Available online at www.sciencedirect.

com

ScienceDirect
Solar Energy 107 (2014) 351–364
www.elsevier.com/locate/solener

Comparative analysis of the variability of fixed and tracking

photovoltaic systems
Simon Heslop a,⇑, Iain MacGill b
a
School of Electrical Engineering and Telecommunications, University of New South Wales, UNSW Sydney, NSW 2052, Australia
b
School of Electrical Engineering and Telecommunications and Centre for Energy and Environmental Markets (CEEM),
University of New South Wales, Sydney 2052, Australia

Received 16 August 2013; received in revised form 17 January 2014; accepted 7 May 2014
Available online 28 June 2014

Communicated by: Associate Editor Bibek Bandyopadhyay

Abstract

Growing penetrations of PV generation in electrical networks pose new challenges for electricity industry operation and planning.
Characterising the variability of PV generation can assist in these tasks. This paper presents a comparative short-term (5 min) variability
analysis for fixed-tilt, single-axis tracking, dual-axis tracking and concentrator PV systems using time-synchronised data from The Desert
Knowledge Australia Solar Centre (DKASC) in Alice Springs, Australia. A number of analysis and data presentation techniques are
presented to assess different aspects of this variability over the course of the day, and across different seasons. Results highlight the very
different variability characteristics of fixed tilt and tracking PV systems, and hence the importance of differentiating between different PV
system types when analysing their potential operational impacts on the electricity grid.
Ó 2014 Elsevier Ltd. All rights reserved.

Keywords: Photovoltaics; Solar tracking; Solar power generation; Intermittent generation

1. Introduction for power system integration (Panel on Climate Change,

Grid connected photovoltaics (PV) has made remark-
able progress over recent years and is now achieving signif-
icant penetrations in a number of electricity industries
around the world (Energy Policy Network, 2012). The
technology presents a major opportunity to reduce the
environmental impacts of electricity industries, and its
costs have fallen markedly over recent years (Morgan
et al., 2012). However, the very variable and somewhat
unpredictable nature of the solar resource by comparison
with the conventional energy option poses some challenges
⇑ Corresponding author. Mobile: +61 406049674; fax: +61 2 93857226.
E-mail addresses: s.heslop@student.unsw.edu.au (S. Heslop), i.macgill
@unsw.edu.au (I. MacGill).
2011). PV system output can vary significantly from time
frames of seconds through hours to days, seasons and even
years. Secure and reliable power system operation requires
that supply precisely meet demand (and losses) at all times
and locations within the network. There are ever present
challenges for maintaining this supply/demand balance.
Load varies on daily, weekly and seasonal cycles and exhib-
its considerable uncertainties. A wide range of possible
contingencies also have to be considered, including the
sudden loss of a large conventional generator or network
element. PV’s operational characteristics therefore do not
represent an entirely new challenge for power system
operation, however it does potentially add to the complex-
ity of operational decision making. PV generation exhibits
marked daily and seasonal cycles. It also exhibits

