Escolar Documentos
Profissional Documentos
Cultura Documentos
Solar Energy Materials and Solar Cells 26 (1992) 51-69 and Solar Cells
North-Holland
The reliability of stand alone photovoltaic systems is analyzed in terms of the loss of load
probability, LLP. A variety of numerical and analytic models for calculating the LLP are
described and evaluated using data for three Spanish locations: Madrid, Murcia and Santander,
selected because they represent different climatic conditions. It is concluded that numerical
models are accurate but complex to use, while analytic models exhibit significant lack of
accuracy.
A new analytic model, as accurate as the numerical models and as simple as analytic models, is
proposed. For each location, the model requires as input 4 different coefficients.
1. Introduction
rAG C
CA L and Cs = L-' (1)
where A is the PV-array area, r/ is the PV-array efficiency, (~ is the mean daily
irradiation on the PV-array, L is the mean daily energy consumption and C is the
useful accumulator capacity.
Note that C A depends on the meteorological conditions of the location. That
means that the same PV-array for the same load can be "large" in one site and
"small" in another site with lower solar radiation.
Given a location and a load, two general ideas are intuitive: First, it is possible
to find many different combinations of C A and C s leading to the same LLP value.
Second, the larger the PV-system size is, the greater the cost and the lower the
LLP.
The task of sizing a PV-system consists of finding the better trade-off between
cost and reliability. Very often, the reliability is an a priori requirement from the
user, and the PV-engineer problem is formulated as follows: Which pair of C A and
C s values leads to a given LLP value at the minimum cost? To solve this problem,
this paper presents a critical review of different methods, that have been referred
in the literature and proposes a new one that attempts to be accurate and simple
to use.
Special attention is paid to the study of the relation between C A, C s and LLP.
This p a p e r does not deal with the estimation of the PV-system cost. In fact, the
last is a classical problem of economics and the PV-sizer can opt for any of the
many methods that are found in the literature [1].
Because long term averages of daily irradiation incident on surfaces other than
horizontal are not generally available, it is useful to define a new p a r a m e t e r
~Ad(0) G(0)
c~i- L CA d ' (2)
where Gdm and Gdm(0) are, respectively, the monthly averages of the tilt and
horizontal daily irradiation. Moreover, it has been shown [2] that the location
dependence of the results can be reduced by correlating the collector area to the
irradiation worst month, defined as the one showing the lowest relation between
solar irradiation and energy consumption. For simplicity, in this paper, we are
concerned with loads exhibiting a constant daily energy consumption during the
whole year. Then, the worst month is the one corresponding to the lowest value of
Gdm(0). Hence, in the following we use for C A and CA the lowest values of CA,~
and C~,m, respectively. Other cases can be easily derived from this one.
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 53
Each point of the C~,-C s plane represents a size of a PV-system. This allows to
map the reliability as fig. 1 shows. The lines are the loci of all the points
corresponding to a same LLP value. Because of that, we call them isoreliability
lines.
Note that the definition of C A and C s implies that this map is independent of
the load and depends only on the meteorological behavior of the location. We will
see that the isoreliability lines are, very nearly, hyperbola with their asymptotes
parallel to the x and y axis, respectively.
On the other hand, given a LLP value, the plot of the cost of the PV-systems
corresponding to the isoreliability line is, approximately, a parabola having a
minimum that defines the optimal solution to the sizing problem (fig. 2).
This section reviews the methodologies for estimating the relation between CA,
C s and LLP, that, as far as we know, have been referred to in the literature. For
explanation purposes, we divided them into three different groups:
.~ LLP = 0.01
. . . . . . . . . . .
Storage Capacity
Fig. 1. Reliabilitymap.
