Observed behavior of distributed photovoltaic systems during major voltage disturbances and implications for power system security

Abstract

As distributed photovoltaics (PV) levels increase around the world, it is becoming apparent that the aggregate behavior of many small-scale PV systems during major power system disturbances may pose a significant system security threat if unmanaged. Alternatively, appropriate coordination of these systems might greatly assist in managing such disturbances. A key issue is PV behavior under extreme voltage events. PV connection standards typically specify aspects of inverter voltage behavior. However unresolved questions remain regarding compliance, ambiguity and transition between versions of these standards. In addition, how major voltage disturbances manifest in the low voltage network is complex, and analysis of operational system data could be particularly useful for establishing the behavior of distributed PV in the field. Our study utilizes 30 s operational PV generation data from 376 sites during two major voltage disturbances in Australia. Australia has one of the highest penetrations of distributed PV worldwide, and as such provides a useful case study. Results show that an aggregate ~30–40% reduction in distributed PV generation occurred during these events, but individual inverter behavior varied markedly. To the authors’ knowledge, this is the first time the aggregate response of distributed small-scale PV to voltage disturbances originating in the transmission system has been demonstrated. Four novel techniques for analyzing events are proposed. Results show a potential increase in system security service requirements as distributed PV penetrations grow. Our findings would seem to have major implications for development of composite load models used by power system operators and for contingency management.

Full text

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
1
Observed behavior of distributed photovoltaic systems
during major voltage disturbances and implications for
power system security
Naomi Stringer
a,b
, Navid Haghdadi
a,b,c
, Anna Bruce
a,b
, Jenny Riesz
d
and Iain MacGill
b,c
a
School of Photovoltaic and Renewable Energy Engineering, University of New South Wales Sydney, Australia
b
Centre for Energy and Environmental Markets, University of New South Wales Sydney, Australia
c
School of Electrical and Telecommunications Engineering, University of New South Wales Sydney, Australia
d
Australian Energy Market Operator, Brisbane, Australia
Key words: Distributed PV, power system security,
voltage disturbance, operational data, voltage response
Abstract
AbstractAbstract
Abstract
As distributed photovoltaics (PV) levels increase around
the world, it is becoming apparent that, if unmanaged,
the aggregate behavior of many small-scale PV systems
during major power system disturbances may pose a
significant system security threat. Alternatively,
appropriate coordination of these systems might greatly
assist in managing such disturbances. A key issue is how
distributed PV might be expected to behave under
extreme voltage events. While PV connection standards
typically specify aspects of inverter voltage behavior,
there are still unresolved questions regarding ambiguity
in these standards, compliance with them, transition
between versions of these standards, and the impact of
transmission level events on voltage in the low voltage
network where the systems reside. Given all of these
complexities, analysis of operational system data could
be particularly useful for establishing the behavior of
distributed PV in the field and forms the motivation for
our study. Our study utilizes a dataset of 30 second
distributed PV operational generation data from 376
sites during two major voltage disturbances in the
Australian National Electricity Market. Australia has one
of the highest penetrations of distributed PV worldwide,
and as such provides a useful case study. Results show
that an aggregate ~30-40% reduction in distributed PV
generation occurred during these events, but individual
inverter behavior varied markedly. To the authors’
knowledge, this is the first time the aggregate response
of distributed small-scale PV to voltage disturbances
originating in the transmission system has been
demonstrated. Four novel techniques for analyzing such
events are proposed and demonstrated; the first two
consider the diversity of individual PV system response in
terms of depth and duration of response. The third and
fourth present a means for analyzing aggregate trends.
Results show a potential increase in system security
service requirements as distributed PV penetrations
grow. A wide range of inverter responses and interaction
between distributed PV and load highlight an increased
complexity in managing power system security. Our
findings would seem to have major implications for
future development of the composite load models used
by power system operators and planners as well as for
contingency management. It is clear that these models
will need to more formally incorporate distributed
generation responses as penetrations of these
technologies continue to grow.
1
11
1 Introduction
IntroductionIntroduction
Introduction
Over recent years there has been unprecedented growth
in the deployment of distributed small-scale rooftop PV
systems within numerous electricity industries around
the world. Whilst this PV uptake has contributed to
reducing sector emissions and energy consumer
electricity bills, it does raise several technical challenges
for safe and secure power system operation. An early
concern and hence focus of research, has been the
technical challenges of distribution network operation
and planning with growing distributed PV penetrations,
particularly maintaining network voltage within safe
limits [1-6]. However, as the penetration of distributed
PV continues to climb, system level security implications
are emerging.
With grid connection through a power electronic
‘inverter’ interface, PV systems have the potential for fast
and precise response, according to local grid
requirements. Prior work has focused on grid codes for
utility scale wind and PV plants [7-9], whilst efforts
considering distributed PV has focused on response to
frequency disturbances [10, 11]. However there is a
growing awareness that high penetrations of distributed
PV can materially impact the power system following
major power system events and that voltage
disturbances also pose a serious challenge [12].
Distributed PV presents a particular challenge due to its
highly dispersed nature, with limited visibility and

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
2
typically very limited, if any, control afforded to power
system operators.
Major grid disturbances such as the loss of a large
generator, network element or load can affect both
system frequency and voltage. Early attention has
focused on PV inverter behavior during frequency
disturbances [13], given the generally system wide
impact of such events. Specific examples include the
German 50.2Hz challenge resulting in substantial, highly
correlated, distributed PV disconnection [10, 11]. As well
as widespread under-frequency load shedding in Hawaii
exacerbated by legacy distributed PV [14] and under
frequency PV tripping in the remote Alice Springs grid in
Australia [15]. However inverter response to voltage
disturbance is also significant, as recognized in the
considerable efforts being undertaken internationally to
develop fault detection and ride through strategies [16-
18], as well as review [10, 11, 19-24] and test [25, 26]
inverter connection standards.
Along with inverter technology progress and other
market developments, revision of grid connection
standards has resulted in a diverse fleet of installed
inverters, with potentially highly varied responses to
major voltage disruptions, in some jurisdictions. As
distributed PV penetrations grow, there is a clear need to
better understand the behavior of their inverters during
possible power system security events.
In an effort to address these challenges, recent work [27-
29] has developed composite load models incorporating
distributed energy resources (DER), predominantly solar
PV and battery energy storage systems, to better predict
the aggregate response of load following contingency
events including voltage disturbances. Dynamic models
that accurately capture the behavior of load, and
increasingly also DER, during disturbances play a key role
in power system operation, including for the
determination of constraints and power system limits for
security and stability, as well as studies to assure the
stable connection of new generation, and the correct
allocation of reserves for ancillary services. However, the
incorporation of DER into such models is particularly
challenging given the limited experience with these
technologies to date, their potentially varied operation
and the complexities of Low Voltage (LV) system
operation during disturbances. As such, there is
considerable value in actual DER operational data
analysis during disturbances to both inform as well as
validate these models.
As PV uptake continues worldwide, Australia provides a
unique snapshot into the future for many grids due to its
high penetration of distributed PV. More than 20% of
residential dwellings in the Australian National Electricity
Market (NEM) have PV installed [30, 31], with over two
million systems in total. The NEM regions of Queensland
and South Australia have some of the highest residential
penetrations in the world with systems installed at over
30% of standalone dwellings [32]. Experiences in the
NEM are therefore relevant to power systems worldwide
as a window into a possible future with very high
distributed PV penetrations.
Despite its growing role, direct system-wide SCADA is not
available for distributed PV [33], which is therefore
currently seen as reduced load, to be met by utility-scale
generation. The industry’s early focus has been on
increased understanding of inverter frequency response
set points [34]. Further, whilst there is some publicly
available operational data for distributed PV [35], it is
typically reported on 5 minute or longer intervals and is
therefore of limited benefit for assessing response to grid
disturbances.
The analysis set out in this paper utilizes a novel
operational data set, reporting individual household and
business distributed PV generation, load, and local
network voltage at 30sec time intervals, from solar
monitoring company Solar Analytics [36]. It examines the
behavior of distributed PV systems during two major
system voltage disturbances following large, non-
credible contingency events in the NEM. The first, in
South Australia, was due to a large gas generating plant
failure, whilst the second, in Victoria, followed a
transmission fault during extreme heat conditions.
Our study makes two key contributions. First, it presents
four novel techniques for analyzing this operational data
set. Two techniques are presented for analyzing
individual PV systems with regards to depth and duration
of their response to these events. Two further techniques
are presented for analyzing aggregate response. These
include spatial analysis of response to voltage
disturbance, and upscaling observed behaviors from
monitored PV systems to the total installed distributed
PV capacity. These techniques provide a possible
framework for future analysis of large operational data
sets examining the behavior of many small DERs.
Secondly, this analysis highlights that the response of
distributed PV during voltage disturbances can be
material, with a ~30-40% reduction in aggregate PV
power output during the two events studied, and
therefore has significant implications for management of
power system security. The observed responses are also
highly diverse across different inverters in different