http://dx.doi.org/10.1016/j.solener.2014.05.015
0038-092X/Ó 2014 Elsevier Ltd. All rights reserved.
352 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
potentially considerable variability within these cycles
depending on the weather and, particularly, cloud cover.
The actual operational characteristics of particular PV sys-
tems therefore depend greatly on these conditions at their
particular location. However, they will also be impacted
by other design and engineering choices including the tech-
nical specifications of key equipment such as the PV panels
and system size, and its spatial arrangement. A particular
issue is the orientation and tilt of fixed panel systems,
and the potential use of tracking mechanisms or concentra-
tor systems. Tracking systems orient the PV panels so that
they follow the sun across the sky over the day. Concentra-
tor systems orient reflecting surfaces towards the sun in a
similar manner, but then concentrate this direct solar inso-
lation onto the PV cells. The overall amount and general
daily and seasonal timing of PV generation is, naturally,
of key interest to PV system owners and operators as well
as other electricity industry participants. This has been a
key consideration in the choice between fixed and tracking
systems given the generally greater output and extended
morning and evening performance of tracking systems,
yet also their greater complexity and cost (Drury et al.,
2013). However, the variability and unpredictability of this
PV generation is also a key issue for power system opera-
tions. Such variability changes supply/demand balance
and hence system frequency generally, and local network
flows and hence voltages, locally. Power systems are gener-
ally required to operate within strict frequency and voltage
standards and high penetrations of PV may often increase
the challenges of meeting these. However, PV outputs that
are well correlated with load can actually reduce power
quality challenges while modern PV inverters offer oppor-
tunities to both mitigate adverse PV impacts, and even
reduce underlying-pre-existing voltage problems, through
active and reactive power management (Demirok et al.,
2009; Liu et al., 2008; Kerber et al., 2009; Hen-Geul
et al., 2012; Wenxin et al., 2013; Goodwin and Krause,
2013; Huijuan et al., 2012). Characterising the expected
‘unmanaged’ output of PV generation yet also its variabil-
ity, is required before these potential adverse impacts and
management opportunities can be ascertained. Of particu-
lar focus in this paper, are the potential implications of
fixed orientation and tilt PV systems versus tracking and
concentrating configurations on the nature of such short-
term output variability. The solar insolation reaching a
PV panel generally reflects some mix of direct (that is,
directly transmitted) and indirect (reflected) insolation.
The amount and mix of these depends on the particular
weather conditions, and the orientation of the panels with
respect to the position of the sun. Sunny skies and panels
oriented directly towards the sun maximise the direct com-
ponent of total insolation. For fixed tilt panels, mornings
and evenings will often see a high component of indirect
insolation falling upon the panels. Adding to the complex-
ities, concentrator systems can only concentrate direct
insolation onto the PV receiver. As such it might be
expected that the variability of fixed versus tracking and
concentrator systems might differ significantly. Solar mon-
itoring has long recognised the importance of accurate esti-
mates of both direct normal insolation (DNI) and diffuse
insolation in determining expected system performance,
although the expense of DNI stations has often limited
their application. With respect to actual experience with
PV systems, however, there have been challenges in per-
forming comparisons of the variability of these different
possible system designs given all of the other factors driv-
ing variability. Work to date addressing the variability of
PV systems has generally been concerned with the impacts
of PV generation on the electricity grid at the system and
distribution level due to its intermittent nature. Typically
the PV generation profile used for the impact analysis is
presented but without an associated analysis of its dynamic
behaviour. Such a dynamic characterisation might include
for example, how the level of PV generation variability
changes over the course of a day or season, ramp rates,
temporal and spatial correlation and variability across dif-
ferent time scales from seconds to minutes to hours and
beyond (see, for example, Bai et al., 2007; Bebic et al.,
2008; Lew et al., 2010; Saadat et al., 2011; Ueda et al.,
2008; Miyamoto and Sugihara, 2009; Papaioannou, 2008;
Enslin and Heskes, 2004; Tan and Kirschen, 2007;
Batrinu, 2006; Enslin, 2010). Work which does provide
some discussion on the characterisation of PV variability
includes (Renne et al., 2008) which present an analysis of
ramp rates. The PV output data set used in this analysis
is computed using the PVWatts model for 1 min irradiance
measurements fixed at a given latitude. Mills et al. (2009)
provides a brief analysis on the aggregate variability of a
number of single-axis tracking systems. As part of an inves-
tigation utilizing battery storage to mitigate PV intermit-
tency, (Mossoba et al., 2012) presents the results of three
deployments of one, three and 17 irradiance sensors finding
that ramp rates reduce with spatial diversity, and that the
frequency and magnitude of ramp rates are much greater
over 1 min averages compared to one-second measure-
ments. In Curtright and Apt (2008) the data set used for
analysis includes three single-axis tracking systems and a
distribution function on the frequency of the percentage
change in output is presented along with the power spec-
trum and a correlation measure between the three sites.
Spatial and temporal correlation of PV output is again dis-
cussed in Lave and Kleissl (2010) where data from four
sites across Colorado is analysed, finding that correlation
reduces with distance and increases with time period.
Hoff and Perez (2010) presents a model showing how the
aggregate variability for a fleet of PV systems reduces
according to the number of systems, cloud transit speed
and the area of which the systems span. Hoff and Perez
(2012) expands on this work and verifies the model against
actual measured irradiance data. In Click et al. (2012), a
