Content uploaded by Kanyawee Keeratimahat
Author content
All content in this area was uploaded by Kanyawee Keeratimahat on Nov 30, 2020
Content may be subject to copyright.
1
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Analysis of short-term operational forecast deviations and
controllability of utility-scale photovoltaic plants
Kanyawee Keeratimahata,c,*, Anna Brucea,c, Iain MacGillb,c
a School of Photovoltaics and Renewable Energy Engineering, The University of New South Wales
(UNSW), Sydney, Australia
b School of Electrical Engineering and Telecommunication, The University of New South Wales
(UNSW), Sydney, Australia
c Centre for Energy and Environmental Markets, The University of New South Wales (UNSW),
Sydney, Australia
* Corresponding Author
email: k.keeratimahat@unsw.edu.au
1. Introduction
The falling costs and environmental benefits of utility-scale photovoltaic (PV) plants are driving
growing deployment in electricity industries around the world. However, their highly variable and
somewhat unpredictable output raises some challenges for electricity industry operation. A particular
challenge is short term frequency management. Deviations of PV generation from forecasts will
contribute to frequency deviations that then need to be corrected [1]. In many electricity industries
such short-term balancing is primarily managed through frequency control ancillary services (FCAS).
While arrangements vary across jurisdictions, these are often divided into regulation or load following
services to address inherent ongoing load variation, and contingency services to address infrequent
major supply-demand balance disturbances such as large generating plant or network failures. Utility
PV plants have particular operating characteristics that would seem to create both challenges and
opportunities for frequency management. The output of even large utility PV plants which have
inherent spatial smoothing, can vary greatly in a matter of seconds under particular cloud patterns,
with limited predictability; while these plants have effectively no energy storage to smooth output,
unless battery energy storage is deployed. On the other hand, their power inverter interface typically
does mean the output of PV plants can be rapidly and finely controlled between zero and their
maximum achievable output depending on available solar irradiance.
A range of strategies are now being deployed to improve the integration of variable renewable energy
(VRE) generation such as PV into short-term electricity industry operation. Implementation of many
of these requires characterising solar resource and PV generation variability (change in output from
one time interval to another) and uncertainty (unpredictability of the output). [2]–[5] characterise PV
variability at 1-second and [6] at 1 minute with comparison to point-source measurement and plant
output. All these studies quantify variability in terms of statistical distribution which is useful for
quantifying frequency regulation reserve requirements [7]. A smaller number of studies have assessed
PV variability compared to the balancing capabilities in the power system. For example, [8] compared
PV variability with the ramping capability of a gas turbine unit while [9] modelled the impact of PV
siting in Ontario on aggregated PV output variability at hourly resolution and observed its correlation
to the system-wide demand. The uncertainty of PV generation has been characterised in the form of
2
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
deviation from a reference line which could be a forecast target [10] or a line of moving average [11],
[12] at a timeframe chosen to reflect the frequency control services provided in that system. These
studies quantified and compared the aggregated deviation to the variability of load to assess the
requirement for additional FCAS services, which are traditionally based on load variability. [10] also
investigated different short-term market operational strategies to reduce reserve requirements due
to increasing PV penetration. Nonetheless, understanding of the characteristics of short-term PV
operation relevant to the power system operator and required to improve integration strategies for
VRE is still limited, particularly short-term deviations from the generation forecasts used in generator
dispatch. More generally, as discussed in [4], there is a need to better understand how this variability
compares with that of other generation technologies, given that all generators typically exhibit some
uncertainty. Furthermore, there is the question of what the impacts of this uncertainty around
dispatch targets (or for wind and PV short-term forecasts) are on power system FCAS requirements.
The degree of uncertainty around the expected output of PV plants would also be useful to indicate
volatile periods [13]. Note that our interest is not specifically in the accuracy of such forecasts although
improving forecast accuracy could significantly reduce the need for frequency control reserves [14],
[15]. Instead, we seek to better understand the short-term variability and uncertainty that might be
expected around these forecasts, and its potential impacts on frequency control requirements.
In addition to variability characterisation, there is the question of how power system operational
arrangements might be improved, including greater participation by VRE generation. A number of
jurisdictions with significant wind generation have for some time been looking to require such plants
to be able to provide regulation services [16]. Key barriers to entry to the FCAS markets for VRE
generation have been identified in [17] for the Spanish market. [18], [19] developed strategies for VRE
to participate in FCAS markets, based on a range of factors including the resource availability, system
security constraints and market price. These studies were based on the assumption that PV inverters
had a great capability in controlling PV output. However, even though the necessary inverter
capabilities to provide advanced control for utility PV plant are well established [20], [21], there is
limited evidence to date of their actual application in real power system operation.
The Australian National Electricity Market (NEM) presents an interesting opportunity for investigating
issues of PV variability characterisation, and both its implications for managing frequency under
existing FCAS arrangements as well as opportunities to contribute such services, because the NEM
includes utility-scale PV as a market participant and the market operator publishes high resolution
SCADA data from every participating plant. The NEM has had four utility PV plants operating for over
four years with an aggregate capacity of 231 MW in the region of New South Wales. At the end of
2019, the NEM-wide installed capacity of utility-scale PV plants was 2.8 GW and many more are now
coming on-line [22]. Utility wind and PV generators are required to participate in wholesale market
arrangements as “semi-scheduled” units that must follow “dispatch” targets when required but can
otherwise generate up to their maximum level. They are required to contribute to the cost of FCAS
regulation (through “Causer Pays” arrangements) according to how their variability within five minute
dispatch periods contributes to short-term supply-demand imbalance, while also being allowed to
provide FCAS services (although to date, the only participation of this type has been through trials
where wind farms have participated in both regulation and contingency, raise and lower, FCAS
markets [23]). The NEM is facing growing challenges in frequency control with questions of whether
3
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
growing renewable penetrations or declining primary frequency control from scheduled plants might
be contributing [24].