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
3
locations, highlighting the criticality of operational data
for security modelling and analysis as distributed PV
penetrations increase. We particularly consider the
implications of these findings for the provision of raise
frequency control services.
The rest of this paper is structured as follows. Section 2
provides an overview of data sets used in the analysis and
section 3 summarizes the case studies examined and
notes expected behavior given inverter connection
standards. Section 4 sets out the method, whilst section
5 sets out the detailed method and findings. The detailed
method presents four novel analytical tools; two for
assessing the diversity of individual system behaviors
(depth and duration of response), and two for assessing
the overall power system impact (spatial analysis and
upscaling). Section 6 concludes the study.
2
22
2 Data sets
Data sets Data sets
Data sets
2.1 Solar Analytics data set
Data including PV generation, local network voltage and
frequency from 376 individual sites at 30s time intervals
and 428 individual sites at 5min intervals was provided by
solar monitoring company Solar Analytics [36].
Systems with erroneous data, including zero generation,
non-zero generation outside sunshine hours, periods of
negative generation, constant 50Hz frequency
measurements (i.e. no deviations or ‘stuck’ values), and
missing data during the period immediately following the
event were removed from the analysis. Missing data
(other than immediately following the event) was
forward filled. This is particularly relevant for the
Victorian data set due to apparent monitoring and
communication issues with the Solar Analytics system on
the day of the event. Voltage measurements from the
Solar Analytics devices are V
rms
recorded over the final
100ms in each measurement interval. Note, therefore
they provide a snapshot rather than continuous visibility
of local network conditions. PV generation was
converted from energy to average power per time
interval.
Table I summarizes the number of sites in the Solar
Analytics data set (after filtering).
Table I – Solar Analytics data summary: number of sites
South Australia
Victoria
30s data set 214 162
5min data set 260 168
2.2 Public PV data sets
A publicly available database of historical PV generation
from the PVOutput.org website [37] was accessed via the
Australian Photovoltaic Institute (APVI) live map
database [32, 38]. This database contained 400 sites in
Victoria and 268 in South Australia with sites either
reporting on 5min, 10min or 15min basis. The missing
intervals and likely invalid data (due to monitoring issues)
were not considered in the analysis. This dataset was
primarily used for validation of the Solar Analytics data.
The Australian Clean Energy Regulator collects
information regarding distributed PV installations which
are installed under the Small Scale Renewable Energy
Target [39]. Data collected includes PV capacity,
installation postcode and installation date, hence
applicable inverter connection standard. This dataset
was used for the upscaling described in section 5.8.
2.3 Spatial data sets
Australian Bureau of Statistics Australian postcodes [40],
and boundaries of Greater Capital City statistical areas
[41] are used as the basis for spatial analysis and
upscaling. This data is also publicly available.
2.4 System operational data sets
The statewide load in South Australia and Victoria has
been estimated using 4s SCADA data publicly available
through AEMO’s NEMWeb portal [42], accessed using
the NEMOSIS open source tool [43]. This data set
contains 4s data for all registered generators and
interconnectors in the NEM.
Historical Frequency Control Ancillary Services (FCAS)
data was also obtained through AEMO’s NEMWeb portal
[42], accessed using the NEMOSIS open source tool [43].
The data set obtained contains FCAS enablement for
each NEM region over the period November 2017 to
November 2018.
3
33
3 NEM event c
NEM event cNEM event c
NEM event case studies
ase studiesase studies
ase studies
and
and and
and specified
specifiedspecified
specified
inverter
inverterinverter
inverter
response
responseresponse
response
The low population density in Australia has resulted in a
long grid with relatively low levels of transmission
interconnection between states. The first case study
region of South Australia has two points of
interconnection with the neighboring region (Victoria),
and is a relatively small power system, serving a typical
peak demand of 3GW. The South Australian generation
fleet contains significant wind capacity, along with a large
gas thermal plant, a CCGT, gas cogeneration and some
peaking gas (and diesel) plants.
The state has an extremely high penetration of
distributed PV, which can occasionally supply more than
40% of demand across South Australia for short periods
of time [44, 45]. Significantly, AEMO forecasts PV growth

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
4
will continue to the extent that periods of negative
demand will be observed in South Australia as early as
2025-26 [46].
Victoria, by contrast, is a comparatively large grid
(around 10GW peak demand), with interconnections to
the neighboring regions of South Australia, New South
Wales and via an undersea HVDC cable, Basslink,
Tasmania. Generation capacity is primarily coal, and
whilst distributed PV uptake has grown consistently in
terms of absolute installed capacity [47], it has lagged
South Australia in terms of penetration (Table II). Again,
however, strong uptake of distributed PV is considered
feasible for the region over the next two decades [47].
Table III summarizes the situation at the time of the non-
credible contingency events being studied here.
Table II – Case study region characteristics [48-51]
South
Australia
Victoria
Number of customers
1
858,647
2,797,458
Transmission network length
1
5,524km 6,559km
Wind generation (%
statewide
rated generation capacity)
2
36%
13%
Thermal generation (% statewide
rated generation capacity)
2
51% 64%
Peak load
2
3.4GW 10.5GW
Distributed PV capacity
2
746MW 1,221MW
Distributed PV penetration
2,3
29% 15%
Forecast distributed PV 2036-37
1
2.1 GW 4.3 GW
1
2017,
2
2018,
3
Penetration means the proportion of ‘free standing
and semi-detached’ dwellings with PV installed
Table III – Event conditions [32, 39, 52, 53]
South Australia
Victoria
Event date and time 15:03:46 AEST
Fri 3 Mar 2017
15:18:54 AEST
Thurs 18 Jan 2018
Event day maximum
temperature
28°C 40°C
Load at time of event 2.0GW 8.7GW
3.1 South Australian event 3 March 2017
The first case study considers a major voltage
disturbance in South Australia on 3 March 2017 following
the explosive failure of a capacity voltage transformer in
the switchyard of Torrens Island (the largest gas
generator in the state). This triggered a number of faults
starting at 15:03:46 Australian Eastern Standard Time
(AEST). An initial loss of 410MW thermal generation
occurred within 1.5s, with a further 200MW lost shortly
afterwards.
The voltage levels fell to around 0.1pu on one phase at
Torrens Island and flow across the Heywood
Interconnector from Victoria increased [54]. Notably,
conditions caused by the contingency event shared some
similarities to those which caused a system black event
in South Australia on 28 September 2016 [54].
Although the Under Frequency Load Shedding scheme
did not operate, demand was observed to reduce by
400MW, believed to be caused by ‘shake off’ of load in
response to under voltage conditions [54]. This was
almost immediately followed by a 150MW increase in
state demand, suggested by AEMO to have possibly been
caused by distributed PV systems “shutting off in
response to the voltage disturbance” [54]. The analysis
presented in this paper provides evidence that the
150MW increase was indeed likely due to PV response
during the event, although we are unable confirm
whether the PV inverters responded to voltage
conditions directly, or perhaps other aspects of the
disturbance.
3.2 Victorian event 18 January 2018
At 15:18:54 AEST on 18 January 2018, during extreme
high temperature conditions, a series of transmission
faults occurred at Rowville terminal station (a major
substation) and nearby sections of the transmission
network on the outskirts of Melbourne [55]. These faults
followed the failure of a 500kV single phase current
transformer. As result, voltage levels fell as low as 0.51pu
in the 500kV network and 0.45pu in the 200kV network.
Although load shedding was not instructed and indeed
no single point of load loss was observed, there was an
aggregate ~550MW reduction in demand observed
across Victoria, followed shortly afterwards by an
increase in load of approximately ~110MW. A modest
frequency rise was observed across the NEM following
this event due to load loss. Again it was suggested that
distributed PV may have contributed to this apparent
load increase, however prior to our study, there was no
evidence to support this hypothesis.
3.3 Inverter connection standards for voltage
excursion
Internationally, there has been considerable effort to
update connection standards for small inverters to
incorporate requirements similar to those existing for
utility scale generators, such as frequency, voltage and
phase angle jump (or vector shift) ride through
requirements. Despite these efforts and ongoing
development of power system models, the small and
distributed nature of devices makes it infeasible to audit
all connections in the manner used for utility plant.
Therefore, there is limited evidence to support the
efficacy of these standards, and as penetrations increase,
this necessitates analysis of operational data to support
development of power system models.

