Documente Academic
Documente Profesional
Documente Cultură
Aristotle University of Thessaloniki, Department of Electrical and Computer Engineering, 54006 Thessaloniki, Greece
a r t i c l e i n f o
Article history:
Received 18 October 2009
Received in revised form 7 March 2010
Accepted 20 April 2010
Available online 15 May 2010
Keywords:
Photovoltaic plants
Solar inverters
Inverter size
a b s t r a c t
A new simple analytical method for the calculation of the optimum inverter size in grid-connected PV
plants in any location is presented. The derived analytical expressions contain only four unknown param-
eters, three of whichare relatedto the inverter andone is relatedto the locationandto the nominal power
of the PV plant. All four parameters can be easily estimated from data provided by the inverter manu-
facturer and from freely available climate data. Additionally, analytical expressions for the calculation of
the annual energy injected into the ac grid for a given PV plant with given inverter, are also provided.
Moreover, an expression for the effective annual efciency of an inverter is given. The analytical method
presented here can be a valuable tool to design engineers for comparing different inverters without hav-
ing to perform multiple simulations, as is the present situation. The validity of the proposed analytical
model was tested through comparison with results obtained by detailed simulations and with measured
data.
2010 Elsevier B.V. All rights reserved.
1. Introduction
The production of electricity from photovoltaic (PV) systems is
still expensive, when compared to conventional production meth-
ods. This fact necessitates the careful selection of each individual
part of a grid-connected PV plant in order to achieve maximum
energetic and economic performance. Part of the energetic opti-
mization is the proper selection of the rated power of the dc/ac
inverter with respect to the total nominal power of the PV array.
The common practice among many design engineers is to select
the rated power of the inverter equal to the total nominal power of
the PV array. Although this practice leads to PV systems that work
without problems, it is not the optimum from an energetic point
of view design, for the following reasons [15]: (1) the nominal
power of theinstalledPVmodules is achievedonlyunder their Stan-
dard Testing Conditions (STC), which are very unlikely to happen in
real conditions, (2) the solar irradiance varies, during the day, from
zero up to a maximumpoint which depends on the time of the year
and the geographic location of the PV installation, and (3) the ef-
ciency of any inverter is not constant over its operating range but
drops signicantly when the inverter operates at or below 1020%
of its ratedpower. Thus, aninverter withrelatively large power will
operate at low efciency during the sunrise and sunset or during
cloudy days. On the other hand, an inverter with relatively small
N to 60.3
to 90
19
N, 25
07
E), for
two typical days in June and December. The solar irradiance is at a
plane facing south with 28
in Iraklion, Greece, for typical days in June (a), and December (b).
It is well known that the photovoltaic panels are character-
ized by their currentvoltage (IV) curve, which varies with the
solar irradiance and the cell temperature. From the IV curve, the
powervoltage (PV) curve canbe derivedfor everysolar irradiance
and cell temperature value. The maximum power point tracker
(MPPT) is a device integrated into the dcac inverter that imposes
on the PV panels the appropriate voltage so that the maximum dc
power is extracted from them. The action of the MPPT is very fast
compared with the variation of solar irradiance or cell tempera-
ture. Therefore, it canbe assumedthat the panels operate always on
their maximumpower point for any given solar irradiance and cell
temperature. The electric power at the dc terminals of a PV array
is directly proportional to the solar irradiance and varies linearly
with the solar cell temperature [10]. For the commonly used crys-
talline silicon cells, the electric power reduces linearly with the cell
temperature for temperatures above 25
C [10]. in
C,
is the rise of the temperature of the cell above 25
C. Usually, the
cell temperature rises 30
19
N) was assumed 28
57
N), 39
.
The PDC for Cape Town (34
14
14
19
25
57
N.
facingnorthwithxedtilt angle equal tothe optimumone, whichis
21
25
N,
25
30
E) showed that T
max
is equal to 4380h. This is a natural
consequence of the fact that the year has 8760h with approx-
imately half of it being day and the other half being night.
Therefore, assuming that T
max
is constant for every locationand
equal to 4400h, introduces an error of less than 0.6%. The small
differences in T
max
between the various locations are due to
tilt of the axis of earth which makes the northern hemisphere
to accept sunlight for longer periods within a year than the
southern.