variability analysis of various PV generation deployment
scenarios for Florida, USA is presented. In contrast to
(Renne et al., 2008; Mills et al., 2009; Mossoba et al.,
2012; Curtright and Apt, 2008; Lave and Kleissl, 2010;
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
Hoff and Perez, 2010, 2012), detailed characterisation
methods are applied, including hourly and sub-second his-
tograms on ramp-rates, as well as average hourly changes
in PV output for each month. Whilst (Renne et al., 2008;
Mills et al., 2009; Mossoba et al., 2012; Curtright and
Apt, 2008; Lave and Kleissl, 2010; Hoff and Perez, 2010,
2012) certainly provide valuable insights into the opera-
tional behaviour of PV systems, most are concerned with
spatial correlation (the smoothing effect) and only one
study utilises data for tracking systems. None of these
studies break up their analysis into seasonal or hourly com-
ponents. In this paper we utilise data from a unique PV test
facility located in Central Australia, The Desert Knowl-
edge Australia Solar Centre (DKASC), to investigate and
characterise the relatively short term (five minute by five
minute) variability of a range of PV systems including fixed
tilt, tracking and concentrator system configurations. As
the systems are all located on the one site, they see nearly
the same solar insolation hence supporting direct compar-
ison of the performance of different technologies. The
desert climate at the site includes both dry and wet seasons,
hence well differentiated periods of generally low and high
cloud cover. Five minute output data is available for each
system over a period of up to three years. We characterise
and compare the five minute variability for the different
system configurations by hour and season in order to better
understand their possible implications for power system
operation over these time scales. In the following Section we
outline the DKASC site and available systems. Section 3
describes the different methods utilised to develop the PV
variability characterisation. Section 4 presents the findings
from the methods described in Section 3. Section 5
discusses the results, gives thoughts on future work and
offers some concluding comments.
2. PV test systems at the desert knowledge Australia solar
centre (DKASC)
DKASC is located in central Australia in the town of
Alice Springs. Five minute generation output data is avail-
able for 21 PV systems of varying technologies and config-
urations, sized between 2–30 kW, located at one site at the
edge of the town. Alice Springs has a desert climate with
very little rainfall and largely clear skies during the dry sea-
son and therefore comparatively little output variability in
the PV systems due to cloud cover during that period.
What rainfall there is generally occurs between November
and February during the wet season with corresponding
common cloud cover, and hence increased PV variability.
The difference in cloud cover between the wet and dry sea-
son allows for an interesting variability comparison
between the two seasons. Of particular value for our study,
the site allows comparison of system performance across a
range of fixed, tracking and concentrating PV systems. The
systems analysed consist of 16 fixed tilt systems including
amorphous, hybrid, thin film, polycrystalline and mono-
crystalline panels, 1 single-axis tracking system, 3 dual-axis
353
tracking systems and 1 concentrating system. See Table 1
for details of these systems. The output data used for the
analysis consists of time synchronised, 5 min average kW
samples for the years 2009–2011. It should be noted that
not all systems have a full 3 years of data depending on
the date of their installation and installed instrumentation.
“Fig. 1” illustrates the different performance of fixed ori-
entation versus dual-axis tracking systems at DKASC over
a particular day. In this case both systems use the same
Kyocera polycrystalline technology and are rated at
5.4 kW. It is clear that tracking systems provide a greater
overall output than fixed tilt systems under such conditions,
with the difference in generation occurring during the morn-
ings and afternoons due to the more appropriate orienta-
tion of the panels on the tracking system as it follows the
sun across the sky. Single axis tracking systems follow the
sun from East to West at a fixed tilt angle to the horizon
while dual axis systems also track the sun’s height about
above the horizon. Concentrating systems also track the
sun whilst concentrating the sun’s direct normal insolation
(DNI) – that is, insolation arriving direct at and normal to
the reflection surface – onto a small yet highly efficient PV
panel (http://www.solfocus.com/en/technology/).
As noted earlier, a key question for power system
integration is what the potential implications for PV output
variability are of deploying tracking or concentrator
systems by comparison with fixed plate installations.
Certainly, cloud cover lying between the sun and the panel
can almost entirely disrupt the arrival of DNI, greatly
reducing the output of a concentrator system. More gener-
ally, different patterns of clouds for different positions of
the sun in the sky can have very variable and somewhat
surprising impacts on the total amount, and proportion
of DNI versus diffuse insolation arriving on PV panels of
different orientations. In this study we seek to characterise
such differences through analysing real system performance
for a range of PV systems located at the one site and hence
experiencing largely the same insolation. As an example,
“Fig. 2” highlights the different performance of fixed,
tracking and concentrator systems for a highly variable
solar insolation day at DKASC. The behaviour displayed
by the concentrating system is clearly very different to that
of the other systems with output often dropping to zero
during periods of cloud cover while the other systems
maintain at least some level of output.
3. Methodology
The time synchronised 5 min average kW data for all PV
systems is publicly available and was downloaded from the
DKASC website as a CSV file. The data was then cleaned
and imported into Matlab with all data analysis executed
using that software. The following sections describe a num-
ber of procedures developed for this work to concisely
summarise the variability of PV generation for the systems.
The results of applying these procedures are presented in
Section 4.
354 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364