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
5
In Australia the most widely adopted inverter connection
standard is AS/NZS 4777 ‘Grid connection of energy
systems via inverters’ [56, 57]. Australian standards are
not mandatory unless specified in legislation, however
compliance with AS4777 is typically required by
distribution system operators for connection to their
grids. This standard has gone through a number of
revisions over the past 15 years. As result, the Australian
PV fleet is comprised of systems installed under the
current standard (‘post 2016’) and a substantial legacy
fleet of inverters installed under the superseded
standard (‘pre 2015’) as well as during a 12 month
transition period in which either standard could be
applied. Notably, anti-islanding requirements feature in
both the superseded and current standard (Appendix A).
The superseded standard lists an allowable range for
anti-islanding set points, where Pre 2015 systems over
and under voltage set points can fall between 0.87-1pu
and 1-1.17pu respectively. In comparison, the current
standard states specific over and under voltage set points
at 0.87pu and 1.13pu respectively. Post 2016 systems are
also required to ramp at no greater than 16.67% of rated
capacity per minute after reconnection (6min to rated
capacity).
Without knowledge of local voltage conditions it is not
possible to assess compliance with anti-islanding
requirements. This would require detailed modelling of
the LV network and is outside the scope of this study.
However, given the severity of the voltage disturbances
studied (with 0.1pu observed at Torrens Island in the
South Australian case and 0.45pu on the 200kV network
in the Victorian case), local under voltage is likely and
therefore the anti-islanding requirements are a plausible
cause of PV response. More PV disconnections are
expected close to the source of the disturbance,
consistent with the voltage dip being deepest at the site
of the disturbance. However, given the widespread loss
of load following each event localized over voltages may
also have occurred and it is unclear whether anti-
islanding due to over voltage may also have caused at
least some of the PV response.
In addition to the anti-islanding requirements set out in
Appendix A, the current standard also includes
requirements regarding active anti-islanding protection,
limits for sustained operation and volt-watt response, as
well as specifications for volt-var response, voltage
balance mode, fixed power factor and reactive power
mode. It is possible that these response modes may have
impacted some inverter behaviors in the operational
data set. Notably, the present standard is silent on
voltage phase angle jump which may also have been a
cause of PV response during the voltage disturbances
studied.
4
44
4 Methods
MethodsMethods
Methods
The analysis approach aims to provide insight into the
diversity of individual PV system behaviors, as well as
characterizing aggregate trends. Custom algorithms were
developed and implemented in Python and QGIS to
analyze the operational data set, with data cleaning
completed as set out in section 2. Four novel techniques
are presented.
The first two techniques characterize individual system
responses with regards to depth and duration of
response. This provides insight into the variability of
power output response across the sample and highlights
the importance of developing tools to accurately
estimate the operation of the broader PV fleet.
Section 5.2 sets out the detailed method for
characterizing depth of response, whilst section 5.3
presents findings from the South Australian and Victorian
events. Section 5.4 sets out the detailed method for
characterizing duration of response and section 5.5
presents findings.
The third and fourth techniques consider aggregate PV
response. A spatial analysis method is developed and
then used as the basis to upscaling. This upscaling
technique models the behavior of the entire installed
distributed PV capacity. The upscaled PV loss estimates
are shown to present substantial implications for
managing power system security.
Section 5.6 provides detailed method for spatial analysis,
with section 5.7 presenting findings. Section 5.8 then sets
out the detailed upscaling method and section 5.9
presents findings and considers power system security
implications.
5
55
5 Detailed m
Detailed mDetailed m
Detailed method
ethodethod
ethods
ss
s
and
and and
and event findings
event findingsevent findings
event findings
5.1 Initial event characterization
In the first instance, the aggregate PV behavior observed
in both the South Australian and Victorian datasets was
examined, and the 4s state wide load profile was
estimated using AEMO SCADA data for all generators and
interconnectors.
5.1.1 South Australia
Fig. 1 shows approximate state wide load in South
Australia at the time of the event, after an initial
reduction in load of around 400MW, load increased by
around 150MW.

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
6
Fig. 1. South Australia: approximate statewide demand during event,
4s data
The Solar Analytics operational PV data set used for this
study shows a clear PV response to the disturbance with
a 42% reduction in aggregate generation observed (Fig. 2
(a) and (b)). As stated in 2.1, the voltage data provides a
‘snapshot’ of voltage conditions and therefore the
average voltage presented in Fig. 2(a) does not capture
the true local voltage variation.
(a)
(b)
Fig. 2. South Australia event: (a) Aggregate average power (top) and
average voltage RMS (bottom) over the day (b) aggregate average
power during the event, measured at Solar Analytics sites, 30s data
5.1.2 Victoria
Fig. 3 shows the approximate Victorian statewide load at
the time of the Victorian case study event with an initial
reduction of ~550MW followed by an increase of
~110MW.
Fig. 3. Victoria: approximate statewide demand during event, 4s data
Similarly to the South Australian case there was no
evidence that the increase in load was due to reduced
distributed PV generation prior to this study. However
the data set examined indicates substantial PV response,
with an aggregate reduction of 28% observed (Fig. 4 (a)
and (b)).
(a)
(b)
Fig. 4. Victorian event: (a) Aggregate average power (top) and
average voltage RMS (bottom) over the day (b) aggregate average
power during the event, measured at Solar Analytics sites, 30s data
5.2 Depth of response analysis
The diversity of response across individual PV systems
was initially assessed in terms of fractional power loss to
nadir (%), where seven categories were developed to
characterize response. These categories are summarized
in Table IV where % is calculated as per (1) for
individual premises. 



 is the operational power
immediately prior to curtailment and 



 is the
minimum operational power after the event interval for
system :

%





















1
The first category ‘Ride through’ indicates sites which are
assessed as having been relatively unaffected by the
event. Whereas the seventh category ‘Disconnect’
indicates sites for which generation was reduced to less
than 0.1kW for at least one measurement interval
immediately following the event.
Categories 2 to 6 exhibit a range of partial responses.
Without more detailed time resolution data it is not
possible to assess precisely what is actually occurring at
these sites. For instance, it is feasible that power output
at these sites reduces to zero without triggering actual
disconnection, and therefore not waiting the full 60s until
reconnection (see Appendix A). Given the 30s data
resolution, cases where PV systems reduce generation to
zero briefly may be seen to exhibit modest curtailment.
3:02 PM 3:04 PM 3:06 PM 3:08 PM 3:10 PM 3:12 PM
Timestamp [3 March 2017]
1400
1600
1800
2000
2200
8 AM 10 AM 12 PM 2 P M 4 PM 6 PM 8 PM
Time [3 March 2017]
200
400
600
240
245
250
2:50 PM 3:00 PM 3:10 PM 3:20 P M 3:30 PM
Time [3 March 2017]
300
400
500
600
700
3:18 PM 3:20 PM 3:22 PM 3:24 PM 3:26 PM 3:28 PM
Time [18 January 2018]
8200
8400
8600
8800
7 AM 9 AM 11 AM 1 PM 3 PM 5 PM 7 PM
Time [18 January 2018]
500
1,000
1,500
240
245
250
3:05 PM 3:15 P M 3:25 PM 3:35 PM 3:45 PM
Time [18 January 2018]
1,200
1,400
1,600
1,800
~150MW increase in demand
following disturbance
~110MW increase in demand
following disturbance

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
7
Table IV – Definition of categories for diversity analysis
Category Fractional power
loss PL%
Additional criteria
1 – Ride through ≤ 4% Not applicable
Partial responses
2
–
Slight dip
≤ 10%
Cat 1 is false
3 – Mild >10% and ≤25% Not applicable
4
–
Medium
>25% and
≤50%
Not applicable
5 – Significant >50%, and ≤75% Not applicable
6
–
Severe
>75%
Cat 7 is false
7 - Disconnect Not applicable