Therefore, PDC can be expressed as,
P
dc
(t) =
P
max
T
max
t +P
max
(3)
where t is the annual duration in h/year that the dc power is larger
or equal to P
dc
(t), P
max
is the maximum dc power (15min average)
that will appear in a specic location for a given installed PVpower.
The error that is introduced when using (3) instead of the actual
values of P
dc
(t) can be estimated by the Root Mean Squared Error
(RMSE), which is generally calculated by
RMSE =
_
k
i=1
[f (x
i
) y
i
]
2
k
(4)
where y
i
is the actual values of a variable and f(x
i
) is the respective
values obtained by a tting function. k is the number of data points
that were used for the tting. In this case, y
i
is the actual value of
the dc power while f(x
i
) is the tting function given in (3). RMSE
is calculated to be 2.3, 1.22, 1.27, and 1.03 for Cape Town, Iraklion,
Milano, and Edinburgh respectively. The respective data points are
230, 240, 196, and 153.
Although Eq. (3) is valid for every location and for PV panels
with crystalline silicon cells, it involves some inaccuracies that
should be mentioned. The rst inaccuracy is in Eq. (1) where it
was assumed that the dc power is directly proportional to the
solar irradiance, which means that the efciency of the PV panels
is constant and independent of the level of solar irradiance. In
reality, the conversion efciency of a solar panel drops slightly
as the incident-to-its-surface solar irradiance drops. For exam-
ple, the efciency of a typical PV panel with mono-crystalline
silicon cells drops from 13.2% at 1000W/m
2
to 12% at 200W/m
2
.
The second inaccuracy is in Eq. (2) where it was assumed that
the increase in the cell temperature is 30
C independently of
the power level it works. Actually, the increase in temperature
is in the range 2237
C or 37
C
is, [10.005x(T
amb
+3025)]/[10.005x(T
amb
+2225)] and
[10.005x(T
amb
+3025)]/[10.005x(T
amb
+3725)] respec-
tively. Obviously, the error depends on the ambient temperature.
Thus, for T
amb
=20
Euro
, is dened by
Euro
= 0.03
5%
+0.06
10%
+0.13
20%
+0.10
30%
+0.48
50%
+0.2
100%
(6)
where it is assumed that the inverter will operate at its nominal
power only for 20% of its operating time over a year, at 50% of its
nominal power for 48% of its operating time over a year, and so on.
It is obvious fromFig. 4 that a given inverter will operate, at a given
power level, for different time periods which depend on the loca-
tion. Thus, the time coefcients in (6) will differ from location to
location for a specic inverter. For this reason, the Euro-efciency
cannot generally represent the effective inverter efciency over its
whole operating range. This assertion will be proved in Section
4.
A simple mathematical function may describe, with very good
accuracy, the efciency curve of any solar inverter. The functionhas
the form,
(P
dc,pu
) = A +B P
dc,pu
+
C
P
dc,pu
(7)
where is the efciency of the inverter in, %, P
dc,pu
>0, the pu
value of the dc power. A, B and C are parameters to be deter-
mined. It is obvious from (7) that three pairs of (, P
dc,pu
) values
are needed to determine the A, B and C parameters. These pairs are
readily available fromthe inverter efciency curve provided by the
manufacturer. A very good choice are the pairs that correspond to
P
dc,pu
=0.1, 0.2 and 1pu, because, (as shown in Fig. 5), P
dc,pu
=0.1pu
corresponds to the rising front of the efciency curve, P
dc,pu
=0.2pu
corresponds toaregionnear thepeak of thecurveandP
dc,pu
=1pu
corresponds to the tail of curve. Parameters A, B and C can now
be easily calculated by solving a system of three linear equations
with three unknowns a task that can be accomplished even with
a pocket calculator. Of course, the best estimation of the A, B and C
parameter can be found with a least-squares method (LSM) which,
however, is very laborious. Table 1 shows, in comparison, the esti-
mated A, B and C parameters of all inverters shown in Fig. 5. The
estimation was made by LSM and by the simple system of three
linear equations mentionedpreviously. The correlationcoefcients
(R
2
) andtheRMSEfor bothestimationmethods aregiven. TheRMSE
for each inverter was calculated using (4) where now, y
i
are the
actual values of (P
dc
) given by the inverter manufacturer while
f(x
i
) are the values obtained by the tting curve given in (7).