Table 1
List of PV Systems at DKASC which have been included in this study.
Manufacturer Material Size (kW) Tracking Other
BP solar Polycrystalline 4.95 None Roof mounted
First solar CdTe Thin Film 6.96 None
Kaneka Amorphous 6 None
Sunpower Monocrystalline 5.805 None
BP Solar Polycrystalline 4.95 None
BP Solar Monocrystalline 5.1 None
Trina Monocrystalline 5.25 None
Kyocera Polycrystalline 5.4 None
Sanyo Heterojunction with Intrinsic Thin Layer (HIT) 6.3 None
Sungrid Monocrystalline 5.04 None
Sungrid Polycrystalline 5.04 None
Evergreen Polycrystalline 4.92 None
Calyxo CdTe Thin Film 5.4 None
Q-cells Upgraded metallurgical grade (UMG) silicon 5.85 None
Q-cells Polycrystalline 5.64 None
Q-cells Monocrystalline 5.64 None
Kyocera Polycrystalline 5.4 Single axis
ADES Unspecified 26.52 Dual axis
DEGERenergie Monocrystalline 31.5 Dual axis
Kyocera Polycrystalline 5.4 Dual axis
Solfocus High efficiency “triple junction” cells 16.8 Dual axis Concentrating

Fig. 1. Production profile comparison for a dual-axis and fixed-tilt PV system at DKASC during a sunny day with only two brief periods of cloud. Both
systems utilise the same polycrystalline Kyocera panels and are rated at 5.4 kW.