54 M. Egido, E. Lorenzo / The sizing of stand alone PV-systems
-- ~ 0
\ I)
S O C j = m i n ( S O C ~ _ 1 +~TAGa/_C;1) (s)
Table 1
Typical values of Fsl and Fs2
Fsl / Fs2 Application
Domestic Telecomunications
North of Spain 1.2/5 1.3/7
South of Spain 1.1/4 1.2/5
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 55
BATTERY LOAD
AUXILIARY
GENERATOR
Fig. 3. Schematic of the photovoltaic system used for simulation.
or
where Gdi is the daily irradiation over the collector surface on the day j and C, is
the nominal capacity of the accumulator. It is calculated from the corresponding
horizontal value using the algorithm proposed by Hay and McKay [3]. Further-
more, let us assume that the auxiliary generator is managed in such a way that, at
the end of the day j, it fulfills the battery if the stored energy is lower than the load
requirements. Then,
SOCj > / 1 / C s ~ EAUXj = 0 (7)
and
SOCj ~ 1 / C s ~ EAuxj = (1 -- SOCk) L / C s and SOCj = 1, (8)
where EAuxj is the energy supplied by the auxiliary generator in the day j.
If the simulation is carried out over a great number of days, N , in order to be
statistically meaningful, then the LLP value corresponding to the stand alone
PV-systems (i.e. the system of fig. 3 excluded the auxiliary generator) is given by
LLP - ~2N=1EAUXj
NL (9)
Eq. (5) implies the use of a constant PV-array efficiency. The validity of such
assumption for the present purposes, is supported by the results of Ambrosone [4],
who found the same LLP values using daily timesteps and r / = constant as they did
using hourly timesteps and considering the r/ dependence on solar irradiance,
ambient temperature and battery state of charge, particularly for C s > 2. Some
authors [5,6] have presented results supporting the opposite idea, i.e. that simula-
tions in hourly basis always lead to more accurate LLP values than in daily basis.
56 M. Egido, E. Lorenzo / The sizing of stand alone PV-systems
Table 2
Parameters of the historical horizontal daily irradiation sequences: Gdy(0) is the yearly mean and
Gdm(0), Ktm , and ~1 are the monthly mean of the irradiation, the clearness factor and the first order
autoregressive parameter, respectively
Gd v(0) Gd m(0) Kt m ~,
(kWh/m 2) (kWh/m 2)
Murcia 4.713 2.108 0.510 0.334
Madrid 4.568 1.463 0.398 0.405
Santander 3.183 1.100 0.290 0.241
Nevertheless, a careful analysis of such results discloses that the differences are
due to the use of different and not exactly equivalent transposition models from
horizontal to tilted radiation, in such a way that the daily irradiation on the
PV-array is different when they calculate in hourly and in daily basis. This is a
[2 0 5 0
E3
CL
",, ~.
~ 060 \
..
\
"'-b.
Q~045 --. ~.
(D \ -
O5O
" 0 40
<~040
<ass
030 050
0 2 5
.. .. .. .. .. .. .. .. .. ~AI . . . . . . . ~ ......... ~'2. . . . . . . . % ........ ~......... ~ ......... ,'2'
Storage Capacity Storage Capacity
LPP = 0 0 1 , MAORID
200 ................................. "~
Historical Date
[~ - - Aguior's Model
...... Oraham's Model ( m )
- -- Graham's M o d e l
150
\\\
)100
<< 0 5 0 ~-~ ~
collateral effect that cannot be attributed to the timestep selected for simulation
and that can be taken into account in more simple ways.
Since the simulation is run for a long time period, the initial state of the battery
has no effect. Moreover, as Gordon [7] noted, the assumption of a 24 h distributed
load leads to the same LLP value as that obtained from our nighttime load if C s is
reduced by 1.
These methods offer the advantage of the accuracy and the capability of
improvements by the incorporation of more complex models of the elements of the
system. This way, they are useful for studying many other aspects in addition to the
sizing; for example, the benefits from the use of maximum power trackers, etc.
Some algorithms to help the size optimization have been suggested [8].
The drawbacks are the large computing time and the need of long sequences of
daily irradiation data, as input. Motivated by the rather scarce availability of such
sequences of data, several authors have investigated models to generate synthetic
radiation sequences, from widely available information, usually monthly mean
i i i , , i J , i~ . . . . . . . . . [ . . . . . . . . . I .... i i I i i
040 8 12 16 035 ........ ~. . . . . . . . . . 8 ......... I'2 . . . . . . . . . 16
Storage Capacity Storage Capacity
- - H~storicel Dote
- - Agulor's M o d e l
~/ . . . . Grahem's Model ( m )
1 O0 - - - Grohern's M o d e l
Ca- ,
0 ",.. \ x \
>~060
0.40
values. The generated sequences keep some statistical properties whose validity is
assumed to be general. In particular, the persistence of solar radiation, i.e. the
dependence of today's solar irradiation on the solar irradiation of the precedent
days is adequately described by a first-order autoregressive process [9,10]. More-
over, the probability function of the clearness index for any given period has a
form associated with only its average value for that period [11,13].