<0.1kW
5.3 Depth of response findings
Initial diversity analysis was based on depth of response,
where Fig. 5 (indicating normalized average profiles for
each category in South Australia) serves to confirm the
categorization process as per Table IV.
Fig. 5. Average PV output power by category, normalized to pre
event power, 30s data
5.3.1 South Australia
Examination of Fig. 6 shows that the majority of
individual systems across South Australia exhibited
disconnect (38%) or ride through (41%) behaviors, with
relatively few partial responses. The majority of sites
which disconnected were located in the Adelaide region,
which is to be expected given Adelaide’s proximity to
Torrens Island Power Station and thus, the source of the
original transmission level voltage disturbance.
Fig. 9 maps the number of sites in each category per
postcode region and also indicates the approximate fault
location. It shows that whilst the majority occurred
within Greater Adelaide, disconnection responses were
observed across the entirety of the state. Given South
Australia’s relatively low level of system strength [58],
this is unsurprising, however does emphasize the
criticality of effectively managing PV response to major
disturbances under such circumstances.
Fig. 6. South Australia event: Proportion of sites in each response
category, 30s data
5.3.2 Victoria
By contrast, the majority of sites across Victoria rode
through the contingency event (68%) whilst around one
fifth of systems disconnected (21%), as shown in Fig. 7.
The higher proportion of ride through sites is consistent
with the higher level of system strength in Victoria
compared with South Australia, as well as the less
extreme voltage disturbance observed. Similarly to the
South Australian case, sites that exhibited a moderate
response (did not disconnect or ride through) tended to
occur in Category 2 or 3 with very few examples of
curtailment beyond Category 3 without disconnection.
Fig. 7. Victoria event: Proportion of sites in each response category,
30s data
In contrast to the South Australia event, Fig. 10 shows
that the sites located outside Greater Melbourne and
which disconnected (Category 7) were comparatively
close to the original disturbance.
5.4 Duration of response analysis
Next, diversity of response across individual system was
assessed in terms of ‘total response time’ (
,
),
or the period between the curtailment interval (

) and
the interval in which generation returns to within 10% of
the pre-event level, as indicated in Fig. 8 and expressed
in (2) and (3).
Fig. 8 Total response time illustrative example
The time at which generation returns to within 10% of
the pre-event level, 

for each system , is
calculated as follows:



,















!"








#

$






%


#
&


'

The total response time is then:












3


!"



)





*"
+,-

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
8
Fig. 9. Mapping the South Australian event: count of PV sites by category and postcode. Insets show sites located in the Greater
Adelaide region only, filtered for (a) Category 7, (b) Categories 2-6 and (c) Category 1. Pink lines indicate transmission network. Red
triangle indicates approximate fault location
Fig. 10. Mapping the Victorian event: count of sites by category and postcode. Insets show sites located in the Greater Melbourne
region only, filtered for (a) Category 7, (b) Categories 2-6 and (c) Category 1. Pink lines indicate transmission network. Red triangle
indicates approximate fault location
5.5 Duration of response findings
Substantial diversity is observed in the duration of
individual systems’ response times across the 30s
measurement intervals. To manage the frequency during
contingency events, AEMO procures Frequency Control
Ancillary Services (FCAS) on a 6s, 60s and 5min basis for
contingency raise and lower services [59]. As
penetrations increase, loss of distributed PV during
Category 7 Disconnect site,
~420km direct route from fault
Category 7 Disconnect
site, ~170km direct
route from fault

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
9
disturbances may need to be considered when procuring
FCAS.
5.5.1 South Australia
Fig. 11 shows that the most commonly occurring total
response time was 150s (2.5min), which was observed at
28% of the 106 sites for which a response occurred (that
is, a reduction greater than 10%). Fig. 11 also shows that
typically sites that fully disconnected took longer to
return. It indicates that additional 60s and 5min FCAS
raise may be required in future given that 37% of sites
had a total response time above 5min, whilst 53% of sites
exhibited a response time between 60s and 5min
(considered further in section 5.10). To assess the need
for additional 6s FCAS raise services would necessitate a
higher resolution data set.
Fig. 11. South Australia event: distribution of total response times,
106 sites, 30s data
Fig. 12. South Australia event: distribution of total response times for
disconnect sites, indicating AS4777 version, 82 sites, 30s data
Examining only sites that disconnected, Fig. 12 shows the
distribution of total response times by AS4777 version
1
.
Fig. 12 suggests that systems installed post 2016 make up
the majority of response times greater than or equal to
300s. This is to be expected, given the current standard
requires a 6min ramp to rated capacity (Table IV). It
suggests that the overall fleet response profile will
change as the proportion of systems installed under the
current standard increases.
Further, Fig. 12 shows that a significant proportion of
post 2016 inverters exhibited a total response time of
less than 360s, which may indicate non-compliance with
the standard’s ramp rate requirements. However, the
total response time alone cannot be used to assess
1
The AS4777 version is determined using installation date.
compliance with the standard ramp rate, since the ramp
rate is a function of system capacity, whilst total response
time is determined using pre-event operating level, and
systems are unlikely to be operating at full capacity. This
is an area for further investigation.
5.5.2 Victoria
The distribution of total response times in Victoria is
shown in Fig. 13 for sites that curtailed by 10% or more.
Similarly to the South Australia event, generally those
sites that disconnected (Category 7) took longer to
recover compared with the other response categories.
Notably, 71% of sites had a total response time of
between 60s and 5min, whilst 22% had a total response
time of greater than 5 minutes. As noted for South
Australian case, this may result in increased FCAS
requirements for both 60s and 5min contingency raise in
the future, this is considered further in section 5.10.
It is important to note that there are fewer Victorian sites
in Categories 3-7 compared with South Australia, since
the majority of Victorian sites rode through the event.
Fig. 13. Victoria event: distribution of total response times, 45 sites,
30s data
Fig. 14. Victorian event: distribution of total response times for
disconnect sites, indicating AS4777 version, 35 sites, 30s data
Examining only sites that disconnected, Fig. 14 shows
that inverters installed post 2016 typically returned more
slowly compared with pre 2015 or transition period
systems. Similarly to the South Australian case, this is
consistent with the new 6min ramp rate requirement in
the current standard, and assessing compliance is an area
for further investigation.
0 60 120 180 240 300 360 420 480 540 600 660 720
Total response time (s)
0%
10%
20%
30% Cat 3 - Mild curtail
Cat 4 - Medium curtail
Cat 5 - Significant curtail
Cat 6 - Severe curtail
Cat 7 - Disconnect
6
0
s
6s
5min
6
0
s
6s
5min

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
10
5.6 Spatial analysis method
Inverter response is dictated by the inverter settings and
the voltage conditions experienced. Local voltage
conditions are affected by how the disturbance manifests
across the network and therefore depends upon a wide
range of factors including system strength at the time of
the event. Both voltage disturbance and phase angle
jump (or vector shift) will be deepest at the site of the
disturbance and then decrease for sites that are more
electrically remote from the disturbance.
The purpose of this study’s spatial analysis is threefold;
firstly, to establish whether PV response exhibits a spatial
pattern thereby indicating likely response to voltage
disturbance or vector shift. Secondly, spatial analysis can
be used to assess the severity of PV response for the
whole power system (as an input to upscaling). Thirdly, if
local voltage conditions can be estimated, the degree of
response across different regions may indicate the
degree of compliance with AS4777 voltage set points
(outside the scope of this study).
Spatial analysis consists of two steps: initially, zone
boundaries are determined and PV systems are
categorized using system location. Then, the degree of
PV system response is determined, with two metrics for
the degree of response assessed (proportion
disconnecting and power output reduction).
5.6.1 Setting zone boundaries
The zones should ideally reflect graduating disturbance
severity and are therefore expected to depend heavily on
electrical distance. Without extensive power system
modelling and access to a detailed LV network model
(outside the scope of this study), it is not possible to
determine such spatial boundaries accurately. Instead,
boundaries are set using concentric circles as a proxy for
electrical distance. Three criteria are applied in this
process: (1) a minimum number of sites per zone, (2) a
minimum distance between zones given the high density
of sites in some districts
2
, and (3) consideration of large
generators online at the time of the event. The weighting
of each criteria is considered on a case by case basis,
given the limitations of available data.
5.6.2 Analyzing the degree of response
The degree of response is assessed using two metrics: the
proportion of sites that disconnected and reduction in
generation within each zone.
2
This can otherwise result in concentric circles with small
differences in radii, and subsequently, systems being classified
5.6.2.1 Proportion disconnect
The proportion of sites that disconnected in zone . is
expressed in (4) where /. is the number of sites in
zone . for which Category 7 Disconnect (Table IV) is true,
and 0. is the total number of sites in zone .:


12+-3,-

.