Of particular importance to the study reported in this paper, the Causer Pays arrangements for
regulation FCAS are based around the ability, or failure, of generators to follow their target trajectories
within 5-minute dispatch periods. Hence, 4-second SCADA data of the generation of every scheduled
and semi-scheduled generator in the NEM is collected and made publicly available, as well as all details
of wholesale market dispatch targets and FCAS “enabled” market participants. This high-resolution
operational data permits analysis in this paper of the generation variability of a number of
geographically dispersed utility PV plants with 4-second sampling. As such, the impacts of PV
generators on short-term supply-demand balance and FCAS market operation, and the potential
implications for these generators can be assessed. Also, the requirement for these plants to meet
“upper bound” dispatch targets in some circumstances provides some evidence of the potential
controllability of utility PV generation. Importantly, the high data transparency of the NEM also allows
us to compare the potential controllability of utility PV against other NEM generation technologies
including utility wind as well as gas and coal-fired units and hydro generation.
The novelty of this paper is summarised as follows. Most existing studies have analysed PV variability
independently of electricity market operation, or in the absence of high-resolution generation data,
have relied on modelled PV generation. Our paper is the first as far as we are aware to analyse real-
world utility PV plant deviation around short-term forecast dispatch targets compared to the
deviations of all other participating generators, the potential implications of this for frequency control,
and the potential controllability of these plants to assist in managing their deviations. Uncertainty
analysis across all generator types contributes to the understanding of the scale of PV variability
compared to dispatched generation uncertainty in large power systems, and its influence on
frequency deviation and therefore frequency control requirements. Observation of PV behaviour
under constraint provides a greater understanding of the current controllability of PV plants in the
existing grid operation and therefore their potential to provide frequency control services. A 4-second
timeframe is commonly used for automatic generation control (AGC) in power system operation. The
results from this analysis therefore have direct implications for frequency regulation and short-term
balancing within the electricity market dispatch timeframes of a range of power systems as PV
penetrations increase. There is very limited work to date internationally that has used such detailed
operational and market data. While individual plant variability and deviation observed will be context
specific, the insights of this analysis, therefore, have broader electricity industry relevance for other
jurisdictions.
The structure of the rest of the paper is as follows. NEM generator classification and generation
capacity are outlined in Section 2. The data used for this study and the methods for analysis of
uncertainty of generators in meeting their dispatch target trajectories and their controllability when
under instruction are explained in Section 3. Section 4 presents an overview of SCADA output profiles
from the main generation technologies exhibited in the NEM. Section 5 analyses forecast deviations
of wind and PV plants while Section 6 and Section 7 carry out the analysis of MW deviations of all
market participants and their relationship to frequency deviation. Section 8 investigates the
controllability of the PV plants. Lastly, Section 9 presents the implications of the observed short-term
operation of PV for electricity market design.
4
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
2. Australian NEM context
The Australian Energy Market Operator (AEMO) categorises generators in the NEM according to Figure
1. Scheduled generators include thermal power plants, hydro and battery energy storage while semi-
scheduled generators are wind and solar farms. Both of these participant types are allowed to provide
frequency regulation services.
Figure 1 Participant categories in the NEM [25].
The installed NEM generation capacity by type of registration and generation technology (as of 2019)
are listed in Table 1 and Table 2, respectively. The main providers for regulation services are at present
a subset of the steam sub-critical coal, hydro, and a few CCGT and OCGT plants, although utility battery
storage is also now playing a greater role. Utility wind and solar do not currently offer regulation
services although they can, and have done so in trials [23]. Background information on the energy
market structure and arrangement of the NEM is provided as supplementary material.
Table 1 Number of generators, installed capacity and the percentage of total NEM generation capacity categorised by
participant types (as of 2019)
Participants (as of 2019)
Number of
plants
Capacity
(MW)
Capacity (% of
Total)
Non-scheduled
127
2,705
5.2%
Semi-scheduled
76
7,163
13.7%
Semi-scheduled (providing regulation in several
trials)
3
317
0.6%
Scheduled (not providing regulation)
83
9,193
17.6%
Scheduled (providing regulation)
96
32,744
62.8%
Total
385
52,123
100.0%
Generators
Scheduled
Non-scheduled
Semi-scheduled
Regulation
No regulation
Dispatch market
Regulation market
(FCAS)
5
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Table 2 Number of generators, installed capacity and the percentage of total NEM generation capacity categorised by
generation technology (as of 2019)
Technology
Number of
plants
Capacity
(MW)
Capacity (% of
Total)
Steam Sub-Critical
58
22,155
42.5%
Hydro - Gravity
57
7,155
13.7%
Open Cycle Gas turbines (OCGT)
66
6,575
12.6%
Wind
69
6,072
11.6%
Combined Cycle Gas Turbine (CCGT)
12
3,082
5.9%
Steam Super Critical
6
2,879
5.5%
Photovoltaic
42
2,726
5.2%
Compression Reciprocating Engine and Spark
ignition
56
609
1.2%
Hydro - pump storage
3
510
1.0%
Battery
4
193
0.4%
Hydro - Run of River
12
168
0.3%
Total
385
52,123
100.0%
3. Data and method
3.1 SCADA generation data
The instantaneous power output at 4-second intervals of all 229 scheduled and semi-scheduled
generators (as of 2016 to consider a full year of data) in the NEM was extracted from the AEMO
Ancillary Services Market Causer Pays Data which is publicly available on the AEMO website [26]. For
each generator, for each 5-minute dispatch interval, the dataset contains 4-second SCADA output, 5-
minute dispatch target, measured system frequency deviation from the nominal value (usually 50 Hz),
and assigned 4-second MW for frequency regulation, which is the increase or decrease in output that
AEMO requires from generators participating in regulation FCAS.
This study assesses the short-term variability and control of all scheduled and semi-scheduled
generators in the NEM as of 2016, particularly the performance of the three utility-scale PV plants that
were operating in 2016, and hence for which there is more than a full year of SCADA data. A summary
of the utility PV plants, their DC and AC rating, and technology is presented in Table 3.
Table 3. PV plants that were operating in 2016 and are classified as semi-scheduled generators (> 30MW)
Plant
DC Capacity (MW)
AC Capacity (MW)
DC:AC ratio
Data available from
Technology
Nyngan,
NSW
132.4
102
1.29
May, 2015
Fixed-tilt
Moree,
NSW
69.5
56
1.25
February, 2016
Single-axis
tracking
Broken
Hill, NSW
70
53
1.31
August, 2015
Fixed-tilt
6
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
All three PV plants have a DC:AC ratio higher than 1 which means that during peak solar irradiance
periods, the plant generation will be constrained down to the limit of the AC “inverter” capacity, or
potentially network or market output constraints. This results in a flat generation profile over such
periods, reducing their variability [27] while also providing an example of the output control that can
be achieved with these plants, as we explore further in Section 8.