It is evident fromTable 1 that Eq. (7) can describe accurately the
efciency curve of an inverter even when the A, B, C parameters are
C. Demoulias / Electric Power Systems Research 80 (2010) 11971204 1201
Table 1
Comparison of the estimated A, B and C parameters of Eq. (7) by a systemof linear equations and by LSM(in shaded cells). The respective R
2
coefcient, the RMSE, the number,
k, of data points and the standard deviations, , are also shown.
A B C R
2
RMSE k
A
B
C
Solar Konzept, 2kW 98.592 3.420 0.277 0.914 0.47 47
97.33 1.801 0.141 0.915 0.44 47 0.185 0.060 0.007
Sunways, 3kW 96.827 1.953 0.347 0.997 0.18 52
97.157 2.272 0.398 0.997 0.07 52 0.037 0.050 0.003
SMA, 5kW 97.644 1.995 0.445 0.993 0.29 42
97.004 1.580 0.362 0.994 0.19 42 0.098 0.077 0.006
SMA, 11kW 99.000 2.225 0.184 0.991 0.26 67
98.641 1.782 0.153 0.991 0.12 67 0.056 0.085 0.002
Satcon, 50kW 100.583 3.611 0.972 0.993 0.27 6
99.799 2.977 0.892 0.995 0.15 6 0.389 0.077 0.047
Satcon, 100kW 99.967 3.222 0.644 0.989 0.22 6
99.316 2.697 0.578 0.992 0.11 6 0.292 0.087 0.035
Siemens, 1000kVA 98.570 0.761 0.088 0.979 0.32 34
98.778 0.873 0.105 0.980 0.11 34 0.118 0.054 0.004
estimated by a simple system of linear equations. It should also be
noted that always, A>0, B<0 and C<0.
The standard deviations,
A
,
B
, and
C
in the estimation of the
A, B and C parameters respectively, are also shown in Table 1. The
standard deviations will be used later to estimate the variation of
the calculated optimum inverter size.
4. Analytical expressions for the optimum inverter size and
the energy yield
To calculate the optimum from an energetic point of view
inverter size of a PV installation with given PV modules in a spe-
cic location, the annual energy yield must rst be calculated. The
annual energy that will be injectedintothe ac gridcanbe calculated
withthe aidof Eq. (3) and(7). Since the locationandthe PVmodules
is known, the PDC can be constructed as shown in Section 2. Such a
typical curve is repeated in Fig. 6. The annual energy injected into
the ac grid must be calculated for two distinctive cases. In the rst
case it is assumed that the nominal dc power of the inverter, P
inv,N
,
is larger than the maximum available dc power of the PV array,
i.e., P
inv,N
>P
max
, while in the second case (also shown in Fig. 6)
P
inv,N
P
max
.
If P
inv,N
>P
max
, the annual energy yield, E, is given by
E =
Tmax
_
0
P
dc
(t) (P
dc,pu
)dt=
Tmax
_
0
P
dc
(t)
_
A +B
P
dc
(t)
P
inv,N
+C
P
inv,N
P
dc
(t)
_
dt
(8)
Using (8) and (3) and after some simple integrations, we get,
E =
A
2
P
max
T
max
+
B
3
T
max
P
2
max
P
inv,N
+C T
max
P
inv,N
(9)
where the annual energy yield will be in kWh, if T
max
is in hours,
P
max
and P
inv,N
in kW and the A, B and C parameters are expressed
as decimal numbers (i.e., not in %).
The optimum rated power of the inverter, P
opt
inv,N
, is found from
dE
dP
inv,N
= 0 P
opt
inv,N
= P
max
_
B
3C
(10)
In this case, the maximum annual energy injected into the ac grid,
E
max
, is found by substituting (10) in (9),
E
max
=
P
max
T
max
2
_
A 4
_
BC
3
_
(11)
The rst term in (11) is the surface beneath each curve in Fig. 4
and represents the maximum energy that could be delivered to
the ac grid if the efciency of the inverter was 100% during the
whole operating range. Therefore, the term in brackets in (11) is a
measure of the effective annual inverter efciency (provided that
the optimum inverter size is selected).