To analyze the validity of such models, we selected two of them as representa-
tives and we compared the LLP values obtained from the simulation described by
eqs. (4)-(6) using as input sequences of real daily irradiation data from several
Spanish sites and sequences generated by the models. In particular, we selected
Murcia, Madrid and Santander located, respectively, at the south, center and north
of Spain. Table 2 shows some representative parameters of the corresponding
historical sequences of data. The different values of ~bI indicate that the selected
locations are representative of different types of climate.
In fact, this way of validation is limited by the length of reference data
sequences. Klein and Beckman [14] showed that with the at present available
historical data ( ~ 20 years) only LLP > 10 -2 can be validated. Below this value
there is a need to rely on extrapolations. The reason for that is the natural
variability associated to the climate. That is, equally probable series of 20 years
solar radiation data, can lead to different LLP values for LLP < 10 2. See also ref.
[10] for a good discussion on this.
Note that this way of validation identifies the characteristics of the data that
influence the PV-system performance rather than simply matching statistical
properties.
The first model, proposed by Graham [14], produces sequences of daily clear-
ness index, K,, using a simple autoregressive model of the form
X N = ~J1XN 1 -~ WN' (10)
where X u is related to K t by the transformation function given by
[1 + e r f ( X N / f 2 ) ] =F(K,, F,,), (11)
where F ( K t, K,t) is the well known probability distribution function for K r WN is
an uncorrelated Gaussian random number having a mean of 0 and a variance of 1
and bl is the first-order autoregressive parameter. The input requirements of this
model are the values of Gdm(O)and ~bl. If the latter is not available, an average
value of 0.29 is recommended.
The second model, proposed by Aguiar [15], also produces sequences of K t,
using a library of transition matrices, each corresponding to a specific interval in
K,. From the analysis of observed sequences of data in several stations, the authors
of this model propose a library which validity is supposed to be universal. The only
input requirement are the values of Gdm(O).
Figs. 4, 5 and 6 show some results. We believe that both models lead to an
adequate degree of accuracy. As expected, the larger differences between C~
values from the models and from the historical data correspond to very little C s
values, generally far from that can be considered as practical. Moreover, the
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 59
LPP = 0.1, SANTANDER LPP = 0 0 5 , SANTANDER
0.70 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
<2 <
030 O40
LPP = 0 0 1 , SANTANDER
200
"\ \ \
\
- -
- - - -
Historical Data
Aguiar's M o d e l
Graham's Model (m)
i, \ \ - Graham's M o d e l
150 ', \
h \ \
\, \ \
', \
',..~ \
Ck
O 1 DO
(3
< 050
Storage Capacity
Fig. 6. Comparison of isoreliability lines in Santander from historical and from generated data.
250
- - Historical Data
.4~_ 2 O0 -- -- Monegon's Model
0
CL
Q~)I 5 0
~ 1O0
<
050
available, it can be obtained from the mean according to the procedure described
in ref. [8]. In fig. 7 the results of this model are compared with the simulation for a
LLP = 0.01, in the case of Madrid. It has a good accuracy for storage capacity
greater than 4. The principal disadvantage of this model is that use of the error
function with an iterative process, requires a large computing time.
Barra et al. [16] considered the fraction fo the energy load covered by the
PV-system.
Y = 1 - EAux/L (12)
and they assume that its monthly average, Ym, relates with the size of the
PV-system through the formula
In fact, this represents very logical limit conditions. The first means that, for
small size PV-systems, all the energy produced by the PV-array is transferred to
the load. The second comes from the consideration that, for very large field areas,
the energy supplied by the array is always able to satisfy the load.
