Вы находитесь на странице: 1из 9

Energy Conversion and Management 50 (2009) 1227–1235

Contents lists available at ScienceDirect

Energy Conversion and Management


journal homepage: www.elsevier.com/locate/enconman

Statistical comparison of monthly mean hourly and daily diffuse and global
solar irradiation models and a Simulink program development for
various Algerian climates
M. Koussa *, A. Malek, M. Haddadi
Renewable Energy Development Center, Postal Box 62, Observatory Road Bouzaréah, Algiers, Algeria

a r t i c l e i n f o a b s t r a c t

Article history: This work has two main objectives. The first objective is to carry out a statistical comparison of specific
Received 30 June 2008 models for estimating monthly mean daily and hourly global and diffuse irradiations on a horizontal sur-
Accepted 27 January 2009 face from the monthly average per day of the main meteorological data and to recommend the most
Available online 12 March 2009
accurate models to be used for estimating solar radiation in sites belonging to the Algerian climate areas
represented by each of the three sites presented in this work. The second objective is to propose a Sim-
Keywords: ulink program where the methodology allows for reconstituting daily and hourly monthly mean horizon-
Irradiation
tal solar radiation data. From the 10 global radiation models proposed in several journal papers,
Diffuse
Meteorological
Rietveld’s, Garg et al.’s and McCulloch’s relations are retained, respectively, for estimating the monthly
Bright sunshine hours mean daily global ratio for Adrar, Bouzareah and Ghardaia sites and Page and from the nine retained dif-
Global fuse relations, Page’s, Liu et al.’s and Erbs et al.’s relations are used, respectively, for the same sites for
calculating the monthly averages per day of diffuse irradiation. On the other hand, the models of Jain
and Liu et al. are, respectively, recommended for the reconstitution of the monthly averages per hour
of the global and diffuse irradiations. However, the obtained results are used to develop a Simulink
program which can be used in the estimation of long-term performance of flat plate solar radiation
systems.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction adjustment to allow for the seasonal progression path of the sun
across the sky, especially for the diffuse component measurement.
Data of global and diffuse solar radiation at a location are of sig- This is the main case in Algeria where few of the 35 existing mete-
nificant use in the design and in the long-term evaluation of the so- orological offices record the solar radiation data.
lar energy conversion systems performances. Furthermore, However, this paper has two objectives: the first consists in car-
monthly mean daily data are needed for the estimation of long- rying out statistical comparison of specific models for estimating
term solar systems performances, but for more precision and the horizontal monthly mean daily and hourly of global and diffuse
according to the considered system application, generally, scien- irradiations and retaining the most accurate models. The second
tists use the hourly data. Thus, for this, the required solar radiation one consists in developing a simple program in Simulink in which
component is the total solar irradiation incident on an inclined sur- the retained models are used to reconstitute the monthly mean
face according to the slope of the considered sensor. If this compo- hourly and daily diffuse and global irradiations from the monthly
nent was not available, then, the horizontal global and diffuse average per day of a minimum main meteorological data.
radiation data are the best source of information for its calculation. Many approaches have been proposed in several journals pa-
However, measurements of the main meteorological parameters pers and retained in our work to estimate monthly mean daily dif-
are generally available, but horizontal diffuse and global solar fuse and global irradiations as function of the main meteorological
radiations are only measured at few locations due to the cost of data. On the other hand, some approaches have been used to ob-
the equipment required, and due to the care needed for regular tain estimations of the monthly mean hourly global and diffuse
irradiations from monthly mean daily data. Furthermore in our
paper, we first performed a literature review of all the existing
models and we made a description of each retained model. This
was followed by a statistical comparison of the monthly mean
* Corresponding author. Tel.: +213 021 90 15 03; fax: +213 021 90 16 54.
E-mail addresses: mustaphakoussa@cder.dz (M. Koussa), amalek@cder.dz daily and hourly retained models to the measured data obtained
(A. Malek), mourad.haddadi@enp.edu.dz (M. Haddadi). from three Algerian sites.

