2):

df/dt = -1.5Hz/s


Time delay = 0.2s

Results


Table 14 –Summarised results for tests 16 to 19

Relay Test Trip Test Trip Test Trip Test Trip Test date

A 16a N** 16b N** 16c N** 16d N** 12-Jun

B 16a N 16b N 16c N 16d N 2-May

D 16a N 16b N 16c N 16d N 2-May

E 16a N 16b N 16c N 16d N 1-May

F 16a N** 16b N** 16c N** 16d N** 12-Jun

G 16a * 16b * 16c * 16d *

* Test 16 has yet to be carried out for this relay.

** The Relay F and the Relay A were subject to these tests with df/dt set points of 0.1 Hz/s and with no guard frequency. Neither of the relays tripped with these settings.

AUFLS III - Appendix B - Transpower New Zealand · System Operator Report: AUFLS RoCoF Relay Performance Requirements Page 1 of 46 AUFLS III - Appendix B Rate-of-Change-of-Frequency

Documents