From the analysis of Ym values obtained by simulations performed with histori-
cal sequences of irradiation data corresponding to several Italian locations, the
authors of the model proposed for Ym the following expression which validity is
assumed to be universal:
Obviously
m-12
LLP = ~ Y'. (1 - Ym). (16)
m=l
Later on, Bartoli et al. [17] proposed a different expression to relate y and Cs:
aAmKtm
Cs = ( b + Amgtm) 2' (17)
where
a = al exp(-Ym/31) + 1 and b = a2y~2. (18)
Again, the analysis of several sequences of historical data, leads to the proposi-
tion of the supposed universal values:
a I =0.695, ~2 = 0.274, /31 = 71.2, /32= 0.136.
06
\
~06
CL Cl
(3 0
LD (D
>~04 ;~0.4
3
02 02
O.O ....... ~, . . . . . . . . . 8 ......... 1~2. . . . . . . 16 00 ........ ~ ......... & ......... ~'t . . . . . . ~,s
Storage Capacity Storage Capacity
- - Historical Data
16 . . . . . Barro's M e t h o d
- - -- Bortoii's M e t h o d
- -- Bucciarelli's Method
",\
G_
O
L~ X ,,,.. I /
>~08
{3
t \ . . . . . . . . .
04
I; h
065 - - H i s t o r i c a l Data - - Historical Data
...... B a r r o ' s Method 07 {II ..... Berra's Method
Bartoli's M e t h o d ", -- Bortoli's Method
Bucclorelli's Method
~060 /L ', -- BucciorellPs Method
,,
(0 ",..
,, "0~06
Ci0.55 eL
I ",,
(D o
(J (J \ ---.. .............
050
>~05
o
~045 \\ . . . . . . .
o4
040
LPP = 0 0 1 , MuRCIA
10 .... " 'l ........
I - - Historical Data
Borra's Meihod
O9 \ '~t _-_-~.%t~%'L,,iMg%%%od
',
~08 I
",,
g_
(~O7
~\ "'" -....
>-,
~DG
<
05
Bucciarelli [18,19] presents a model for the LLP derived by approximating the
probability density function of the difference betwen the daily PV-array output and
the load with two events and by assuming the daily storage charge/discharge
process can be represented as a one-step Markov process. A similar model to that
of the Bucciarelli but using a three-event representation of the difference between
the daily output and the load was presented by Gordon [7], later on. The inputs to
this models are the long term mean, Gd,., the variance, G and the first autocorrela-
tion parameter, 4~1 of the daily solar irradiation impinging on the PV-array. If the
value of ~bl is not available a universal coefficient equal to 0.29 is recommended.
Figs. 8, 9 and 10 show the results of these models for the locations of reference.
All the models lead to differences between their predictions and the results from
historical data that exceed the expected variability in some part of the C s range of
interest. Barra's model systematically leads to oversizing. The model of Bartoli
exhibits a too low sensibility to the battery size, that is, changes of the C s value,
while keeping constant the LLP, do not reflect changes of C~. Finally, the models
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 63
LPP = 0 1, SANTANDER LPP = 0 0 5 , SANTANOER
10 ,,i ................................ 1.0
'I'I
II
[11 - - Historicol D a t a 09 - - Historical Data
--- Barra's M e t h o d ", -- Barra's M e t h o d
O8 ',,
[Ii -- -- Bartoli's M e t h o d --- Bartoli's M e t h o d
', -- Bucciarelli's Method - Buccle rell;'s Method
>~ /I ",,
~06
o_
'-..
g~o~
o O ~\ "-.
<D (.9
o6
o
~o5
02
LPP = 0 0 1 , SANTANDER
20 ,i ................ ~ ................
i
I ", - - H i s t o r i c a l Data
16 ', ..... Borra's M e t h o d
I , Bortoli's M e t h o d
',- -- Bucciore,i's Method
~12 !s ","',,
C/_ \ I
;~08
04
of Bucciarelli and Gordon display very much alike behavior (because of that, the
last has not been represented). They lead, for C), values larger than about 1.3, to
the absurdity that both the array and the battery sizes have to increase simultane-
ously in order to keep the LLP constant.