0196-8904/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enconman.2009.01.035
1228 M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235

As results, for the three considered Algerian sites and for the where q, qa, and qc are, respectively, the ground, cloudless sky and
locations included in the same climatic zones, we propose, respec- cloud albedo. SS0o is the modified day length which excludes the
tively, Rietveld’s [1], Glover’s and McCulloch’s [2] and Garg’s and fraction of day where the solar zenith angle is greater than 85° be-
Garg’s [3] relations for reconstituting monthly mean total solar cause the Campbell–Stokes sunshine recorder used by most meteo-
energy for Adrar, Bouzareah and Ghardaia, and Liu’s and Jordan’s rological stations does not function when the solar elevation is less
[4], Erbs et al.’s [5] and Page’s [6] models for reconstituting the than 5°. The proposed modified day length can be obtained from
monthly mean daily diffuse solar energy for Bouzareah, Ghardaia  
2 cosð85Þ  sinðuÞ sinðdÞ
and Adar. Furthermore, Jain’s [7] and Liu’s and Jordan’s [4] models SS0o ¼ cos1 ð6Þ
15 cosðuÞ cosðdÞ
are retained to reconstitute, respectively, the monthly mean hourly
global and diffuse irradiations. In this way, a Simulink program has
been developed to calculate the monthly average per day and per 2.1.1.2. Global irradiation, relative humidity, sunshine hours and
hour of the horizontal global and diffuse irradiations from only temperature relations.
the main meteorological data. 2.1.1.2.1. Reddy’s [9] model. Reddy [9] proposed a relation that
takes into consideration some theoretical parameters such as lati-
2. Models description tude u, mean sun declination for the considered month, and the
maximum possible bright sunshine hours in addition to the main
2.1. Monthly mean daily models meteorological parameters such as the measured bright sunshine
hours, relative humidity and the number of rainy days in a month.
2.1.1. Total solar energy models The formula is as follows:
2.1.1.1. Global irradiation bright sunshine hours relations.   
DI ð1  0:2tÞ
2.1.1.1.1. Rietveld’s model [1]. Rietveld [1] examined several linear Gh ¼ K 1 þ 0:8 pffiffiffiffiffiffiffi ð7Þ
SSo HR
correlations developed in different locations in the world and relat-
ing the clearness index KT to the sunshine duration ratio (DI/SSo), where
was used, and noted that the a and b coefficients are, respectively, K ¼ 100ðkSSo þ wij cosð/ÞÞ ð8Þ
related linearly and hyperbolically to the appropriate mean values
of DI/SSo such that k is the latitude factor given by the following relation:
 
DI 1
a ¼ 0:10 þ 0:24 ð1Þ k¼ ð9Þ
SSo ð1 þ 0:1/Þ
 
SSo u is the latitude of the place in degrees, SSo is the theoretical day
b ¼ 0:38 þ 0:08 ð2Þ
DI length in hours, DI is the daily bright sunshine hours, and t is de-
fined by the following ratio:
and by substituting these two relations in the Angstrom equation
form he proposed the following equation: number of rainy days in the month
  t¼ ð10Þ
total number of days in the month
Gh DI
¼ 0:18 þ 0:62 ð3Þ HR is the mean relative humidity per day in the month, and wij
Gho SSo
is the seasonal factor (i = 1, 2 and 3) for inland stations, coastal sta-
This formulation is believed to be applicable anywhere in the world. tions and hilly stations, respectively, while (j = 1, 2, . . ., 12) stands
2.1.1.1.2. Glover’s and McCulloch’s [2] relation. In their model, for January–December. The wij values are listed out in Table 1.
attempting to improve on the total irradiation–sunshine correla- 2.1.1.2.2. Relation of Sayigh [10]. The model of Sayigh [10] was
tion, Glover and McCulloch [2] have included the latitude effects developed by using the data relating to several Indian sites whose
and presented the following formulation: latitude lies between 8° North and 28.6° North. Thus, Sayigh’s [10]
  model is expressed by the following relation:
Gh DI
¼ 0:29 cosðuÞ þ 0:52 ð4Þ       
Gho SSo DI HR 1
GH ¼ KN exp u   ð11Þ
where u is the latitude of the considered site, and this relation is va- SSo 15 T max
lid for u 6 60°. where
2.1.1.1.3. Hay’s [8] model. The coefficients a and b of the linear cor- N ¼ 1:7  0:458u ð12Þ
relation form relating the global irradiation to the sunshine duration
are affected by the optical properties of the atmosphere constitu- K, DI, SSo, and HR were previously defined, and Tmax is the monthly
ents, ground reflectivity and average air mass. Garg and Garg [3] maximum dry temperature. In Eq. (12), the latitude u of the consid-
developed the second type of equation relating the global irradiation ered site is expressed in radians. Furthermore, Sayigh [10] subdi-
to the sunshine duration, where he took into account the considered vides the entire range of relative humidity into three subranges:
meteorological factors and proposed the following expression: HR 6 65%, HR P 70% and 65% < HR < 70%, and corresponding to
h i each subrange, a graph has been drawn between the humidity fac-
Gh 0:1572 þ 0:5566 SSDI0 tor wij and each month of the year.
o
¼      ð5Þ 2.1.1.2.3. Relations of Swartman and Ogunlade[11]. Swartman and
Gho 1  q q DI0 þ q 1  DI0
a SS c
o SS o Ogunlade [11] have proposed two empirical equations. These