/

.

0

.

(4)
5.6.2.2 Generation reduction
The generation reduction in zone . is calculated using the
average PV capacity factor observed in each zone, in
order to avoid large systems distorting results. The
capacity factor profile over time for individual system, 
is calculated as follows:
45










6
7



89
(5)
Where 

 is the average power generation profile in
interval  for system , and where 6
789
is the DC
capacity of system . The average capacity factor profile
over time is then calculated for zone .:
45
:

.




;
45




<

=
>
0

?
@
,A,

0


+2,+

2

.
#
(6)
Finally, the generation loss in zone . is calculated as per
(7), that is, the change in average capacity factor for zone
. at the time of the event. Where 
BB
is the time
at which the aggregate nadir occurs and noting that

BB
is based on the total aggregate profile and is
consistent across zones.


C,

A,1D-23

.


45
:

.






45
:
E
.


BB


F
45
:

.




(7)
5.7 Spatial analysis results
5.7.1 South Australia
5.7.1.1 Setting zone boundaries
Given the limited number of synchronous generators
operating at the time of the event and limited number of
sites available in the data set, the zone boundaries were
set to ensure a reasonable spread of sites per zone, and
to ensure zones are set a reasonable distance apart.
Characteristics of the zones used in South Australia are
summarized in Table V and mapped in Fig. 15.
in different regions despite likely being a similar electrical
distance from the disturbance.

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
11
Table V – South Australian zone boundaries: data set summary
Number of sites Capacity (kW
DC
) Radius (km)
Zone 1 120 577 34
Zone 2 56 259 182
Zone 3 38 222 942
Total
214
1,058
-
Fig. 15. South Australia zone boundaries, pink indicates transmission
network
5.7.1.2 Spatial response
Fig. 16 shows a clear spatial trend in terms of the
proportion of sites that disconnected and the reduction
in generation with the most severe response close to the
disturbance in zone 1. Zone 1 experienced an extremely
high rate of disconnections with just under 50% of
systems disconnecting.
Fig. 16. South Australia spatial response of distributed PV systems
5.7.2 Victoria
5.7.2.1 Setting zone boundaries
In the Victorian case, zone boundaries were set primarily
based on the location of large synchronous generating
units operating at the time of the disturbance, with
consideration also given to the number of sites per zone.
Characteristics of the zones applied in Victoria are
summarized in Table VI and mapped in Fig. 17.
Table VI – Victorian zone boundaries: data set summary
Number of sites
C
apacity (kW
DC
)
Radius (km)
Zone 1 63 821 38
Zone 2
48
997
127
Zone 3 51 1,238 506
Total 162 3,056 -
Fig. 17. Victoria zone boundaries, pink indicates transmission
network, diamonds indicate generators operating at >400MW
(approximate operating level at time of disturbance is shown)
5.7.2.2 Spatial response
Similarly to the South Australia case, there is a clear
spatial trend (Fig. 18) in the percentage of generation
reduction and proportion of PV system disconnections
immediately following the disturbance. Notably, the
response is not as severe as that in South Australia,
where 49% of zone 1 systems disconnected, compared
with 30% of sites in the Victorian zone 1 region.
Fig. 18. Victorian spatial response of distributed PV systems
5.8 Upscaling Method
Upscaling is an important tool for projecting observed
behavior from a sample to an entire installed fleet of PV
systems. It is similarly applied for cases where only a
sample of distributed PV data is available in [32, 35, 38,
60].
Clearly, there are a number of variables that could factor
in the upscaling technique applied; namely inverter
connection standard version, response ‘zones’ (section
5.6) and system capacity, with potential for inverter
manufacturer and model to also feature. The most
significant factor for response observed is distance from
the event origin, therefore the spatial analysis has been
utilized here as the basis to upscaling. Future iterations,
particularly those concerned with the recovery profile,
should take into account standard version, given the total
response time results set out in 5.5.
South Australian
state border
Torrens Island
Zone 3
Zone 2
Zone 1
Victorian state
border
Rowville terminal station
Zone 3
Zone 2
Zone 1

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
12
Upscaling involves two steps; first, the upscaled profile in
zone . is calculated using the average capacity factor
profile (5.6.2.2), and the fleet installed capacity
6
789
.:
G

.




45
:

.



H

6
7

89


.

(8)
The state wide upscaled profile G is then calculated
as the sum across all spatial zones:
G





I
G

.



J
K
=
>

?
@
,A,

L


+MN2NO

.3,+
#
(9)
5.9 Upscaling results
5.9.1 South Australia
The upscaling results for South Australia show an
estimated 45% or 205MW loss in distributed PV
generation (Fig. 19, Fig. 20). This reduction is slightly
greater than the 42% observed in the raw data, perhaps
due to the sample containing a smaller proportion of
sites (by capacity) in the most affected region of zone 1
compared with the overall fleet.
Fig. 19. South Australia upscaled PV generation estimate
Fig. 20. South Australia upscaled PV generation estimate, time of
event
The upscaled reduction in generation in the three zones
is summarized in Table VII. The upscaled total PV
generation loss estimate of 205MW is 36% greater than
the observed 150MW increase in state wide load. There
are several possible explanations for this discrepancy
including a) that the loss of load which also occurred
following the event effectively ‘masked’ total PV
response, b) that filtering out zero generation sites
during data cleaning reduces representativeness of the
sample since some sites in the fleet will also be operating
at zero, and c) that the sample is not sufficiently
representative, particularly regarding differences due to
inverter connection standard version and system
capacity (larger systems may be subject to additional
requirements compared with smaller systems).
The upscaled power loss estimate is approximately 10%
of the load observed immediately prior to the event, and
it is important to stress that this was a near miss event,
with similar conditions to those resulting in a system
black event in South Australia in 2016. If the forecast
2037 PV capacity of 2.1GW had been installed at the time
of this event with a proportional response, the PV
generation loss estimate of ~580MW would be
equivalent to 29% of demand, posing clear challenges for
the system operator to avoid system black.
Table VII – South Australia upscaling results
PV
Fleet
capacity
(MW)
Upscaled
power loss
(MW)
% of load
*
% of load
in 2037**
Zone 1 430 149
Zone 2
196
44
Zone 3 118 12
Total 744 205 10% 29%
* Load prior to the event was ~2.0GW
** Forecast PV capacity of 2.1GW
5.9.2 Victoria
Upscaling in Victoria resulted in an estimated 25% or
177MW reduction in PV generation (Fig. 21, Fig. 22). This
is less than the 29% reduction observed in the raw data
set due to a number of commercial scale systems which
disproportionately impacted the aggregate profile.
Fig. 21. Victoria upscaled PV generation estimate
Fig. 22. Victoria upscaled PV generation estimate, time of event
The upscaled reduction in each zone is summarized in
Table VIII. Similarly to the South Australia case study, the
upscaling method has substantially overestimated PV
response (by approximately 61%) in comparison with the
60s
5min
60s
5min