010 , , [ ,,,
: ,
1
: ;
008 _:
006
L~
i
(D t ,
;!:
>
L~ 004 I ' : i i,
0 ." , , I .:
: I:
0 O2 i :'
00o . . . . . . . . . . . . . "i;
" !
...... t i +~
000I ~ ~OOOl 2 5001 ~ Ol
toss of Load Probabilitie
Fig. 11. The Loss of load Probability versus the geometrical regression error.
(b) All the analytical models proposed up to today allow sizing of pv-systems in
a very simple way by means of straightforward hand calculations. However, they
exhibit significant lack of accuracy which disfavors their general use.
The motivation of our work has been the development of a sizing procedure as
accurate as the numerical models and as simple as the analytical models. To that
end, we have done an in depth analysis of the solar radiation based on Spanish
data [20] and we mapped the reliability for 42 different locations and for 11 LLP
values uniformly distributed between 0.5 and 5 10 - 4 . S o , 462 isoreliable curves
have been studied. We found all of them can be well represented by the expression
CA =mCs n, (19)
where m and n are depend on the LLP value and on the location. Fig. 11 shows
the relative error due to this geometrical regression for all the curves. The mean
value of the coefficient of determination, R e, is 0.98 and it varies between a
minimum of 0.93 and a maximum of 0.999.
For a given LLP, m can be defined as the C A value corresponding to C s = 1. n
takes into account the effect of the persistence of the solar radiation. Note that C s
decreases when n increases, while keeping constant all the other parameters.
Moreover, we found that for each location, the values of m and n are related
whit LLP, through the simple regressions
/~_--/ 4 ~ / -L....... )
.iX f "
/ / /'oo: ; ) ~ (0)\
~ ~ o~ / J--oj3 ,L 'l
/ ) ~_/" / Cordoba / ---~mrcla/ ,
~-f~'~-"-q /--'-~-~,
j__~" /Almer.ia'/L ",
N coefficient (I,PP 0 1 )
v S tan
/ ~oo o ,~<>--4 -
I / / ;LoJ~--~., ~.z" ,u
L)/~ ,., /" // Aliean~F$-
/ ' "" ~t I rcJ~ 7
/ / .----" Co.eba / "5
l/ / / ,,oivs , ~ o o~ - - / / -- --
Table 3
T h e r e g r e s s i o n c o e f f i c i e n t s o f eq. ( 2 0 )
mI m2 nl n2 R2n R~
Madrid -0.2169 -0.7865 - 1.2138 - 15.280 93.72 99.80
Murcia 0.0483 - 0.9684 - 1.1329 - 36.415 93.78 99.95
Santander - 0.2026 - 0.7527 - 0.9759 - 14.289 94.81 99.18
\ / ~, (
/./-.bo.,,lo " \ / b
i , - ,///
N coefficient (DPP 0 . 0 1 )
? -_../ ....
17~"-
/ t tt\k"- \.
F i g . 13. C o e f f i c i e n t S p a n i s h m a p s f o r L L P = 10 - 2 .
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 67
2.00
1 60
.~ 1 20
0
.::..- ..
. ,
(0) 0 8 0
040
I i l i . . . . . . I I , I I I I I I I l I I I I I I L I~
0 0[~
n 20 0.40 0.60 080
Clearness Lndex of the Worst Month (Kt)
variation of m versus Ktm. It is clear that the use of any correlation linking both
parameters leads to significant accuracy losses Similar comments merit all the
correlation we analyzed between m or n with other radiation parameters (Gdm,
etc.) and between m and n. This research is still being undertaken by including
other widely available meteorological data like the number of rainy days, etc.
Meanwhile, the present results suggest that the location dependence of the
PV-size method cannot be neglected without paying a significant price in terms of
accuracy. This agrees with the fact that the monthly clearness index over a period
of 20 years shows different distributions for the different locations
Example. Estimate the PV-array area, A, needed to obtain a LLP = 0.05 in
Sevilla in supplying a constant load of 500 W h / d a y and for a battery wich useful
capacity is 1500 Wh. The monthly average daily horizontal irradiation in Sevilla is
2.154 k W h / d a y for the worst month (December).