If P
inv,N
P
max
, the inverter will operate normally for
T
max
T
inv,N
h/year but, for T
inv,N
h/year (see Fig. 6), will deteriorate
the injected power to its nominal power. Thus, the annual energy
yield, E
, will be,
E
= P
inv,N
T
inv,N
(P
inv,N,pu
) +
Tmax
_
T
inv,N
P
dc
(t) (P
dc,pu
)dt (12)
For the pair (T
inv,N
, P
inv,N
), Eq. (3) gives:
P
inv,N
=
P
max
T
max
T
inv,N
+P
max
(13)
The inverter efciency at nominal power is, from Eq. (7),
(P
inv,N,pu
) = A +B +C (14)
Now, using (12), (3), (13) and (14), we get,
E
=
T
max
P
max
_
P
inv,N
P
max
(A +B +C)
_
A
2
+
2B
3
_
P
2
inv,N
_
(15)
Fig. 6. Typical dc-power duration curve. (a) Actual curve, (b) straight-line tting.
1202 C. Demoulias / Electric Power Systems Research 80 (2010) 11971204
Table 2
Inuence of the shape of the efciency curve of an inverter on its optimumsize and on the maximumac energy yield. Calculations according to the analytical model and LSM
(shaded cells). Possible variation of the inverter optimum size is shown in parentheses.
Inverter type P
opt
inv,N
/Pmax
eff
= Emax/(PmaxTmax/2) max euro
Solar Konzept, 2kW
2.02
0.964 0.968 0.962
2.06 (2.241.90)
Sunways 3kW (Inv3)
1.37
0.950 0.953 0.947
1.38 (1.421.34)
SMA 5kW
1.22
0.954 0.958 0.952
1.20 (1.281.13)
SMA 11kW (Inv1)
2.00
0.976 0.977 0.973
1.97 (2.091.85)
Satcon 50kW
1.11
0.962 0.964 0.952
1.06 (1.140.98)
Satcon 100kW
1.29
0.966 0.967 0.960
1.25 (1.371.14)
Siemens 1000kVA (Inv2)
1.69
0.980 0.981 0.979
1.66 (1.831.51)
where the annual energy, E
dP
inv,N
= 0 P
opt
inv,N
= P
max
A +B +C
A +
4
3
B
(16)
and the maximum energy,
E
max
=
T
max
P
max
2
(A +B +C)
2
A +
4
3
B
(17)
Since, P
opt
inv,N
P
max
, it follows from (16) that B3C. Therefore, the
optimuminverter size could be less than P
max
, only if the efciency
curve of that inverter is such that B3C (it is reminded that both B
and C are negative numbers, which means that |B| 3|C|). It should
be noted, however, that an inverter with B3C is a rather rare case.
All the inverters investigated by the author (see Fig. 5 and Table 1)
had efciency curves such that B3C. The only cases with B3C
were found in the literature [6,8], which, however, were models of
inverters, i.e., the authors did not mention the inverter manufac-
turer.
Table 2 shows the normalized optimum inverter size, the
effective annual efciency, the euro-efciency, and the maximum
efciency of the inverter types examined in this paper. It is clear
from Table 2 that the shape of the efciency curve of an inverter
(i.e., its A, B and C parameters) inuences signicantly its optimum
size and the maximum ac energy yield of a given PV installation,
i.e., given P
max
and T
max
. It can also be noticed that using the ana-
lytical method presented here, instead of the laborious LSM, results
in practically the same optimum inverter size. The possible varia-
tion in the estimation of the optimum inverter size is also shown
in Table 2 (values in parentheses), if it is assumed that the A, B and
C parameters can vary from their average value by 2
i
, i =A, B, C.
The respective
i
is shown in Table 1.
It can also be seen from Table 2 that the optimum size of an
inverter, for agiveninstallation, canvaryfromP
opt
inv,P
= 1.11 P
max
to
P
opt
inv,P
= 2.02 P
max
while the effective annual efciency when the
optimuminverter size is selected is larger than its euro efciency.