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
13
observed 110MW increase in state load. Possible causes
of the overestimation are consistent with those set out in
5.9.1. The PV response constituted only 2% of state load
at the time of event. If the same event occurred in 2037
where the forecast uptake of PV had been realized, the
upscaled impact would be around 600MW, which would
still only constitute around 7% of total demand, however
this may still have implications for FCAS procurement.
Table VIII – Victoria upscaling results
PV Fleet
capacity
(MW)
Upscaled
power loss
(MW)
% of load
1
% of load
in 2037
2
Zone 1
449
96
Zone 2 425 59
Zone 3 362 22
Total
1,237
177
2%
7%
1
Load prior to the event was ~8.7GW
2
Forecast PV capacity of 4.3GW
5.10 Operational implications
The analysis in this paper suggests that distributed PV
could contribute to contingency FCAS requirements in
several ways:
1. A deep fault near a metropolitan center could
cause a significant proportion of region-wide
distributed PV to disconnect. With ongoing
growth in distributed PV, the loss of the
quantities of PV indicated here could start to
approach credible contingency sizes due to loss
of a unit.
2. A loss of a large unit close to a metropolitan
center could be associated with a voltage dip,
causing additional disconnection of distributed
PV. This would increase the size of the original
contingency, summing the loss of the unit, with
the loss of distributed PV. This may mean that
the largest credible contingency is no longer
defined simply by the size of the largest unit, but
could be defined by a slightly smaller unit, if it
happens to be close to a metropolitan center.
In typical periods, the largest credible contingency (loss
of a single unit) in the NEM is in the range 600-750MW.
Examination of historical FCAS data shows the global
maximum FCAS contingency enablement at around 750 -
900MW (enabled under unusual circumstances).
However, the lowest level of 60s contingency FCAS
enabled was around 200MW. The FCAS enablement
level is smaller than the anticipated largest credible
contingency, because AEMO assumes load relief of 1.5%
of system demand [61].
It is important to note that estimates of PV contingency
sizes are not directly comparable with FCAS raise
enablement, given complexity around several factors
including load relief. Note also that the two events
considered in this study are classified as non-credible
contingencies. AEMO only enables FCAS based upon
contingencies classified as credible, under the National
Electricity Rules.
Fig. 23. Historical FCAS requirement duration curve (1 Nov 2017 – 30
Nov 2018) for all hours and during sunlight hours (07:00 – 17:00 AEST)
5.10.1 South Australian event
Examining the upscaled PV profile in Fig. 20, it is possible
to estimate PV generation reduction from pre-event
levels at 60s and 5min after the event. It is useful to
consider the potential PV contingency size in context of
contingency requirements.
An estimate of 60s and 5min generation deficit due to PV
behavior is presented in Table IX, along with an estimate
for the future based on projected PV uptake in South
Australia.
Table IX – South Australian estimated generation deficit due to PV
behavior during disturbances, at time of event and forecast
Generation
deficit
(MW)
2017
(event)
2020 2025 2030 2035
60s after
event
205 241 344 461 536
5min after
event
40 47 67 90 105
Note: estimates based on upscaled profile. For instance, 60s estimate
shows the difference between pre-event upscaled PV generation and
PV generation 60s later. Forecasts scaled based on projected PV
capacity.
Table IX indicates the largest PV generation deficit may
be approaching the typical largest contingency size in the
NEM (600-750MW), especially under low load conditions
where larger units may not be operating at high levels.
This analysis only considers a similar fault in the Adelaide
metropolitan area; similar events could occur near other
metropolitan centers, and may produce similar results.
This should also be explored in future work.
These results indicate that it may already be appropriate
to consider distributed PV when setting FCAS

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
14
requirements during some periods, which is not current
practice in the NEM.
As shown in Fig. 1 and Fig. 3, considerable loss of load
occurred along with the reduction in PV generation.
Understanding the balance between load ‘shake off’ and
reduction in aggregate PV will be important for setting
appropriate FCAS requirements, considering both depth
as well as duration of response.
5.10.2 Victorian event
Table X summarizes the current and projected
generation deficit estimates, given the upscaled PV
profile for the Victorian event (Fig. 22). Considering the
projected PV uptake in Victoria, it is estimated that a
similar event occurring in around 2034 could result in
~535MW of PV reduction over 60s. As for the South
Australian case, this is approaching the realm of typical
contingency sizes (600-750MW), particularly if larger
units are not operating, or operating at lower levels
during low load periods.
Table X – Victorian estimated contingency sizes due to PV behavior
during disturbances, current and forecast
Generation
deficit
2018
(event)
2020 2025 2030 2035
60s after
event
177 236 372 479 550
5min after
event
66 88 139 180 206
Note: estimates based on upscaled profile. For instance, 60s estimate
shows the difference between pre-event upscaled PV generation and
PV generation 60s later. Forecasts scale based on projected PV
capacity.
5.11 Verification using independent data set
An independent and publicly available data set was
accessed from PVOutput.org and compared against the
Solar Analytics data. The PVOutput.org data set is only
available at 5min or greater time increments, and as
result the PV response following the disturbance is
muted compared with that of the 30s data analyzed in
this paper. PV system capacity factor (ratio of generation
to DC capacity) is used as the basis for comparison.
Fig. 24 and Fig. 26 show the aggregate performance
observed in both data sets in South Australia and Victoria
respectively. Whilst Fig. 25(a) and Fig. 27(a) show the
event period for both regions.
The correlation analysis shown in Fig. 25(b) and Fig. 27(b)
indicates a close match between the PVOutput.org and
Solar Analytics data set as indicated by large r-squared
values. The residual (Fig. 27(b)) show that the there is a
consistent difference between the data sets for both the
South Australian and Victorian event in the second time
interval immediately following the event. This is likely
due to the higher proportion of pre 2015 systems in the
PVOutput.org data set, compared with the Solar
Analytics data set and new ramp rate requirements for
post 2016 systems (Appendix A).
Fig. 24. (a) South Australia comparison of PVOutput.org data set (268
sites) and Solar Analytics data set (260 sites), time is AEST
(a) (b)
Fig. 25. (a) South Australia comparison during event period of
PVOutput.org data set (268 sites) and Solar Analytics data set (260
sites), time is AEST, (b) correlation and residuals
Fig. 26. Victoria comparison of PVOutput.org data set (400 sites) and
Solar Analytics data set (168 sites), time is AEST
(a) (b)
Fig. 27. (a) Victoria comparison during event period of PVOutput.org
data set (400 sites) and Solar Analytics data set (168 sites), time is
AEST, (b) correlation and residuals
6
66
6 Conclusions
ConclusionsConclusions
Conclusions
As distributed PV penetrations continue to grow, power
system operators world wide will need to manage the
very real system security challenges posed by mass
response of distributed inverters to disturbances
originating in the transmission system. Whilst early work
focused on frequency disturbances, there is a growing
Possibly due to
AS4777 version
Possibly due to
AS4777 version

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
15
awareness that large voltage disturbances also pose a
significant risk to secure power system operation.
This study presents analysis of two case studies
examining major power system voltage disturbances.
Upscaling for the more extreme South Australian event
showed distributed PV generation reduced by 45%
following a major voltage disturbance, constituting
approximately 10% of regional demand at the time of the
event. If a similar event occurred in 2035 and the
projected level of distributed PV installation were to be
realized, then the PV response could represent
approximately 29% of the total demand or ~536MW,
with larger fractions possible if the event occurred at
midday or during a period of low demand. It is of course
possible to envisage more severe voltage disturbances
that would involve even greater distributed PV power
loss.
Clearly, this poses a significant security challenge and it is
critical that power system planning and operation takes
into account distributed PV in regions of high
penetration. Critically, this study suggests that it may be
appropriate to consider distributed PV response to
disturbances when setting FCAS requirements.
Given that distributed PV is not currently considered
when settings FCAS contingency requirements in the
NEM, this represents a significant operational change for
operators.
This study emphasizes the need for analysis of actual
operational data in order to capture legacy issues (for
instance systems installed under superseded connection
standards), the diversity of installed inverter models, and
the complexity of events within the low voltage network.
For example, both case studies presented here consider
voltage disturbances in which severe undervoltage was
observed in the transmission network. However, both
events resulted in substantial loss of load and as result,
the observed local voltages show both under and over
voltage, suggesting localized over voltage due to load
‘shake off’. Modelling such interactions is challenging,
and in this context, operational PV data is likely to
provide an important opportunity to test and verify
composite load models developed to capture these
impacts.
The study presents four novel techniques for analyzing
operational distributed PV data. Two techniques are
presented for analyzing individual responses and it is
shown that there is considerable diversity in both depth
and duration of response to the events. Two further
techniques are presented for analyzing aggregate
response, where the first sets out an approach to spatial
analysis and the second upscales observed behaviors
from monitored PV systems to the broader installed PV
capacity. The spatial analysis shows clear trends in
severity of response with systems closer to the source of
disturbance more likely to be affected. The upscaling
technique is a powerful tool for assessing fleet behavior,
however is shown to overestimate PV response
compared with the overall change in state load. This may
be due to a number of factors, including the potential for
load ‘shake off’ to mask PV response.
Importantly, it is not possible to establish whether the
observed PV disconnections were caused by over
voltage, under voltage, vector shift or other inverter
responses by examining this data set and further, it is not
possible to directly assess compliance with the inverter
connection standards. To better understand the cause of
inverter behaviors and assess compliance, in lab testing
is required. Powerful further insights may be possible
through coupling operational data, system modelling and
lab testing.
Acknowledgment
AcknowledgmentAcknowledgment
Acknowledgment
The authors gratefully acknowledge support provided for
this research by an Australian Government Research
Training Program Scholarship, the Faculty of Engineering
at UNSW, CRC LCL project RP1023u1 and ARENA project
‘UNSW, Addressing Barriers to Efficient Renewable
Integration’.
The authors gratefully acknowledge the contributions of
R. Bunder, R. Egan and J. Dore of Solar Analytics for the
provision of data. The authors also gratefully
acknowledge the contributions of N. Gorman for support
accessing data, T. Barton for data cleaning, and A.
Millican for discussion throughout analysis.
Declarations of interest
Declarations of interestDeclarations of interest
Declarations of interest
None.