3.1. Data clean-up
It is important to note that there was only one concen-
trator system at the site as compared with 4 tracking sys-
tems and 16 fixed plate systems. Hence comparative
results for the concentrator system need to be treated with
care. Furthermore, the difference in variability between the
dual and single-axis systems proved to be minimal and
therefore there is little comparative comment on these
two different technology approaches. Most comment is in
relation to the difference between fixed-tilt and dual-axis
or concentrating systems. For a comparative variability
analysis to be possible it was crucial for all data to be time
synchronised. For occasions where small amounts of data
was missing the values for the same times from the previous
or following day were inserted. An output of zero was
assigned to larger sections of missing data – this data is
later filtered out and does not contribute to the variability
analysis.
3.2. Normalisation
As each system is of a different size, the data set for each
system had to be first normalised for a comparative
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364 355

Fig. 2. Plot of concentrating, dual-axis tracking and fixed tilt PV system output at DKASC over a particular day highlighting the greater variability and
reduced output of the concentrating system during periods of cloud cover, and hence reduced direct insolation. The output for each system has been
normalised, refer to Section 3.2 for an explanation of the normalisation method.

analysis. Normalisation was done against the rated capac-
ity specified by the system installers, typically installed PV
panel capacity, of each system.
3.3. Variability
The shortest time period possible for assessing variabil-
ity was, of course, five minutes given the five minute data.
Variability data sets were constructed from the normalised
data sets by shifting each PV system data set forward by
one 5 min time sample and then subtracting the shifted
data set from the non-shifted. This resulted in a new data
set, consisting of 5 min percentage change in output values
relative to the rated capacity.
3.4. Month v hour variability surface plot
Using the absolute of the variability data sets, from this
point the data is grouped into fixed, single-axis, dual-axis
output profile of PV systems. “Fig. 3” below gives a
schematic of the process for fixed systems.
The last step in this process involves some curve
smoothing, where each data point is set to the value of
the mean of the 3 point average in both the x-axis direction
(month) and y-axis direction (hour). This method is also
repeated to create a magnitude surface plot, where the
normalised magnitude is used instead of the variability.
3.5. Monthly and hourly magnitude and variability
For each system type (fixed, single-axis, dual-axis,
concentrator), the data to create the surface plots is used
to calculate the monthly and hourly magnitude and vari-
ability. The 5 min mean of the variability and magnitude
for each month and each hour is taken to achieve this.
Eqs. (1) and (2) define this process for variability
( ),
X
h¼1

and concentrating data sets. The data sets were then VM ¼ V mh H ð1Þ
H
grouped by month. Finally, the mean 5 min variability
( ),
for each hour of each month was calculated. We consider X
m¼1
that the month v hour surface plot gives a good indication VH ¼ V mh M ð2Þ
of how climate and seasons impact on the variability M

Fig. 3. Procedure for resolving data point for month v hour variability surface plot.
356 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
VM = Mean variability for a certain month m
VH = Mean variability for a certain hour h
H = Total number of hours
M = Total number of months
m = month
h = hour
3.6. Weekly scatter plot with mean
For a particular technology, all the 5 min variability data
points from the variability data set are collected, the 16
fixed-tilt variability data sets for example. The data for a
particular hour is then extracted. The data for this particu-
lar hour is then grouped into weeks producing an array con-
sisting of 52 columns. In the case of 16 fixed-tilt systems, we
would have 4032 rows; given there are 12 5 min samples in
an hour, 7 days in a week, 3 years and 16 systems. All zero
values are then removed. Two 5 min weekly variability
means are then calculated, one for all the negative data
points for each week and one for all the positive.
3.7. Histogram
The final analysis step is a magnitude and 5 min
variability analysis for any particular month and hour. A
histogram is used to achieve this. A histogram is a function
that counts the number of observations that fall into each
value range, known as bins. A mathematical definition is
given below
X
k
n¼ mi ð3Þ
i¼1
where n is the total number of observations, k the total
number of bins, i represents a value range and mi is the
number of observations that fall within the value range
of i. Looking at the variability histogram, all values are
between 1 and 1 and a bin size of 0.025 is used. The same
is done for the magnitude except that a bin size of 0.01 is
used. For a particular technology, all the 5 min variability
data points and the normalised magnitude data points are
collected. The data points for a particular hour during a
particular month are then extracted. These data points
are then assigned to a bin and the histogram developed.
4. Results
The aim of this analysis is to present the difference in
variability between fixed, single and dual axis and concen-
trating PV systems. The first graph “Fig. 4” gives an indi-
cation of how tracking systems can be more variable than
fixed system. The red plot is for tracking systems (1 single
and 3 dual) while the blue plot is for a group of fixed tilt
systems. This weather on this day is likely to have been
overcast as the ramp up in the morning and ramp down
in the afternoon for both types of systems is similar. If
the weather was fine then the tracking systems would have
far steeper ramp gradients. An overcast day can also be
assumed as the spikes in output are up instead of down,
indicating a gap in the clouds as opposed to a passing cloud
in a clear sky. Looking at the spikes at around 6:45 am,
7.30 am and around 5 pm we can see how much more a
tracking system can vary during the morning and evening
periods due to these changes in cloud cover. As expected,
with the difference in tilt between the fixed and tracking
systems reducing towards the middle of the day, the varia-
tion due to changes in cloud cover is much the same over
those times.
The next two figures, “Figs. 5 and 6” give a month v
hour normalised output surface plot for fixed-tilt and
dual-axis tracking systems, created using method Month v
Hour variability surface plot but using normalised output
instead of variability. Output data from all systems was
used to produce the figures. These two plots were included
to communicate how output for the two system types com-
pares across the course of the year. The greater output of
the dual-axis tracking systems in the morning and after-
noon is made most apparent by the length of the bottom
two segments, 6–7 am and 7–8 am. Looking at December
for example, the fixed-tilt system is producing at 0.3 while
the dual-axis tracking system is producing at 0.55. The var-
iation in the length of these segments in “Fig. 6” reveal the
degree to which the output for dual-axis tracking systems
varies across the year for this time of day, whilst staying
relatively constant for fixed-tilt.
4.1. Variability surface plots – month v hour
The first set of results presents the month v hour 5 min
variability surface plots, these results are obtained through
method Month v Hour variability surface plot. These plots
give a good indication of how the 5 min variability profiles
of PV systems vary throughout the year. Plots of this nat-
ure may be useful for the electrical industry in assessing the
potential variability in network flows, associated power
quality issues, and potential supply–demand challenges
associated with significant deployment of these different
PV configurations. “Figs. 7 and 8” below show these plots
for fixed-tilt and dual-axis tracking systems.
Examining the above plots, the months of November to
February are when Alice Springs receives the most rainfall
and therefore has greater cloud cover; this period clearly
exhibits greater variability relative to the remainder of
the year. At around 2 pm the variability (8%) is almost
double that of the April to September months (4%).
These surface plots would vary greatly according to
geographical location. Sydney for example has monthly
rainfall levels of over 4 in. 8 months of the year and would
give a very different surface plot. It is likely the difference in
variability throughout the year would not be as obvious as
that of Alice Springs which experiences quite pronounced
wet and dry seasons. As comparing the two plots is difficult
without seeing them from different angles “Fig. 9” shows
the difference in variability between fixed and dual for all
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364 357