3.2 Assessing deviations around plant targets
Our analysis of variability and controllability is based around deviations from generation plant target
trajectories. At present, in the NEM, target trajectories for a semi-scheduled plant are calculated in
the same way as for scheduled plant. As discussed below, limitations in current arrangements for
semi-scheduled plants increase their assessed deviations from target. We, therefore, propose a simple
adjustment to how the target trajectory is set that better reflects semi-scheduled plant ramping
characteristics. It is notable that there are current rule change processes underway that seek to
address this issue as well.
3.2.1 Target trajectory calculation for scheduled plant
According to the AEMO Dispatch procedure [28], the linear trajectory for scheduled plant can be
calculated by using a linear fit between the 5-minute dispatch targets of the previous timestamp and
the current timestamp (as illustrated in Figure 2). The 4-second targets can then be calculated by
substituting the number of data points within that 5-minute period into the linear fit equation for that
interval.
Figure 2 Illustration of the actual generation, dispatch target and the calculated linear trajectory according to AEMO’s
dispatch procedure
The linear equation for each dispatch interval can be presented as (1)
(1)
Where is the 4-second target (MW) in the interval t, is the slope of the ramp, is the data
point of the set of 4-second targets in that 5-minute interval and is the constant coefficient in the
linear equation. and are calculated by (2).
and (2)
Where is the dispatch target of the previous interval , is the dispatch target of the current
interval and is the number of data points contained in that interval. Theoretically, there should be
7
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
75 4-second data points in a 5-minute interval. However, the number of data points can vary slightly
as the dispatch target does not change exactly at the beginning of each 5-minute interval. The MW
deviation from the linear trajectory can then be calculated by (3).
(3)
With this approach, MW deviations for each scheduled generator can be quantified. The summary of
the process and the related result sections is shown in Figure 3.
Figure 3 Methodology diagram for uncertainty analysis (MW deviations) and the related result sections.
3.2.2 Target trajectory calculation for semi-scheduled plant and a proposed
modification
At present, AEMO sets 5-minute dispatch targets for VRE (semi-scheduled) generators in the NEM,
based on forecasts from the Australian Solar Energy Forecasting System (ASEFS), rather than from the
5-minute market dispatch process. The forecast for the beginning of each dispatch interval is used to
set the target for that interval, which must be met at the end of the 5 minutes, and the target
trajectory is a linear fit between the dispatch targets for the current and previous intervals which is
similar to the linear trajectory for scheduled generators (Figure 2). These arrangements create a bias
deviation in the plant output compared to the 4-second trajectory targets because the plant output is
forecast to already be at the target at the beginning of the 5 minutes, and by the end of the 5 minutes
will be higher in the morning or lower in the afternoon due to changes in solar irradiation throughout
8
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
the day. Modified persistence forecast based on the active power output measured at 2 minutes prior
to the current dispatch period is used to set the dispatch target, meaning that the target for the end
of the 5-minute dispatch period is equal to the plant’s MW output 2 minutes prior to the current time,
effectively a 7-minute delayed forecast.
We propose a very simple modification as illustrated in Figure 4. This modification does not suggest
any change to the ASEFS forecasting method but rather a change to the way that wind and solar PV
are accounted for in AEMO’s energy dispatch calculation. Instead of expecting wind and solar PV plants
to meet their targets at the end of the dispatch interval, the targets can be met at the beginning of
each interval. Hence the slope in (2) is changed to defined in (4).
and (4)
Where is the dispatch target of the next interval . The rest of the parameters remain the
same (2). A conceptual comparison between the two linear ramping calculations is shown in Figure 4.
Figure 4 Illustration of the actual generation, dispatch target and linear trajectory with the proposed procedure
The impact of the present AEMO approach and our proposed modification is shown in Figure 5. The
near continuous deviation of Nyngan PV plant from its AEMO “target” in the morning and afternoon
on a clear day can be addressed through our proposed modification. In later analysis we generally use
our proposed approach given that this very simple modification can greatly reduce the perceived
deviations of utility PV and wind against their targets.
3.3 Assessing the controllability of utility PV
To date of the study, PV plants in the NEM are not required to control their output to meet their
dispatch targets and are not providing frequency regulation. The regular controls that the PV plant
would experience in the field are AC inverter limit and response to network constraints. Two aspects
of the controllability of utility PV are, therefore, assessed. The first is the ability of these plants to
produce a steady output when they are limited by inverter capacity despite inevitable solar variation
in typical periods of operation. The second is the controllability these plants show when required by
AEMO’s dispatch instruction to constrain their output at or below a given target. These analyses
assess how accurately a PV plant could respond to market instructions and, therefore, highlight
potential participation in frequency regulation. The overview of the steps is shown in Figure 6 while
the detailed descriptions are explained below.
a. Classification of flat peak generation
In order to assess the controllability of existing utility PV, we selected periods of peak generation
where curtailment in power output occurs due to inverter capacity or network connection limits.
9
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(a) (b)
Figure 5 Example of generation profile on a clear sky day at Nyngan a) with the original linear trajectory and b) with the
proposed linear trajectory correction
During these times, PV is operated to produce a steady output over a period of time, whereas at other
times, plant operators are typically making no effort to control weather driven output variation. The
selection of peak generation periods for assessment was carried out as follows:
1) The dataset was pre-filtered to avoid processing large amounts of 4-second data. The days
with power production meeting 95% of the plant rated capacity for at least one hour in total
were selected to ensure that there is a long period of peak generation in that day.
2) A moving linear fit over 15 minutes period was applied to the data throughout the selected
day. The slope of the linear fit was be used to classify whether each period has a flat
generation profile over the 15-minutes. Additionally, a moving average over 15 minutes was
calculated in order to classify the peak generation.