In any case, the effective annual inverter efciency is between its
maximum and euro efciency.
Three of the inverters in Table 2 were also designated as Inv1,
Inv2, and Inv3 for the purposes of the next section.
5. Validation of the model
The analytical model will be validatedintwo ways: rst through
comparison with results obtained by analytical simulations and,
second, through comparison with measurements.
5.1. Comparison with analytical simulations
Simulations were performed with the PVSYST software for a
hypothetical PV installation of 10kW
p
for the three locations in
northern hemisphere shown in Fig. 4. The climate data that were
given as input to PVSYST were extracted from [9] and were the
same to those used for the extraction of the PDCs shown in Fig. 4.
The Sharp NU-S5E3E, 185Wp, was arbitrarily chosen as PV panel.
The temperature coefcient of that panel is DP=0.42%/
C. The
electrical connection of the panels was assumed to consist of three
parallel strings with each string consisting of 18 panels connected
in series. The same tilt angles to those mentioned in Section 2 were
used. PVSYST has the ability to calculate, besides the temperature
losses, also energy losses due to:
- panel mismatch,
- reduction in panel efciency caused by the variation of the irra-
diance level,
- ohmic resistance in the wiring,
- voltage drops on the diodes of the panels,
- reduced quality of the PV panels.
All the aforementioned losses were deliberately set to zero so
that only the temperature losses are taken into account just like in
the model presented here.
For each location, three types of inverters were investigated
with P
opt
inv,N
/P
max
ratio equal to 2.00, 1.69 and 1.37 respectively. The
st type of inverter was assumedtohave anefciencycurve like the
SMA 11kWinverter shown in Fig. 5. This type of inverter will, from
now on, be called Inv1. The efciency curve was given as input to
thePVSYSTintheformof (, P
dc
) pairs. PVSYSTaccepts upto11data
points. The efciencies that correspond to 1%, 2%, 3%, 5%, 10%, 20%,
30%, 40%, 50%, 70%, and100%of the nominal dc power were given. A
number of simulations were runby changing ineachsimulationthe
nominal inverter dc power from 5 to 16kW with an increment of
1kW. Similar procedures were followed for an inverter with ef-
ciency curve like the Siemens 1000kVA (called Inv2) and for the
Sunways 3kW (called Inv3). PVSYST calculates, on an hourly basis,
the various parameters of the simulated system, including the ac
C. Demoulias / Electric Power Systems Research 80 (2010) 11971204 1203
Fig. 7. (a) Annual ac energy produced by a 10kWp PV plant in Iraklio, Greece, for
various inverter types as a function of the nominal dc inverter power. (a) Inv1, (b)
Inv2, (c) Inv3. (b) Annual ac energy produced by a 10kWp PV plant in Milano, Italy,
for various inverter types as a function of the nominal dc inverter power. (a) Inv1,
(b) Inv2, (c) Inv3. (c) Annual ac energy produced by a 10kWp PV plant in Edinburgh,
UK, for various inverter types as a function of the nominal dc inverter power. (a)
Inv1, (b) Inv2, (c) Inv3.
power injectedintothe ac gridandthus the annual energy. Fig. 7ac
shows the calculated annual energy in kWh, for the three locations
andthethreeinverter types. Theoptimumfromanenergetic point
of view nominal dc inverter power can be found from Fig. 7 for
each location and inverter type.
The optimum nominal dc inverter power can also be calculated
from the simple expressions in (10) or (16). For the three invert-
ers examined here, Eq. (10) was used, since for all of them B3C.
The P
opt
inv,N
/P
max
values for Inv1, Inv2 and Inv3 are already given
in Table 2. The maximum dc power, P
max
, in each location is one
Table 3
Optimumnominal dc inverter power, in kW, as calculated by simulation and by the
analytical method (AM) for a 10kWp PV installation.
Iraklion, Greece Milano, Italy Edinburgh, UK
Simulation AM Simulation AM Simulation AM
Inv1 15 14.40 12 11.8 9 9.34
Inv2 13 12.24 10 9.86 8 7.65
Inv3 10 9.86 8 8.22 7 6.30
tenth of the respective values given in Fig. 4 since in this case a
10kW
p
plant is assumed while the P
max
values in Fig. 4 were calcu-
lated for a 100kW
p
plant. Thus, the maximum dc power of the PV
plant is 7.2kW, 5.8kWand4.6kWfor Iraklio, MilanoandEdinburgh
respectively.