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
16
Appendix A
Appendix AAppendix A
Appendix A
Table XI – AS4777 Passive Anti-islanding requirement summary, current and legacy standards, nominal voltage is 230V [56, 57]
Protective
function
Protective
function limit
Trip
delay
time
Maximum
disconnection time
Standard
Minimum time before
reconnection
1
Reconnection ramp
rate
1
Under voltage 200 – 230V
0.87 – 1p.u.
Not
specified
2s AS4777.3-2005
(superseded)
60s
Not specified
Over voltage 230 – 270V
1 – 1.17p.u.
Not
specified
2s AS4777.3-2005
(superseded)
60s
Under voltage 180V
0.78p.u.
1s 2s AS4777.2-2015
(current)
60s
16.67% of rated power
per minute (nominal
ramp time of 6min)
Over voltage 1 260V
1.13p.u.
1s 2s AS4777.2-2015
(current)
60s
Over voltage 2 265V
1.15p.u.
- 0.2s AS4777.2-2015
(current)
60s
1
When limits are exceeded the disconnection device operates, with a minimum 1min before reconnecting, with ramp rate at reconnection as
specified.
References
ReferencesReferences
References
[1] R. Passey, T. Spooner, I. MacGill, M. Watt, and K.
Syngellakis, "The potential impacts of grid-
connected distributed generation and how to
address them: A review of technical and non-
technical factors," Energy Policy, vol. 39, no. 10,
pp. 6280-6290, October 2011.
[2] International Energy Agency, "Do It Locally: Local
Voltage Support by Distributed Generation – A
Management Summary," in "Management
Summary of IEA Task 14 Subtask 2 –
Recommendations Based on Research and Field
Experience," January 2017.
[3] H. Ji et al., "A centralized-based method to
determine the local voltage control strategies of
distributed generator operation in active
distribution networks," Applied Energy, vol. 228,
pp. 2024-2036, 2018/10/15/ 2018.
[4] H. Ji et al., "An enhanced SOCP-based method for
feeder load balancing using the multi-terminal soft
open point in active distribution networks,"
Applied Energy, vol. 208, pp. 986-995,
2017/12/15/ 2017.
[5] P. Zhang, W. Li, S. Li, Y. Wang, and W. Xiao,
"Reliability assessment of photovoltaic power
systems: Review of current status and future
perspectives," Applied Energy, vol. 104, pp. 822-
833, 2013/04/01/ 2013.
[6] M. N. Kabir, Y. Mishra, G. Ledwich, Z. Xu, and R. C.
Bansal, "Improving voltage profile of residential
distribution systems using rooftop PVs and Battery
Energy Storage systems," Applied Energy, vol. 134,
pp. 290-300, 2014/12/01/ 2014.
[7] J. Susanto, F. Shahnia, and D. Ludwig, "A
framework to technically evaluate integration of
utility-scale photovoltaic plants to weak power
distribution systems," Applied Energy, vol. 231, pp.
207-221, 2018/12/01/ 2018.
[8] A. Cabrera-Tobar, E. Bullich-Massagué, M.
Aragüés-Peñalba, and O. Gomis-Bellmunt, "Review
of advanced grid requirements for the integration
of large scale photovoltaic power plants in the
transmission system," Renewable and Sustainable
Energy Reviews, vol. 62, pp. 971-987, 2016/09/01/
2016.
[9] A. Purvins, A. Zubaryeva, M. Llorente, E. Tzimas,
and A. Mercier, "Challenges and options for a large
wind power uptake by the European electricity
system," Applied Energy, vol. 88, no. 5, pp. 1461-
1469, 2011/05/01/ 2011.
[10] CIGRE C2.16, "Challenge in the Control Centre
(EMS) due to Distributed Generation and
Renewables," September 2017, Available:
https://e-cigre.org/publication/700-challenges-in-
the-control-center-ems-due-to-distributed-
generation-and-renewables.
[11] D. Lew et al., "The Power of Small: The Effects of
Distributed Energy Resources on System
Reliability," IEEE Power and Energy Magazine, vol.
15, no. 6, pp. 50-60, 2017.
[12] NERC, "900 MW Fault Induced Solar Photovoltaic
Resource Interruption Disturbance Report,"
February 2018.
[13] R. Yan, T. K. Saha, N. Modi, N.-A. Masood, and M.
Mosadeghy, "The combined effects of high
penetration of wind and PV on power system
frequency response," Applied Energy, vol. 145, pp.
320-330, 2015/05/01/ 2015.
[14] C. Ching, "Adapting to Change: Hawai'i's Response
to Alternative Energy Sources [In My View]," IEEE
Power and Energy Magazine, vol. 15, no. 2, pp. 93-
96, 2017.