3) The selected intervals have a slope coefficient of no more than 0.001 with average generation
more than or equal to 97% of the plant capacity. Note that the raw 4-second SCADA output
data were collected with up to 2 decimal places. Hence, a maximum of 0.001 will ensure a flat
profile. For curtailment due to overgeneration to occur, the average generation should reach
the rated capacity. The 97% generation is selected instead of 100% to leave some margin for
the discrepancies between the curtailed generation level from inverter to inverter.
10
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
4) Finally, to assess how tightly the generation is controlled, the fluctuations within the selected
intervals are calculated as the difference between 4-second output and the 15-minute moving
average value.
Following this method, uncontrolled variables such as weather variability are excluded from the
analysis and the assessment of output deviations from target will only be based on the technical
capability of the PV plants. The results of this analysis are shown in Section 8.1.
b. Identifying the constrained period
The 4-second SCADA output is analysed to investigate the response of the three utility PV plants to
AEMO dispatch instructions limiting maximum plant output during particular periods. The record of
dispatch cap instructions can be found in AEMO’s 5-minutes dispatch data which is also publicly
available [29].
A semi-scheduled cap can be applied due to network constraints at any time, not necessarily at the
peak generation of utility-scale PV. This may include times when the output profile is ramping up or
down. In this analysis, the average of 4-second SCADA output over the 5-minute dispatch interval is
compared with the dispatch target of the flagged interval. The results of this analysis are shown in
Section 8.2.
Figure 6 Methodology diagram for controllability analysis of PV plant output
11
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
4. Overview of SCADA Outputs and Sample analysis
Figure 7 shows the generation profile for selected NEM plants compared to their dispatch targets for
a 24-minute window on 18th October 2016, a period where both minor under and over-frequency
deviations were present. The output of scheduled generators which were not providing regulation
services can be seen to track their dispatch targets reasonably well. Gas and hydro plants during this
period demonstrated steady output while some slightly larger fluctuations around the targets could
be seen in coal-fired power plants. For semi-scheduled generators (wind and PV power plants), the
generation stayed around the dispatch forecast “target” with larger deviations from the linear
trajectory compared to scheduled generators. This is to be expected given the underlying variability
of the wind and solar resource, and because they are not required to follow these targets.
Meanwhile, the generators with regulation services enabled (denoted with “with regulation”)
responded actively to frequency deviations. Note that when plants are enabled to provide regulation
services, they are given a “regulation” target at 4-second intervals based on how AEMO configures the
generator governors. This target is represented with blue lines in Figure 7. The active response from
generators that were providing regulation services is shown by the actual generation (black) that is
attempting to follow the regulation target (blue) instead of the linear trajectory (green). The regulation
target was above the linear trajectory during the frequency droop below 50Hz (shaded area), while it
was set under the linear trajectory during the over frequency period (white area). Faster ramping
scheduled generation like Hydro gravity, CCGT and OCGT demonstrated significant ability to follow
their dispatch instructions. Slow ramping scheduled generation like coal plants, in contrast, could not
accurately follow their assigned regulation targets. Interestingly, they are typically the main provider
of regulation services in the NEM. Far better control has been exhibited by battery energy storage
plants and even the wind farms which participated in the FCAS trials described earlier [23].
12
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Figure 7 Generation profiles for selected NEM plants on 18th October 2016 from 11:02 to 11:26. The grey shaded area
represents under frequency intervals (<50 Hz) while the white area represents over frequency intervals (>= 50 Hz). The
generators which were providing frequency regulation are denoted as “with regulation”.
13
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
5. Forecast deviations of PV and Wind
Research has highlighted that the very short-term variability of PV is considerably greater than wind
[30]. Both wind and PV are classified under the same semi-scheduled participation status by AEMO
but their characteristics are different. Figure 8 shows the 4-second MW deviations for a wind and PV
plant of roughly equivalent MW size against both the current forecast target and our proposed
timeshift of these forecasts, classified according to the output of the plant, and using a year of data.
The percentile lines (99.3%) bound the cumulative number of hours that the deviations exceeded
these levels. The 99.3% is equivalent to 61.3 hours of the year outside the boundary. The whiskers
represent 1.5 interquartile range which would cover 99.3% of the data if it was a normal distribution.
The large gap between the whiskers and the percentage line for the PV plant (Figure 8(a)-(b)) implied
that the distribution of deviations is not normal.
The deviation of the wind farms (Figure 8(c)) is greatest in the middle range of their capacity factor
while it narrows as the output goes towards maximum output. This is an outcome of the rapid changes
in output during the mid-range of the wind turbine power-wind speed curve, with more of the turbines
operating at rated power and hence less impacted by wind speed changes as wind farm output grows
[31]. Overall, the MW deviation of the wind farm remains within +/- 20% of the plant rated capacity
for 99.3% of the time with the current target setting, and this can be considerably improved with our
modified targets (Figure 8(d)).
In contrast, the deviation of the PV plant is considerably greater when approaching its full capacity
(Figure 8(a)). During full capacity generation periods, usually clear sky periods during the middle hours
of the day, PV is prone to fast moving cloud which can cause a sudden drop in output. Figure 8(a)
shows that the negative deviations (i.e. dispatch target > actual generation) increase when the level
of generation is reaching full capacity. The negative deviations at 99.3 percentile demonstrated an
almost one-to-one linear relationship between the size of MW deviations and the level of generation.
For example, a deviation of 60% could be expected when the PV plant operated at 80% of its rated
capacity. This suggests that the system operator should be prepared for large variability from PV plants
during peak PV generation periods. Our proposed adjustment can improve the level of deviations
(Figure 8(b)), although the improvement may not be as significant as for wind.
It is worth noting that these MW deviations are not only caused by weather variation but also include
forecast errors, unplanned operational issues and errors due to calculation procedure.
14
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(a) (b)
(c) (d)
Figure 8 4-second MW deviation per unit of plant capacity against the dispatch targets per unit of plant capacity with colour
coded system frequency of (a) PV farm as per original data, (b) PV farm with corrected linear trajectory and (c) Wind farm as
per original data and (d) Wind farm with corrected linear trajectory with whiskers indicating 99.3 percentile of a normal
distribution and trendlines indicating the actual 99.3 percentile derived from the data.
6. Deviations of all NEM generators
To put the deviations of PV and wind plants into the wider NEM context, the short-term deviations of
all NEM scheduled and semi-scheduled generators from their targets/forecasts are analysed. This
study categorises 229 generators into 12 technology types and by their participant status (Figure 1).