Fig. 4. Plot showing the output of a number of fixed tilt systems (blue) and 1 single-axis and 3 dual-axis tracking systems (red) over a particular day. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5. Month v hour normalised average output surface plot for fixed-tilt systems.

Fig. 6. Month v hour normalised average output surface plot for the dual-axis tracking systems.

data points; “Fig. 9” highlights that the dual-tracking sys-
tem is more variable for all data points. The difference is
not large but this could be attributed to the large number
of data points used and the smoothing which occurred to
better define the climate and time dependent shape of sur-
face plot. Results in a different form presented later are
more suited for comparison. From the three figures in
can be concluded that the dual-tracking system is clearly
more variable during the mornings and afternoons. It also
obvious that the variability at these times increases during
the wet season, due to the combination of increase cloud
activity and the sun being higher in the sky, and rising ear-
lier and setting later than in the dry season.
4.2. Magnitude and variability by month and hour
The next set of results uses the method described under
Monthly and hourly magnitude and variability and more
358 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364

Fig. 7. Month v hour average variability surface plot for fixed-tilt systems.

Fig. 8. Month v hour average variability surface plot for dual-axis tracking systems.

Fig. 9. Average variability difference between fixed-tilt and dual tracking system surface plots.

clearly show the difference in 5 min variability between the
different technologies. “Fig. 10” shows the mean monthly
magnitude and 5 min variability for all system types. Note
that there is no variability bar for the concentrator system
for February; this data point was removed as its variability
value was so high that the values in the other months lost
resolution. The monthly magnitude values are normalised
against the highest value, this being the concentrator sys-
tem magnitude for August. We can see from the figure that
for every month the variability for tracking systems is
greater than for fixed. During the middle of the year the
variability is reduced and converges for all systems. This
can be explained by the fact that there is very little rainfall
during this period and thus no cloud cover to induce
output change. Where there is more rainfall, during the
months of November to February, the difference in vari-
ability increases. This difference can be attributed to the
morning and afternoon periods when the tracking systems
are still orientated towards the sun whilst the fixed tilt sys-
tems are not. During these periods there is far less direct
irradiance incident on the fixed tilt panels and they are
mostly producing power due to diffuse irradiance. Global
diffuse irradiance is not impacted as greatly as direct irradi-
ance by a change in cloud cover and so the fixed tilt systems
display reduced variability. Tracking systems on the other
hand are receiving greater direct irradiance and so a change
in cloud cover will result in a change in output resulting in
greater variability.
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364 359

Fig. 10. Monthly average magnitude and variability for all technologies.

Fig. 11. Hourly magnitude and variability for all technologies.