Table 3 shows in comparison the optimum inverter size as cal-
culatedby simulationandby the analytical method(AM) presented
here. It is evident from Table 3 that the analytical method is very
accurate incalculating the optimuminverter size. It shouldbe men-
tionedthat, inpractical situations, the designengineer couldhardly
select precisely the inverter size due to the following reasons. The
rst reason is that PV plants are usually designed with multiple
inverters instead of a single one for reliability reasons. Thus, the
total rated power of the installed inverters will be an integer mul-
tiple of the rated power of the respective commercially available
unit. The second reason is that the grid operator requires a bal-
anced three-phase connection of the PV plant, which means that
the number of inverters in the case that single-phase inverters
are used should be an integer multiple of three. For example,
let us assume a 100kW
p
PV plant in Iraklion, Greece, and suppose
that the design engineer selects the SMA 11kW inverter (the rated
dc power is 11.4kW). According to Table 2 and Fig. 4, the opti-
mum inverter size is P
opt
inv,N
= 2.00 P
max
= 2 72 = 144kW which
means that 144/11.4=12.6 inverters are required. Since the spe-
cic inverter is single-phased, the engineer must select either 12
or 15 inverters in order to design a balanced three-phase system.
In the rst case, the P
inv,N
/P
max
ratio will be 1.9 while in the sec-
ond case it will be 2.37. Since the annual energy yield does not
vary signicantly around the optimuminverter size (as can be seen
from Fig. 7), 12 inverters should be selected in order to reduce the
initial capital cost. This means that practically the design will set-
tle to the ratio P
inv,N
/P
max
=1.9 instead of the optimum which is
2.0. Such practical restrictions make the errors in the estimation
of the optimum inverter size, which appear in Table 2, insigni-
cant.
Table 4 shows, in comparison, the maximumac energy injected
into the grid as calculated by simulation and by the analytical
method using Eq. (11) with T
max
=4400h for all cases.
It can be seen from Table 4 that the energy estimated by the
analytical method matches within an 1.42.2% error the energy
estimated by simulation. Therefore, Eq. (11) can predict, with good
accuracy, the maximum ac energy yield.
The calculations with the analytical method can be signicantly
reduced by making the following observations: The rst one is
that the actual values of T
max
vary from 4375h in Cape Town to
4423h in Edinburgh. Therefore, the assumption that T
max
is con-
Table 4
Maximum energy injected into the ac grid, in kWh, as calculated by simulation and
by the analytical method (AM) for a 10kWp PV installation.
Iraklion, Greece Milano, Italy Edinburgh, UK
Simulation AM Simulation AM Simulation AM
Inv1 15.208 15.448 12.198 12.444 9.662 9.869
Inv2 15.302 15.519 12.274 12.501 9.720 9.915
Inv3 14.809 15.036 11.883 12.113 9.404 9.606
1204 C. Demoulias / Electric Power Systems Research 80 (2010) 11971204
Fig. 8. Measured (a) and estimated (b), dc power duration curves for a 20kWp grid-
connected PV installation in Korinos, Greece (40
18
N, 22
36
E).
stant for every location and equal to 4400h, introduces an error
of less than 0.6%. The second observation is that since the PDC can
be approximated very well with a straight line, the only param-
eter needed to dene the duration curve is P
max
. The latter can
be easily found from the maximum solar irradiance and the ambi-
ent temperature at that irradiance, using (1). Therefore, it is not
necessary to build the PDC in order to nd P
max
. Simply search-
ing in [9], or in other climate databases, for the maximum solar
irradiance which is usually during May or June for northern
hemisphere locations and the respective ambient temperature
is sufcient.
Thus, with the analytical method presented here, only four
parameters (P
max
, A, B and C) need to be determined in order to
calculate the optimuminverter size and make a good estimation of
the average annual energy yield.
5.2. Comparison with measurements
A 19.98kW
p
, grid-connected PV installation in Korinos, Greece
(40
18
N, 22
36