Figure 9 illustrates the mean of deviations from the dispatch target and the standard deviation of
these values for each generator. For non-wind and solar plant, deviations can be caused by the
ramping characteristics of each generator and other technical characteristics, as well as their
participation in the provision of frequency regulation services. Most of the synchronous, dispatchable
generators have a tight deviation range (average 1.5%) compared to wind (4.7%) and PV (7.6%).
Furthermore, the deviations for the wind and PV plants are caused mainly by the intermittency of the
resources and the forecast uncertainty, whereas they are not providing any regulation services. PV
plants have the largest mean deviations amongst all technologies.
15
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Figure 9 The mean and standard deviation of output deviations from their linear trajectory for 185 NEM generators (exclude
non-scheduled generators), categorised by technology type. The grey bars indicate the plants that provide regulation FCAS.
The deviations of PV and wind were calculated with the current AEMO linear trajectory.
7. Aggregated power deviation and its relationship with frequency
excursions
Throughout the year, the aggregated MW deviation from target trajectory of all the NEM generators
is within the range of 500 MW which is around 1% of total generation capacity (Figure 10). Some large
generation trips are also visible in the figure. The negative trend of the deviation is shown by fitting a
linear line. This shows a negative correlation between frequency deviations and MW deviations, which
implies that in aggregate, generation deviations from targets are generally working to return
frequency excursions back to 50 Hz. This is of course mostly delivered by those generators providing
regulation services.
Figure 10 Aggregated generation deviations of all generators in the NEM
When considering the contribution of each generator towards assisting in frequency regulation, as
shown in Figure 11, strong negative correlations representing significant corrective actions are mainly
found in Steam Sub-critical power plants followed by Hydro power plants. These two technologies are
the main providers for frequency regulation services but even units that are not enabled for regulation
typically deviate from their targets in a useful manner (plants that provide regulation FCAS are shown
16
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
as ‘*’ in Figure 11). Meanwhile the correlation coefficients of other technologies such as OCGT and
Steam Super-critical units (only a few of the coal plants in the NEM) are, like wind and solar, mostly
deviating in ways that make regulation more challenging in the very short-term, potentially
contributing to the frequency excursions. This result could be partly attributed to plants which
participated in regulation FCAS but were not enabled.
Figure 11 Pearson correlation coefficients between 4-second MW deviations and system frequency deviations of 185
(exclude non-scheduled generators) generators registered in the NEM as of 2016 categorised by generation technologies
(only shown for the main technologies). The correlation coefficients of semi-scheduled generators (PV and wind) were
calculated with the current AEMO linear trajectory
More detailed analysis of MW deviation as a function of frequency deviation, by plant status
(scheduled or semi-scheduled), and whether enabled for regulation was undertaken to better
understand behaviour. Scheduled generators not enabled for regulation (mainly CCGT, OCGT and Coal
Steam Super-critical) generally do not show a clear relationship between deviation from generation
target trajectory and frequency (Figure 12(a)), although there is some evident useful response for
minor frequency droop. At present in the NEM, local governor action to assist in frequency deviations
from generators even when they are not enabled for regulation is not uniformly implemented with
some units doing this and others not [32]. The highly negative MW deviations at -0.15 Hz and below
Figure 12(a) are during contingencies as a result of unit trip, seen as major deviations during those
periods. Although the MW deviations of these scheduled generators are very small (as shown by the
99th percentile lines for individual plants, which are very close to zero), the analysis shows that the
aggregated values are significant. In contrast, Figure 12(b), showing the MW deviation of scheduled
generators that provide FCAS shows, unsurprisingly, a clear correcting impact on frequency deviations.
Figure 12(c) and (d) demonstrates that the deviations of semi-scheduled generators (wind and PV)
have very little correlation to frequency deviation. A weak negative contribution is shown by the
positive slope of the linear fit. This finding is similar to that from Figure 11 and suggests that these
plants’ deviations from forecast are adding, modestly, to frequency deviations in general. While semi-
scheduled generation may lead to more requirement for regulation services to manage the variability
that it produces in the grid, contributions to frequency deviation from non-FCAS scheduled generation
participants are similar as shown by the similar slope coefficients in Figures 12(a) and (c) which is a
negligible difference to the slope in Figure 12(b).
In Figure 13, the negative contribution to frequency deviation is reduced by almost half for both PV
and wind farms with our proposed linear trajectory forecast correction, highlighting the potential
value of improving the integration of forecasts into the setpoints and trajectory in the NEM dispatch.
17
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(a)
(b)
(c)
(d)
Figure 12 Aggregated MW deviations of (a) scheduled generators (non-participants in regulation FCAS), (b) scheduled
generators (participants in regulation FCAS), (c) semi-scheduled PV farms and (d) semi-scheduled Wind farms
18
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(a)
(b)
Figure 13 Aggregated MW deviations calculated according to the proposed linear trajectory calculation of (a) semi-
scheduled PV farms and (b) semi-scheduled Wind farms
8. Controllability of Utility PV plant
8.1 Variability during peak generation
A frequency distribution of the variability of each of the utility PV plants during peak generation
curtailment periods is shown in Figure 14. Most of the fluctuations of all three PV plants during the
peak generation curtailment periods are within ±0.2% of the plant capacity which is around ±0.1 to
0.2 MW. Figure 14(a) shows the results for Nyngan. A zero deviation occurs 85% of the time showing
that the plant can produce extremely smooth output. For the two other plants, output generally sits
a little higher than its forecast target. This could be the effects of DC:AC ratio and discrepancies in
forecast models. As previously shown in Table 3, the DC:AC ratio of Broken Hill is slightly higher than
Nyngan. Nonetheless, Broken Hill experiences significantly higher levels of small fluctuations as shown
in Figure 14(b). It is a smaller plant which may be one factor, however, it also suggests differently
configured control arrangements. Another factor that could influence the fluctuations is the mounting
technology. Figure 14(c) highlights the significant deviations of the single axis tracking plant Moree.
Note that these deviations exclude the dip in single-axis tracking generation profile during the peak
generation time [5]. Despite a peak flat-top generation profile, as shown in Figure 15, the peak
generation does fluctuate. One of the reasons could be the easily disturbed direct normal irradiance
(DNI) which is the main solar insolation component for a single-axis tracking system generation [33].