The evident greater variability of tracking systems
during the morning and afternoons is further highlighted
by “Fig. 11” which shows the variability for each of the sys-
tem types by hour. As per “Fig. 10”, the hourly magnitude
values are normalised against the highest value, the dual-
axis system magnitude for 12 pm. 7–8 am and 4–5 pm show
the greatest difference in variability, indicating that these
are the times where the tracking functionality is making
the most significant difference to output. Higher relative
output will of course give higher relative variability. This
is supported by the magnitude bar graph which shows a
large difference in output between the tracking and fixed-
tilt systems at these times. At 5 pm the tracking systems
are seeing variability of around 5% at a time when output
magnitude is only around 40% of rated capacity. The max-
imum variability is around 5.5% when output is close to
rated capacity. In comparison the fixed-tilt systems are
only exhibiting around 2% (compared to a max of 5.5%
at 1 pm) variability for an average output magnitude of
approximately 15%.
4.3. Weekly scatter plots with mean
The following set of results was produced using method
Weekly scatter plot with mean. The plots were devised to
illustrate the range of possible variability by taking a
“slice” of the surface plot to more clearly show how the
5 min variability changes throughout the year, and provide
further distinction between the fixed and dual-axis variabil-
ity profiles. “Fig. 12” is the weekly scatter plot with mean
for fixed-tilt systems at 2 pm. Looking at the scatter aspect
we can see variability reaching levels of 80%. The two mean
variabilities best illustrate the difference between the wet
and dry season with the positive mean moving from around
6% during the dry season (middle) to levels nearing 15%
during the wet (ends of the plot); the negative mean also
changes from 3% (dry) to 9% (wet). This reflects what
one would expect, with only minor negative variability as
there are no clouds to force drops in output.
To better show how the difference between fixed-tilt and
dual-axis systems a slice at 7 pm is taken, “Figs. 13 and 14”
360 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364

Fig. 12. Weekly scatter plot for fixed-tilt systems at 2 pm.

Fig. 13. Weekly scatter plot for fixed-tilt systems at 7 am.

Fig. 14. Weekly scatter plot for dual-axis systems at 7 am.

show these two plots. At this time of day, the difference in triple that of the fixed-tilt, the positive mean is double
variability between the two systems types is quite obvious. throughout the year and the negative mean could be said
The range of possible variation for dual-axis is almost to be triple that of the fixed-tilt during the wet season.
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
The impact of climate is really apparent at 7 am, especially
in regard to the negative mean. Of concern to the electricity
industry would be if these large drops in output are occur-
ring in the morning when demand is increasing.
4.4. Variability histograms
This next section of results presents what might be
thought of as the final drill down from the original, and
least definitive, month v hour surface plot; where the histo-
gram for an individual “cell” of the month v hour surface
plot is calculated. Again, the purpose of these histograms is
to show the difference in 5 min variability, in a more defin-
itive manner, at a specific time (hour of a month) of the
year. These histograms might well be of most use to elec-
tricity industry planners and operators as they give the
essential detailed information required to take into account
the potential impacts of PV system variability on the like-
lihood of negative and positive ramp rates of all magni-
tudes. The histograms are for both magnitude and
variability and for both fixed-tilt and dual-axis for easy
comparison. Magnitude is given to show at what times
and to what extent dual-axis is outperforming fixed-tilt.
These plots provide answers to questions such as, for exam-
ple, how frequent and how large are the potential drops in
PV output during early morning periods of increasing
load? “Figs. 15 and 16” show the histograms of PV output
and variability for both fixed and tracking systems at 7 am
for the June dry season, and February wet season. For
June there is little concern that large drops in output are
likely to occur with virtually no negative changes greater
than 2.5% for either system; an indication of negligible
cloud activity during this period. Even for February when
there is maximum cloud activity the chance of a negative
change greater than 2.5% is very small.
Comparing the behaviour of the two systems for both
time periods we see that in both cases fixed-tilt systems
have a far greater percentage of samples in the 0–2.5%
bin, indicating less variability. Dual-axis has a greater
Fig. 15. Magnitude and variability histogram for June 7–8 am, fixed-tilt and dual-axis.
361
variability spread with values in bin 10–12.5% for June
and bin 15–17.5% for February; by comparison fixed-tilt
only manages maximum values in bin 7.5–10% for June
and 10–12.5% for February. The magnitude histogram also
gives a useful indication of the level of variability: the
greater the spread the greater the variability. The magni-
tude spread for February reaches 0.8 for dual-axis systems
while fixed-tilt is restricted to less than 0.3. As expected the
majority of the variability is on the positive side and can be
attributed to the normal increase in output as the sun rises.
“Figs. 15 and 16” allow for a variability comparison
between the wet and dry season and “Fig. 16” allows for
a comparison between fixed-tilt and dual-axis system
behaviour. In summary, due to the lack of cloud activity,
the majority of the difference in variability between the
two systems could be attributed to the dual-axis system
having a steeper output gradient. On the other hand in
February, the variability for dual-axis systems is substan-
tially greater and cannot be solely due to this steeper gradi-
ent. This indicates that increased cloud activity during the
mornings and afternoons will result in increased variability
for a dual-axis system. The next figure, “Fig. 17”, is for
2–3 pm in February and shows, as during this period both
systems are exposed to a similar amount of irradiance, how
similar their behaviour despite this being the period of
greater cloud activity. We can see that the magnitude
spread for both systems is similar and in agreement with
the variability spread.
The final histogram plot presented, “Fig. 18”, is for
2–3 pm in June. The purpose of this is to show how predict-
able the output for a PV system can be at this time of year:
the middle of the dry season (no cloud activity) when the
sun is at its peak (minimal change in output). For the
fixed-tilt system over 70% of samples are in bin 2.5 to
0% compared to 40% for February, almost double the
number of samples. The dual-axis system exhibits similar
reductions in variability with an increase in samples from
around 30% to over 50%. This low variability is supported
by a heavy concentration (as opposed to the large spread
362 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364

Fig. 16. Magnitude and variability histogram for February 7–8 am, fixed-tilt and dual-axis.