600 . . . . . . . . ,
Q_
(~) 3 . 0 0
200
~Y
I O0
000
OQO01 0 001 0.01
Loss of Load Probabilitie
Fig. 15. C A versus LLP predicted by analytic models for LLP < 10 -2.
68 M. Egido, E. Lorenzo / The sizing of stand alone PV-systems
From figs. 12 and 13 we obtain the values: m = 0.434, n = 0.02 for LLP = 10 -1
and m = 0.98, n = 0.208 for LLP = 10 -2. Eq. (20) allows then to calculate m I =
-0.112, m 2 = - 0 . 5 4 6 , n 1 = - 1 . 3 1 and n 2 = - 2 6 . 0 2 . For L L P = 0 . 0 5 again eq.
(20) leads to m = 0.598 and n = 0.073.
A value of C A = 0.552 is then calculated from eq. (19). Finally, assuming an
overall system efficiency of 10%, eq. (2) lead to A = 1.281 m 2.
6. Conclusions
An analytic model for accurate sizing stand alone photovoltaic systems has been
developed. For each location, the model requires as input 4 different coefficients
that are obtained from the study of large sequences of historical or generated daily
irradiation data.
Maps of such coefficients have been presented for Spain. From these maps, the
task of PV-sizing becomes very simple and can be accomplished by means of
straightforward hand calculations.
On the other hand, it has been shown that other previous analytic models
exhibit significant lacks of accuracy which disfavors their use.
References
[1] H.L Macomber, I.B. Ruzek, F.A. Costello et al., Photovoltaic Stand-Alone Systems: Preliminary
Engineering Design, Handbook prepared for NASA, L RC, Contract DEN 3-195, August 1981.
M. Egido, E. Lorenzo / The sizing of stand alone PV-systems 69
[2] R.N. Chapman, Design Consideration for Stand-Alone Photovoltaic Systems, Pro. Symp. on
Applications of Solar and Renewable Energy-86, Cairo, Egypt, March 1986
[3] S.E. Hay and D.C. McKay, Int. J. Sol. Energy 3 (1985) 203.
[4] G. Ambrosone, S. Catalonotti, U. Coscia and G. Troise, Sol. Energy 34 (1985) 1.
[5] LL. Mora, M. Egido and M. Sidrach, Utilization of Simulated Hourly Global Radiation Series for
Sizing Stand-Alone Pbotovoltaic Power Systems, Proc. of the 9th E.C. Photovoltaic Solar Energy
Conference, Freiburg, 1989, p. 583.
[6] F. Lasnier and T.G. Ang, Photovoltaic Engineering Handbook, (Adam Hilger, Bristel, 1990).
[7] J.M. Gordon, Sol. Cells 20 (1987) 295.
[8] PH. Tsalides and A. Thanailakis, Sol. Cells 18 (1986) 115.
[9] B.J. Brinkworth, Sol. Energy 19 (1977) 343.
[10] V.A. Graham, K.G.T. Hollands and T.E. Unny, Sol. Energy 40 (1988) 83.
[11] B.Y.H. Liu and R.C. Jordan, Sol. Energy 4 (1960) 1.
[12] P. Bendt, M. Collares-Pereira and A. Rabl, Sol. Energy 27 (1981) 1.
[13] K.G.T. Hollands and R.G. Huget, Sol. Energy 30 (1983) 195.
[14] S.A. Klein and W.A. Beckman, Sol. Energy 39 (1987) 499.
[15] R.J. Aguiar, M. Collares-Pereira and J.P. Conde, Sol. Energy 40 (1988).
[16] L. Barra, S. Catalanotti, F. Fontana and F. Lavorante, Sol. Energy 33 (1984) 509.
[17] B. Bartoli, V. Cuomo, F. Fontana, C. Serio and V. Silverstrini, Appl. Energy 18 (1984) 37.
[18] L.L. Bucciarelli, Sol. Energy 32 (1984) 205.
[19] L.L. Bucciarelli, Sol. Energy 36 (1986) 11.
[20] Radiaci6n Solar en Espafia: Afios 1972 a 1984. Centro de Estudios Meteorol6gicos. Ministerio de
Transportes, Turismo y Comunicaciones.