While most of the fluctuations are between ±0.2% of plant capacity, all the plants do deviate above
target – see for example a sample day of clipping generation at Broken Hill as shown in Figure 16.
19
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Nonetheless, the results suggest good controllability of PV plants for both fixed- and single-axis
tracking. PV plants can produce very steady output when solar insolation permits.
(a) (b)
(c)
Figure 14 Frequency distribution of the difference between 4-second SCADA output and 15-minutes moving average
generation at (a) Nyngan, (b) Broken Hill and (c) Moree
20
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Figure 15 Example of generation profile on a clear sky day at Moree
Figure 16 Behaviour of generation at Broken Hill pushing past its rated capacity
8.2 Generation under dispatch constraint
The accuracy of the response of the PV output to dispatch constraint instructions is visualised by
plotting the average of the 4-second SCADA output against the dispatch target of that interval (Figure
17). Overall, Nyngan and Broken Hill responded well to these periods, staying below their ‘constrained’
dispatch targets in the great majority of periods, although Nyngan did not receive many constraint
orders. Moree’s behaviour under constraints did not show a clear relationship, indicating potential
issues with the plant control system.
21
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(a) (b)
Figure 17 Scatter plots between 5-minutes averaged of 4-second actual generation against 5-minutes dispatch target at (a)
Nyngan and (b) Broken Hill
It is of particular interest to observe how well PV plants can maintain their output under target over
longer constrained periods. For this purpose, this study investigates the periods when a semi-
scheduled PV plant constrained target has been flagged for consecutive 5-minute intervals. The results
of statistical analysis are shown in Figure 18. The analysis is produced by calculating the frequency
distribution of 4-second MW deviation from the dispatch target. Broken Hill experienced many more
constrained periods than the other plants, exceeding the dispatch cap less than 0.27% of the time, of
which, 0.25% was within 1% (0.5 MW) above the plant capacity. This could result from the small
fluctuations that were also found when the plant is under curtailment at inverter capacity (Figure 16).
Nyngan and Moree both have quite symmetrical deviations under and over target. However, Nyngan
has many more deviations away from zero, including 3.88% of analysed 5-minute periods with
deviations greater than 1% (1 MW) of plant capacity. An example profile of Nyngan output fluctuating
around the dispatch target during a long constrained period is shown in Figure 19.
(a) (b)
11.62%
21.98%
38.53%
23.99%
3.88%
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
< -1% >= -1% zero <= 1% > 1%
Frequency of occurence
of plant capacity
Nyngan
Total 5-min intervals: 1,931
2.99%
13.46%
83.28%
0.25% 0.03%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
< -1% >= -1% zero <= 1% > 1%
Frequency of occurence
of plant capacity
Broken Hill
Tota 5-min intervals: 14,929
22
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
(c)
Figure 18 Frequency distribution of 4-second MW deviation of PV plant from dispatch target during consecutive periods
with dispatch constraint flag at (a) Nyngan, (b) Broken Hill and (c) Moree. The horizontal axis indicates the size of MW
deviations as percentage of plant capacity.
Figure 19 Example profile of Nyngan being under dispatch constraint for a consecutive period (from 8.25AM to 4.20PM)
9. Discussion
By explicitly analysing the deviation around dispatch target of individual generators, our analysis
shows that PV has by far the highest mean deviation of all generation technologies. This result is as
expected as PV is known to have higher variability and uncertainty than wind [34]. However, we have
been able to quantify the magnitude of this difference for the Australian NEM the mean deviation of
PV generation (7.6%) is around five times greater than the average of the mean deviations of
dispatchable generators (1.5%). Our study also shows, however, that this is not because PV plants are
not capable of precise dispatch, albeit within the constraints of the changing solar resource
availability. This raises the question of whether PV plants should have higher requirements around
meeting their dispatch target, in a similar manner to other generators.
The way that the electricity market incorporates uncertainty from VRE affects the uncertainty seen by
the system operator as found in [10]. Our study has further explored how these forecast targets could
be incorporated into dispatch and proposed a modification to the linear trajectory for PV currently
used in the NEM which would significantly reduce the deviation of PV around forecast target with no
change to the forecasting method.
1.84% 8.56%
81.09%
7.05% 1.46%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
< -1% >= -1% zero <= 1% > 1%
Frequency of occurence
of plant capacity
Moree
Tota 5-min intervals: 979
23
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
The analysis has also highlighted the potential implications of incoherence between dispatch and FCAS
arrangements. The high deviation observed from forecast dispatch targets indicates that the PV plants
in the NEM generally output their maximum generation available. While they are not currently
required to adhere to a linear trajectory according to dispatch procedures, these results suggest that
they are not currently attempting to avoid the Causer Pays penalties that apply from contribution to
frequency deviations in the NEM. This is likely a result of the relatively much higher prices in the real
time energy dispatch market than in the regulation FCAS market [35]. The pass through of FCAS costs
to generators via Causer Pays penalties does not currently provide sufficient incentive for generators
to trade-off generation in the energy market to follow their target, despite PV plants being
demonstrated to be controllable. As penetrations grow, electricity industries will need to consider
how to incentivise PV to reduce their uncertainty while recognising their inherent resource variability,
which imposes high costs on precise controllability.
When interpreting PV variability and uncertainty according to generation level, the analysis of
deviations presented in Figure 8(a) indicates that the volatile periods for PV plants are when they are
operating at their rated capacity. This results agree with the findings in [5] where a high variability
index occurred during partially cloudy and cloudy days with a medium to high clear sky index which
implies PV generation close to rated capacity with intermittent cloud passing. This is in contrast with
the characteristic found with the wind farm where the output is more certain at its rated capacity
(Figure 8(c)). This implies the potential need for higher FCAS requirements during peak PV generation
periods.