Fig. 17. Magnitude and variability histogram for February 2–3 pm, fixed-tilt and dual-axis.

Fig. 18. Magnitude and variability histogram for June 2–3 pm, fixed-tilt and dual-axis.
 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
seen in “Fig. 17”) of samples of normalised output around
0.7 for fixed-tilt and 0.85 for dual-axis.
4.5. Relationship between dual-axis and fixed-tilt
One final analysis step was to develop a surface plot
which showed the 5 min variability relationship between
dual-axis and fixed-tilt systems. Even though the absolute
variability difference between the two systems varies
through the year, proportionally it should only be depen-
dent upon the position of the sun; “Fig. 19” gives a strong
argument that this is the case. We can see from “Fig. 19”
that the proportional difference in variability is relatively
independent of the month and heavily dependent on the
time of day – with the proportional difference in variability
being at minimum around noon (close to 0%) and increas-
ing towards a maximum of 70% at the extremities of the
day. These findings prompted further work on exploring
this variability relationship. The work predicts the output
of a fixed-tilt system from that of a dual-axis. It uses the
incidence angle of the sun at every 5 min timestamp whilst
accounting for the immunity of diffuse output to passing
cloud to “reduce” the dual-axis output to that of a fixed-
tilt. This work has been separately published (Heslop and
MacGill, 2012). This relationship allows for the output
profiles and potential variability of dual-axis systems to
be modelled based upon data from fixed-tilt irradiance
measurement devices.
5. Discussion
This paper presents a variability comparative analysis
and characterization of four different types of PV technol-
ogies, all located at the same location. This geographical
proximity is the reason this analysis possible, allowing for
a time-synchronised comparison yielding results which
gives insight into the behaviour of fixed-tilt, dual-axis, sin-
gle-axis and concentrating systems. The hour v month var-
iability surface plots for the different technologies
highlights that when investigating the potential impacts
Fig. 19. Month v hour surface plot – percentage difference in average variability between dual-axis and fixed tilt systems.
363
of PV generation it is necessary to take into account the
type of PV technology, the time of day and time of year.
It should be noted that this particular characterisation of
PV variability is geographically dependent. The specific
variability findings could only be said to be relevant for
Alice Springs and surrounds, or other locations that share
a very similar climates. The analysis techniques and associ-
ated presentation plots, however, are of general value. The
expected 5 min variability hour-to-hour, broken down by
month characterises the behaviour of the PV systems in a
way which reflects the influence of the climate and the path
of the sun for different technology types. The weekly scatter
plots takes a “slice” of the month v hour surface plot and
gives the range of variability and shows more clearly the
manner levels to which the variability can vary for a certain
time of day throughout the year. It is also of value in com-
paring different possible system technologies. The final
form, the histogram, gives definitive statistics on the per-
centage chance for particular levels of variability and mag-
nitude. With the presentation of magnitude, all four factors
governing PV behaviour are represented: irradiance; tech-
nology type, sun path and climate. The results presented
in this paper confirm operational characteristics of different
PV technologies that one could intuitively surmise.
However, it also quantifies them for the particular climate
experienced at DKASC. Increasing penetration levels of
intermittent renewable generation and the associated injec-
tion of fluctuating power into the electricity network are of
concern for those responsible for the planning and opera-
tion of our electricity networks. It is hoped that the analysis
and data presentation techniques presented in this paper
can contribute to this task. Possible avenues for future
work include a comparison on the variability characteris-
tics of PV for different locations around the globe. This
would provide further insight into the influence of climate
and sun path on PV behaviour. Also, the work presented
here only deals with individual systems of small scale;
systems which on their own could have only a local impact
on the electricity grid. What is required is an equivalent
analysis on a large number of systems, for both a
364 S. Heslop, I. MacGill / Solar Energy 107 (2014) 351–364
centralised and distributed layout, examining their behav-
iour in aggregate. There are interesting issues regarding
the operation of concentrator systems that also seem to
warrant further investigation.
Acknowledgments
The authors gratefully acknowledge The Desert
Knowledge Australia Solar Centre for providing the data
used for the analysis presented in this paper. The authors
would also like to thank Australian Solar Institute (ASI)
projects for funding support on high PV penetrations, solar
resource characterisation and forecasting research. Simon
Heslop is a recipient of a PhD scholarship from ASI.
References
Bai, X., et al., 2007. Intermittency Analysis Project: Appendix B Impact of
Intermittent Generation on Operation of California Power Grid,
Report CEC-500-2007-081-APB.
Batrinu, F., et al., 2006. Impacts of grid connected photovoltaic plant
operation on the harmonic distortion. In: IEEE Mediterranean
Electrotechnical Conference, pp. 861–864.
Bebic, J., 2008. Power System Planning: Emerging Practices Suitable for
Evaluating the Impact of High-Penetration Photovoltaics, Subcon-
tract, Report NREL/SR-581-42297.
Click, D.K., et al., 2012. Effects of solar resource variability on the future
florida transmission and distribution system. In: Transmission and
Distribution Conference and Exposition, pp. 1–7.
Curtright, A.E., Apt, J., 2008. The character of power output from utility-
scale photovoltaic systems. Prog. Photovoltaics: Res. Appl. 16, 241–
247.
Demirok, E., et al., 2009. Clustered PV inverters in LV networks: An
overview of Impacts and comparison of voltage control strategies. In:
Electrical Power and Energy Conference (EPEC), IEEE, pp. 1–6.
Drury, E. et al., 2013. Relative performance of tracking versus fixed tilt
photovoltaic systems in the USA. Prog. Photovoltaics: Res. Appl.
http://dx.doi.org/10.1002/pip.2373.
Renewable Energy Policy Network for the 21st Century 2012, Renewables
Global Status Report. http://www.map.ren21.net/GSR/GSR2012_
low.pdf>.
Enslin, J., 2010. Network impacts of high penetration of photovoltaic
solar power systems. In: IEEE Power and Energy Society General
Meeting, pp. 1–5.
Enslin, J., Heskes, P., 2004. Harmonic interaction between a large number
of distributed power inverters and the distribution network. IEEE
Trans. Power Electron. 19, 1586–1593.
Goodwin, S.E., Krause, O., 2013. Mitigation of voltage band violations
through distributed active and reactive power control of inverter based
PV generation on LV networks. In: Power and Energy Society General
Meeting (PES), IEEE, pp. 1–5.
Hen-Geul, Y. et al., 2012. Adaptive VAR control for distribution circuits
with photovoltaic generators. IEEE Trans. Power Syst. 27, 1656–1663.
Heslop, S.F., MacGill, I.F., 2012. A simple method for predicting the
output of dual-axis tracking systems from fixed-tilt system outputs. In:
22nd Australasian Universities Power Engineering Conference
(AUPEC), pp. 1–6.
Hoff, T.E., Perez, R., 2010. Quantifying PV power output variability. Sol.
Energy 84, 1782–1793.
Hoff, T.E., Perez, R., 2012. Modelling PV fleet output variability. Sol.
Energy 86, 2177–2189.
Huijuan, L., et al., 2012. Real and reactive power control of a three-phase
single-stage PV system and PV voltage stability. In: Power and Energy
Society General Meeting (PES), IEEE, pp. 1–8.
Kerber, G. et al., 2009. Voltage limitation by autonomous reactive power
control of grid connected photovoltaic inverters. Compat. Power
Electron., 129–133.
Lave, M., Kleissl, J., 2010. Solar variability of four sites across the state of
Colorado. Renew. Energy 35, 2867–2873.
Lew, D., et al., 2010. Impact of High Solar Penetration in the Western
Interconnection, Technical, Report NREL/TP-5500-49667.
Liu, Y., et al., 2008. Distribution system voltage performance analysis for
high-penetration PV. In: Energy 2030 Conference, pp. 1–8.
Mills, A., et al., 2009. Understanding Variability and Uncertainty of
Photovoltaics for Integration with the Electric Power System, Report
LBNL-2855E.
Miyamoto, Y., Sugihara, H., 2009. Demonstrative research on clustered
PV systems. In: 34th IEEE Photovoltaics Specialists Conference, pp.
512–516.
Morgan, R., et al., 2012. The impact of a rapidly changing global PV
market on the PV manufacturing supply Chain. In: 38th IEEE
Photovoltaics Specialists Conference, pp. 1648–1650.
Mossoba, J., et al., 2012. Analysis of Solar Irradiance Intermittency
Mitigation Using Constant DC Voltage PV and EV Battery Storage.
In: Transportation Electrification Conference and Expo (ITEC),
IEEE, pp. 1–6.
Intergovernmental Panel on Climate Change 2011, Special Report on
Renewable Energy Sources and Climate Change Mitigation. <http://
srren.ipcc-wg3.de/report>.
Papaioannou, I., et al., 2008. Harmonic impact of small photovoltaic
systems connected to the LV distribution network. In: 5th Interna-
tional Conference on European Electricity Market, pp. 1–6.
Renne, D., et al., 2008. Solar Resource Assessment, Technical, Report
NREL/TP-581-42301.
Saadat, N., et al., 2011. A statistical approach to quantify the impact on
voltage quality caused by PV generators. In: 4th International
Conference on Electric Utility Deregulation and Restructuring and
Power Technologies, pp. 156–161.
Tan, Y., Kirschen, D., 2007. Impact on the power system of a large
penetration of photovoltaic generation. In: IEEE Power Engineering
Society General Meeting, pp. 1–8.
Ueda, Y. et al., 2008. Analysis results of output power loss due to the grid
voltage rise in grid-connected photovoltaic power generation systems.
IEEE Trans. Ind. Electron. 7, 2744–2751.
Wenxin, P., et al., 2013. Local load power factor correction by grid-
interactive PV inverters. In: PowerTech (POWERTECH), IEEE
Grenoble, pp. 1–6.