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
17
[15] M. Hancock and e. al, "Alice Springs: A Case Study
of Increasing Levels of PV Penetration in an
Electricity Supply System," 2011.
[16] A. Khoshnami and I. Sadeghkhani, "Two-stage
power–based fault detection scheme for
photovoltaic systems," Solar Energy, vol. 176, pp.
10-21, 2018/12/01/ 2018.
[17] A. Q. Al-Shetwi, M. Z. Sujod, and F. Blaabjerg, "Low
voltage ride-through capability control for single-
stage inverter-based grid-connected photovoltaic
power plant," Solar Energy, vol. 159, pp. 665-681,
2018/01/01/ 2018.
[18] I. I. Perpinias, N. P. Papanikolaou, and E. C. Tatakis,
"Fault ride through concept in low voltage
distributed photovoltaic generators for various
dispersion and penetration scenarios," Sustainable
Energy Technologies and Assessments, vol. 12, pp.
15-25, 2015/12/01/ 2015.
[19] Jens C. Boemer, Mohammad Huque, Brian Seal,
Tom Key, Daniel Brooks, and Charlie Vartanian,
"Status of Revision of IEEE Std 1547 and 1547.1,"
presented at the Power & Energy Society General
Meeting, 2017 IEEE, Chicago, IL, USA, 16-20 July,
2017.
[20] N. Schäfer, G. Arnold, and W. Heckmann, "Grid
code compliance testing of renewables – New
requirements and testing experiences," presented
at the 1st International Conference on Large-Scale
Grid Integration of Renewable Energy in India,
New Delhi, 6-8 September, 2017.
[21] "IEEE Standard for Interconnection and
Interoperability of Distributed Energy Resources
with Associated Electric Power Systems
Interfaces," IEEE Std 1547-2018 (Revision of IEEE
Std 1547-2003), pp. 1-138, 2018.
[22] Y. Yang, P. Enjeti, F. Blaabjerg, and H. Wang,
"Suggested grid code modifications to ensure
wide-scale adoption of photovoltaic energy in
distributed power generation systems," in 2013
IEEE Industry Applications Society Annual Meeting,
2013, pp. 1-8.
[23] A. Hoke et al., "Setting the Smart Solar Standard:
Collaborations Between Hawaiian Electric and the
National Renewable Energy Laboratory," IEEE
Power and Energy Magazine, vol. 16, no. 6, pp. 18-
29, 2018.
[24] G. Ghatikar, S. Mashayekh, M. Stadler, R. Yin, and
Z. Liu, "Distributed energy systems integration and
demand optimization for autonomous operations
and electric grid transactions," Applied Energy, vol.
167, pp. 432-448, 2016/04/01/ 2016.
[25] A. Nelson et al., "Power Hardware-in-the-Loop
Evaluation of PV Inverter Grid Support on Hawaiin
Electric Feeders," presented at the Power & Energy
Society Innovative Smart Grid Technologies
Conference, 2017 IEEE, Washington, DC, USA, 30
October, 2017. Available:
http://ieeexplore.ieee.org/document/8085982/
[26] G. B. Huka, W. Li, P. Chao, and S. Peng, "A
comprehensive LVRT strategy of two-stage
photovoltaic systems under balanced and
unbalanced faults," International Journal of
Electrical Power & Energy Systems, vol. 103, pp.
288-301, 2018/12/01/ 2018.
[27] North American Electric Reliability Corporation
(NERC), "Technical Reference Document Dynamic
Load Modeling," December 2016, Available:
https://www.nerc.com/comm/PC/LoadModelingT
askForceDL/Dynamic%20Load%20Modeling%20T
ech%20Ref%202016-11-14%20-
%20FINAL.PDF#search=Load%20modeling%20tec
hnical%20reference.
[28] A. Maitra, A. Gaikwad, P. Pourbeik, and D. Brooks,
"Load model parameter derivation using an
automated algorithm and measured data,"
presented at the 2008 IEEE Power and Energy
Society General Meeting - Conversion and Delivery
of Electrical Energy in the 21st Century, Pittsburgh,
PA, USA, 20-24 July, 2008. Available:
https://ieeexplore.ieee.org/document/4596333/
[29] P. Pourbeik et al., "Proposal for DER_A Model,"
2018, Available:
https://www.wecc.biz/Administrative/DER_A_Fin
al_021518.pdf.
[30] International Energy Agency, "Status of Power
System Transformation 2017 System integration
and local grids," 2017, Available:
https://www.iea.org/publications/freepublication
s/publication/status-of-power-system-
transformation-2017.html.
[31] APVI. (2019, 26 April). Australian PV market since
April 2001. Available: http://pv-
map.apvi.org.au/analyses
[32] APVI. (2018). Australian PV Institute (APVI) Solar
Map, funded by the Australian Renewable Energy
Agency. Available: http://pv-map.apvi.org.au/
[33] AEMO, "Visibility of Distributed Energy
Resources," in "Future Power System Security
Program," January 2017, Available:
https://www.aemo.com.au/-
/media/Files/Electricity/NEM/Security_and_Relia
bility/Reports/2016/AEMO-FPSS-program----
Visibility-of-DER.pdf.
[34] AEMO, "Response of Existing PV Inverters to
Frequency Disturbances," 27 April 2016, Available:
https://www.aemo.com.au/Media-
Centre/Response-of-Existing-PV-Inverters-to-
Frequency-Disturbances.
[35] N. Haghdadi, J. Dennis, A. Bruce, and I. MacGill,
"Real Time Generation Mapping of Distributed PV

Preprint submitted to Applied Energy. Final published version: https://www.sciencedirect.com/science/article/pii/S0306261919319701
18
for Network Planning and Operations," presented
at the IEEE PES Asia-Pacific Power and Energy
Engineering Conference, Brisbane, Australia, 2015.
[36] Solar Analytics. (2018). Solar Analytics. Available:
https://www.solaranalytics.com/au/
[37] PVoutput.org. (2018). Live photovoltaic data.
Available: PVoutput.org
[38] N. Haghdadi, A. Bruce, I. MacGill, and R. Passey,
"Impact of Distributed Photovoltaic Systems on
Zone Substation Peak Demand," IEEE Transactions
on Sustainable Energy, vol. 9, no. 2, pp. 621-629,
April 2018.
[39] Clean Energy Regulator, "Postcode data for small-
scale installation - all data," ed, 2018.
[40] Australian Bureau of Statistics, "Postal Areas ASGS
Ed 2016 Digital Boundaries in ESRI Shapefile
Format (zip file)," in 1270.0.55.003 - Australian
Statistical Geography Standard (ASGS): Volume 3 -
Non ABS Structures, July 2016, ABS, Ed., ed, 2016.
[41] Australian Bureau of Statistics, "Greater Capital
City Statistical Area (GCCSA) ASGS Ed 2011 Digital
Boundaries in ESRI Shapefile Format ", ed, 2011.
[42] AEMO, "NEMweb," ed, 2018.
[43] N. Gorman, N. Haghdadi, A. Bruce, and I. MacGill,
"NEMOSIS – NEM Open Source Information
Service; open-source access to Australian National
Electricity Market Data," presented at the Asia
Pacific Solar Research Conference 2018, Sydney,
Australia, 4-6 Decmber, 2018. Available:
http://apvi.org.au/solar-research-conference/wp-
content/uploads/2018/12/194_D-
I_Gorman_N_2018.pdf
[44] G. Parkinson, "Rooftop solar provides 48% of South
Australia power, pushing grid demand to record
low," in RenewEconomy, ed, 2017.
[45] Energy Transition Hub. (2018, 13 February 2018).
OpenNEM South Australia. Available:
http://opennem.org.au/#/regions/sa
[46] AEMO, "Electricity Forecasting Insights – March
2018 update," in "Electricity Minimum Demand
Operational," 29 March 2018, Available:
http://forecasting.aemo.com.au/Electricity/Mini
mumDemand/Operational.
[47] Walter Gerardi and Damien O’Connor,
"Projections of uptake of small-scale systems,"
Jacobs9 June 2017.
[48] AER, "State of the Energy Market," 30 May 2017.
[49] AER and AEMO, "Seasonal peak demand (region),"
D11/2285335[V2] ed, 2018.
[50] AER and AEMO, "Registered capacity in regions by
fuel source," D11/2285335[V2] ed, 2018.
[51] AEMO, "Rooftop PV and battery storage," 2017,
Available:
https://www.aemo.com.au/Electricity/National-
Electricity-Market-NEM/Planning-and-
forecasting/Electricity-Forecasting-Insights/Key-
component-consumption-forecasts/PV-and-
storage.
[52] Bureau of Meteorology, "Daily maximum
temperature," 2018.
[53] A. O'Neil, "Was South Australia lucky the lights
stayed on last Friday?," in WattClarity, ed:
GLOBAL-ROAM, 2017.
[54] AEMO, "Fault at Torrens Island Switchyard and loss
of multiple generating units on 3 March 2017," in
"Reviewable Operating Incident Report for the
National Electricity Market," 10 March 2017,
Available: https://www.aemo.com.au/-
/media/Files/Electricity/NEM/Market_Notices_an
d_Events/Power_System_Incident_Reports/2017/
Report-SA-on-3-March-2017.pdf.
[55] (2018). Update - Non-credible contingency event -
VIC region - Thursday, 18 January 2018.
[56] AS/NZS 4777.2:2015 Grid connection of energy
systems via inverters, 2015.
[57] AS/NZS 4777.3-2005 : Grid connection of energy
systems via inverters - Grid protection
requirements, 2005.
[58] AEMO, "National Transmission Network
Development Plan," 12 December 2016, Available:
https://www.aemo.com.au/-
/media/Files/Electricity/NEM/Planning_and_Fore
casting/NTNDP/2016/Dec/2016-NATIONAL-
TRANSMISSION-NETWORK-DEVELOPMENT-
PLAN.pdf.
[59] J. Riesz, J. Gilmore, and I. MacGill, "Frequency
Control Ancillary Service Market Design: Insights
from the Australian National Electricity Market,"
The Electricity Journal, vol. 28, no. 3, pp. 86-99, 1
April 2015.
[60] N. Haghdadi, A. Bruce, and I. MacGill, "Assessing
the Representativeness of 'Live' Distributed PV
Data for Upscaled PV Generation Estimates,"
presented at the IEEE PES Asia-Pacific Power and
Energy Engineering Conference, Brisbane,
Australia, 2015.
[61] AEMO, "Constraint Formulation Guidelines," 5
December 2013, Available:
https://www.aemo.com.au/-
/media/Files/Electricity/NEM/Security_and_Relia
bility/Congestion-
Information/2016/Constraint_Formulation_Guide
lines_v10_1.pdf.