Nevertheless, our study found that the correlation between frequency deviation and MW deviation
from the dispatch target of PV is not very different from that of a gas turbine unit (OCGT) that is not
participating in frequency regulation. In aggregate, wind and solar deviations would seem to be
making a contribution towards increasing the frequency deviations to be managed by regulation FCAS
markets, although the impact is currently minor, due to the lack of correlation of 4-second variations
across wind and solar plants. In comparison, scheduled coal, gas and hydro units that are not enabled
for FCAS regulation show a minor tendency for deviations from their target to assist in correcting
frequency deviations, yet this impact also looks modest. This is because generators in the NEM have
not been required to provide primary frequency response unless participating in the FCAS markets. A
stronger correlation between frequency deviation and non-regulation dispatchable generator outputs
would of course be expected if they are required to enable primary frequency response [32]. The
question of whether to require primary frequency control from NEM generators not enabled for
regulation FCAS is receiving growing attention given growing frequency control challenges in the NEM,
and the Australian Energy Market Commission is currently considering feedback from stakeholders on
such a proposal [32].
Although the overall trend (Figure 10) showed that the NEM fleet has been able to correct frequency
deviations, Figure 11 and Figure 12(b) show that it has been heavily relying on a number of large aging
thermal power plants (mainly Steam sub-critical), some of which are planned for decommissioning in
the near future [36]. This will lead to reduction in FCAS capacity, and the system operator will need to
source these services from new market entrants.
The impact of wind farm trips at extreme over-frequency variations in Figure 12(d) provides evidence
that semi-scheduled generators should be considered as potential providers of FCAS regulation
24
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
services. The 600 MW reduction in power output within 4 seconds was equivalent to 18% of the total
rated wind farm capacity in the NEM, and assisted to manage the over-frequency deviation. More
recent incidents in the NEM in 2018 similarly saw PV plants contribute to lowering the frequency
during an over-frequency event [24]. This demonstrates the potential of these highly controllable
inverter connected technologies in providing rapid active power response to frequency deviations. PV
performed well in responding to dispatch instructions as found in the controllability analysis. This
demonstrates that PV could behave like a dispatchable generator by controlling its output. This would
reduce the level of MW deviations around the dispatch target and also reduce the Causer pays cost,
but of course would require PV to curtail down to be able to provide firm generation.
10. Conclusion
Our findings highlight some key aspects of utility PV variability, deviation from forecasts and ability to
follow constraints compared to the behaviour of other generating units. These results have
implications for short-term frequency control and the potential controllability of these units if and
when required.
The uncertainty analysis demonstrates that the NEM’s linear ramp calculation procedure does not
account for the fast characteristics of PV generation, resulting in consistent bias errors in PV
generation compared to targets. This highlights the importance of arrangements for incorporating VRE
forecasts into dispatch targets. We demonstrate a simple modification to the translation of solar and
wind plant forecasts to target trajectories that greatly reduce these deviations. The deviation at 1.5
interquartile is reduced by 57% which reduces the influence on frequency deviation by 48%.
Additionally, PV plant output showed greater uncertainty when operating at peak output. Although
the overall operational behaviour in following the expected linear trajectory produces large
uncertainty, the results in Section 8 showed that the PV plants have the capability to precisely control
their output during curtailment periods and respond well to dispatch target constraints, with most
fluctuations lying within ±1% of the rated capacity.
Of the generator types in the NEM, utility PV exhibits by far the highest deviations of around 7.6%
mean deviation from their five minute ‘forecast’ dispatch targets, with wind the next highest. It is also
notable that all of the scheduled plants also exhibit some level of deviation from targets, although
considerably less than solar and wind on a per unit basis. Although these deviations were analysed in
the context of the NEM, similar trends in the findings could be expected in other industries for the
main generation technologies in electricity industries covered in this study. Additionally, we found
that PV plants had a weak negative contribution to frequency deviation, with correlation coefficient
of 0.06 to 0.09.
A negative contribution to frequency deviation can already be detected at 5% penetration of
intermittent renewable energy generation. As the penetration of intermittent generation in the NEM,
especially utility-scale PV, is expected to increase rapidly, it will be important to estimate the
aggregated variability and correlation of utility-scale PV plants at high penetrations, and the potential
role and participation of utility PV in managing frequency.
25
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Data availability
Datasets related to this paper can be found at https://www.aemo.com.au/Electricity/National-
Electricity-Market-NEM/Data/Ancillary-Services/Ancillary-Services-Market-Causer-Pays-Data and
http://www.nemweb.com.au/Data_Archive/Wholesale_Electricity/MMSDM/, public online data
sources hosted at Australian Energy Market Operator [26], [29]. The detail of data cleaning and
processing is listed in supplementary material.
Acknowledgements
This research is supported by University of New South Wales International Postgraduate Award and
Australian Government Research Training Program Scholarship.
References
[1] J. Vasilj, P. Sarajcev, and D. Jakus, “Estimating future balancing power requirements in wind–
PV power system,” Renew. Energy, vol. 99, pp. 369–378, Dec. 2016.
[2] S. Shedd, B. Hodge, A. Florita, and K. Orwig, “A Statistical Characterization of Solar
Photovoltaic Power Variability at Small Timescales Preprint,” in The 2nd Annual International
Workshop on Integration of Solar Power into Power Systems Conference, 2012.
[3] J. Marcos, L. Marroyo, E. Lorenzo, D. Alvira, and E. Izco, “Power output fluctuations in large
scale pv plants: One year observations with one second resolution and a derived analytic
model,” Prog. Photovoltaics Res. Appl., vol. 19, no. 2, pp. 218–227, 2011.
[4] A. Mills et al., “Dark Shadows,” IEEE Power Energy Mag., vol. 9, no. 3, pp. 33–41, May 2011.
[5] F. Bai, R. Yan, and T. K. Saha, “Variability study of a utility-scale PV plant in the fringe of grid,”
2017 IEEE Innov. Smart Grid Technol. - Asia Smart Grid Smart Community, ISGT-Asia 2017, pp.
1–5, 2017.
[6] R. van Haaren, M. Morjaria, and V. Fthenakis, “Empirical assessment of short-term variability
from utility-scale solar PV plants,” Prog. Photovoltaics Res. Appl., vol. 22, no. 5, pp. 548–559,
2014.
[7] H. Holttinen et al., “Methodologies to determine operating reserves due to increased wind
power,” in 2013 IEEE Power & Energy Society General Meeting, 2013, vol. 3, no. 4, pp. 1–10.
[8] Y. Tang, J. W. M. Cheng, Q. Duan, C. W. Lee, and J. Zhong, “Evaluating the variability of
photovoltaics: A new stochastic method to generate site-specific synthetic solar data and
applications to system studies,” Renew. Energy, vol. 133, pp. 1099–1107, Apr. 2019.
[9] I. H. Rowlands, B. P. Kemery, and I. Beausoleil-Morrison, “Managing solar-PV variability with
geographical dispersion: An Ontario (Canada) case-study,” Renew. Energy, vol. 68, pp. 171–
180, 2014.
[10] E. Ela, V. Diakov, E. Ibanez, and M. Heaney, “Impacts of Variability and Uncertainty in Solar
Photovoltaic Generation at Multiple Timescales,” National Renewable Energy Laboratory,
2013.
[11] H. Asano, K. Yajima, and Y. Kaya, “Influence of photovoltaic power generation on required
26
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
capacity for load frequency control,” IEEE Trans. Energy Convers., vol. 11, no. 1, pp. 188–193,
Mar. 1996.
[12] J. Gari da Silva Fonseca et al., “Regional photovoltaic power fluctuations within frequency
regulation control time frames: A study with high-resolution data,” Prog. Photovoltaics Res.
Appl., vol. 26, no. 6, pp. 402–413, 2018.
[13] S. Sayeef, T. Moore, S. Percy, D. Cornforth, J. Ward, and D. Rowe, “Characterisation and
integration of high penetration solar power in Australia: a solar intermittency study,” in 1st
International Workshop on the Integration of Solar Power into Power Systems, 2011.
[14] A. Kaur, L. Nonnenmacher, H. T. C. Pedro, and C. F. M. Coimbra, “Benefits of solar forecasting
for energy imbalance markets,” Renew. Energy, vol. 86, pp. 819–830, 2016.
[15] Q. Wang, H. Wu, A. R. Florita, C. Brancucci Martinez-Anido, and B. M. Hodge, “The value of
improved wind power forecasting: Grid flexibility quantification, ramp capability analysis, and
impacts of electricity market operation timescales,” Appl. Energy, vol. 184, pp. 696–713, Dec.
2016.
[16] V. Y. Singarao and V. S. Rao, “Frequency responsive services by wind generation resources in
United States,” Renew. Sustain. Energy Rev., vol. 55, pp. 1097–1108, 2016.
[17] C. Fernandes, P. Frías, and J. Reneses, “Participation of intermittent renewable generators in
balancing mechanisms : A closer look into the Spanish market design,” Renew. Energy, vol.
89, pp. 305–316, 2016.
[18] A. Banshwar, N. K. Sharma, Y. R. Sood, and R. Shrivastava, “Market based procurement of
energy and ancillary services from Renewable Energy Sources in deregulated environment,”
Renew. Energy, vol. 101, pp. 1390–1400, 2017.
[19] M. Cui and J. Zhang, “Estimating ramping requirements with solar-friendly flexible ramping
product in multi-timescale power system operations,” Appl. Energy, vol. 225, pp. 27–41, Sep.
2018.
[20] K. Zeb et al., “A comprehensive review on inverter topologies and control strategies for grid
connected photovoltaic system,” Renew. Sustain. Energy Rev., vol. 94, no. July, pp. 1120–
1141, 2018.
[21] 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,” Renew. Sustain. Energy Rev., vol. 62, pp. 971–987, 2016.
[22] Clean Energy Council, “Clean Energy Australia Report 2020,” Clean Energy Council, 2020.
[23] AEMO, “Hornsdale Wind Farm 2 FCAS trial,” Australian Energy Market Operator, 2018.
[24] AEMO, “Final Report – Queensland and South Australia system separation on 25 August
2018,” Australian Energy Market Operator, 2019.
[25] AEMO, “Participant Categories in the National Electricity Market,” Australian Energy Market
Operator, 2016.
[26] AEMO, “Ancillary Services Market Causer Pays Data,” 2016. [Online]. Available:
https://www.aemo.com.au/Electricity/National-Electricity-Market-NEM/Data/Ancillary-
27
Finalised version
0960-1481/©2020 Elsevier Ltd.
K. Keeratimahat, A. Bruce and I. MacGill, Analysis of short-term operational forecast deviations and
controllability ofutility-scale photovoltaic plants, Renewable Energy,
https://doi.org/10.1016/j.renene.2020.11.090
Services/Ancillary-Services-Market-Causer-Pays-Data. [Accessed: 14-Oct-2019].
[27] K. Keeratimahat, A. Bruce, and I. MacGill, “The impact of DC/AC ratio on short-term variability
of utility-scale PV plants,” in 2019 IEEE Power & Energy Society General Meeting (PESGM),
2019, vol. 2019-Augus, pp. 1–5.
[28] AEMO, “Dispatch,” Australian Energy Market Operator, 2017.
[29] AEMO, “DISPATCHLOAD,” Australian Energy Market Operator, 2016. [Online]. Available:
http://www.nemweb.com.au/Data_Archive/Wholesale_Electricity/MMSDM/. [Accessed: 14-
Oct-2019].
[30] I. Graabak and M. Korpås, “Variability Characteristics of European Wind and Solar Power
Resources—A Review,” Energies, vol. 9, no. 6, p. 449, 2016.
[31] J. Riesz, J. Gilmore, and I. MacGill, “Frequency control ancillary service market design: Insights
from the australian national electricity market,” Electr. J., vol. 28, no. 3, pp. 86–99, 2015.
[32] AEMC, “Primary frequency response rule changes - Consultation paper,” Australian Energy
Market Commission, 2019.
[33] S. Heslop and I. MacGill, “Comparative analysis of the variability of fixed and tracking
photovoltaic systems,” Sol. Energy, vol. 107, pp. 351–364, 2014.
[34] M. Anvari et al., “Short term fluctuations of wind and solar power systems,” New J. Phys., vol.
18, no. 6, pp. 1–14, 2016.
[35] AEMO, “Data Dashboard - NEM Dispatch Overview,” 2018. [Online]. Available:
https://aemo.com.au/Electricity/National-Electricity-Market-NEM/Data-dashboard#nem-
dispatch-overview. [Accessed: 10-Jul-2018].
[36] AEMO, “Advice To Commonwealth Government on Dispatchable Capability,” Australian
Energy Market Operator, 2017.