Table 1
Values of relative humidity factors wij as function of the month number and humidity subranges 1, 2, 3.

i/j 1 2 3 4 5 6 7 8 9 10 11 12
1 1.28 1.38 1.54 1.77 2.05 2.30 2.48 2.41 2.36 1.73 1.38 1.17
2 1.46 1.77 2.05 2.15 2.05 2.05 2.10 2.17 2.14 1.96 1.60 1.43
3 1.60 1.81 2.00 2.17 2.25 2.26 2.24 2.20 2.10 1.92 1.74 1.60
M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235 1229

formulas predict global radiation on a horizontal surface from For a given global irradiation Gh, the multiple reflections phe-
bright sunshine hours and relative humidity only. The two rela- nomena increase the received global irradiation, which is quanti-
tions are given as follows: fied by G0h and is obtained from the following equation:
     
DI DI DI
Gh1 ¼ 490 0:357ðHRÞ0:262 ð13Þ G0h ¼ Gh 1  q qa 0 þ qc 1  0 ð21Þ
12 SSo SSo
  
DI
Gh2 ¼ 460 exp 0:607  HR ð14Þ
12 and the diffuse irradiation obtained after the multiple reflection
2.1.1.2.4. Relation of Garg and Garg [3]. Garg and Garg [3] showed phenomena is calculated by using the correlation in following:
that the daily global irradiation data averaged over each month
for 14 Indian measurement stations can be correlated with sun- D0h
¼ 0:9702 þ 1:6688K 0T  21:303ðK 0T Þ2 þ 51:288ðK 0T Þ3
shine duration and atmospheric water vapor content. They found G0h
that a least squares fit of data gives the following relationship:  50:08ðK oT Þ4 þ 17:551ðK oT Þ5 ð22Þ
   
DI G0
Gh ¼ Gh0 0:414  0:4  0:0055Ha ð15Þ where K 0T ¼ G h , and finally Dh can be calculated from the following
SSo ho
relation:
where DI is the sunshine duration, SSo is the day length, and Ha is   
the atmospheric water vapor content per unit volume of dry air DI DI
Dh ¼ D0h þ Gh q qa 0 þ qc 1  0 ð23Þ
which may be computed from the following expression: SSo SSo
 
Ha ¼ HR 4:7923 þ 0:3647T a þ 0:0055T 2a þ 0:0003T 3a ð16Þ
2.1.2.2. Diffuse radiation, sunshine hours and atmospheric water con-
Ta is the dry air temperature in degree Celsius. tent relation.
2.1.1.2.5. Relation of Hussain [12]. Following Garg’s and Garg’s [3] 2.1.2.2.1. Relation of Hussain [12]. In his same paper, Hussain [12]
model, Hussain [12] proposed an empirical expression. Thus, this proposed an expression relating the diffuse irradiation to the abso-
expression makes it possible to estimate the total solar energy as lute humidity and the bright sunshine hours, this relation is as
function of monthly average per day of the absolute humidity follows:
and the monthly average per day of bright sunshine duration. This    
DI
relation is given as follows: Dh ¼ Gh0 0:306  0:165  0:0025Ha ð24Þ
    SSo
DI
Gh ¼ Gh0 0:394 þ 0:364  0:0035Ha ð17Þ
SSo
2.1.2.3. Diffuse radiation and clearness index relations.
DI, Gho, SSo and Ha were previously described. 2.1.2.3.1. Model of Liu and Jordan [4]. Liu and Jordan [4] first
2.1.1.2.6. Relation of Sambo [13]. Based on the main meteorological reported the correlation between the fraction of monthly mean
parameters such as sunshine duration DI, minimum and maximum diffuse to monthly mean total solar energy KD as function of the
daily dry temperatures Tmax and Tmin, relative humidity HR, where monthly mean clearness index KT and gave
all are monthly averaged values and the monthly mean per day
of the theoretical values of day duration SSo, Sambo [13] developed K D ¼ 1:39  4:027K T þ 5:531K 2T  3:108K 3T ð25Þ
a correlation given as follows:
    For 0.30 < KT 6 0.70
Gh DI DI T min
¼ 0:62  0:294 þ 0:178  HR  2.1.2.3.2. Page’s model [6]. Page [6] using data from 10 widely
Gho SSo SSo T max
  spread sites in 40° North to 40° South latitude belt proposed this
DI T min general expression:
þ 0:491 ð18Þ
SSo T max
K D ¼ 1:0  1:13K T ð26Þ
2.1.2. Reconstitution of the monthly mean daily diffuse solar energy
2.1.2.1. Diffuse radiation and sunshine hours relations. 2.1.2.3.3. Collares-Pereira’s and Rabl’s model [16]. Collares-Pereira
2.1.2.1.1. Relations of Iqbal [14]. Iqbal [14] proposed two correla- and Rabl [16] used measured data from five United States loca-
tions where the first one is in a linear form which requires mea- tions, and to test for seasonal variations they grouped the data into
sured global irradiation and the sunshine hours data as input. In bins according to their sunset hours angles xs. And the least
his second proposed correlation, Iqbal [14] removed the global squares fit yields the correlation
irradiation requirement and expressed a correlation which re-
quires only sunshine duration. The two proposed relations are gi-  p h  pi
ven in the following: K D ¼ 0:775 þ 0:347 xs   0:505 þ 0:261 xs 
 2 2
Dh DI  cos ð2ðK T  0:9Þ ð27Þ
¼ 0:791  0:635 ð19Þ
Gh SSo
 2
Dh DI DI According to the yearly variation of the sunset angle hour, the
¼ 0:163 þ 0:478  0:655 ð20Þ
Go SSo SSo data are grouped as follows:

2.1.2.1.2. Hay’s [15] relation. In his work, Hay [15] considered the
increasing magnitude of the global and diffuse solar radiations xs  p2  0:15 for winter season
after multiple reflections between the earth and the cloud cover. p  0:15  xs  2 p  0:15 for spring and fall season
2
The relationships between the radiation before and after multiple p  0:15
reflections are given below. xs  2 for Summer season
1230 M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235

2.1.2.3.4. Relation of Erbs et al. [5]. Erbs et al. [5] used a data set for a ¼ 0:409 þ 0:5016 sinðxs  60Þ ð36:aÞ
four United State stations and developed two third–order seasonal b ¼ 0:6609  0:4767 sinðxs  60Þ ð36:bÞ
polynomial correlations for monthly diffuse fraction as function of
clearness index and gave where x is hour angle in degrees calculated in the mid-point of the
considered hour and xs is the sunset hour angle in degrees calcu-
K D ¼ 1:391  3:560K T þ 4:189K 2T  2:137K 3T ð28-aÞ lated by
for xs 6 1.4208 and 0.30 6 KT 6 0.80 and xs ¼ cos1 ðtgðuÞtgðdÞÞ ð37Þ
K D ¼ 1:311  3:022K T þ 3:427K 2T  1:821K 3T ð28-bÞ where u is the latitude of the considered site and d is the solar dec-
lination calculated for the representative day of the month.
for xs > 1.4208 and 0.30 6 KT 6 0.80, where xs is a sunset hour an-
gle in radians.
2.2.2. Diffuse irradiation model
2.1.2.3.5. Iqbal’s relation [17]. Iqbal [17] used measured data
2.2.2.1. Model of Jain [7]. In the same paper, Jain proposed the fol-
relative to several Canadian sites and developed the following
lowing relation for calculating the monthly mean hourly ratio in
relation:
daily diffuse irradiation rd as function of solar time and the
K D ¼ 0:958  0:982K T for 0:30 6 K T 6 0:60 ð29Þ monthly mean theoretical day length:
" #
1 ðt  12Þ2
2.2. The hourly models rd ¼ pffiffiffiffiffiffi exp  ð38Þ
2pr 2r2

2.2.1. Global irradiation model So, to simplify the calculation of the r values, especially for the
2.2.1.1. Model of Jain [7]. In his work, Jain [7] used 11 years monthly sites where the hourly diffuse irradiation data are not available, he
average values of the ratio hourly to daily of global rt as well as the proposed the following correlation:
diffuse irradiations rd on a horizontal plan data relating to several
Canadian sites and had shown that the distribution of these ratios
r ¼ 0:222SSo þ 0:27 ð39Þ
follows a Gaussian law form. However, he established the follow-
ing relation for global irradiation: 2.2.2.2. Liu’s and Jordan’s [4] equation. As for the monthly mean
" # hourly global irradiation, in their work, Liu and Jordan [4] showed
2
1 ðt s  12Þ
r t ¼ pffiffiffiffiffiffi exp  ð30Þ a related set of curves for rd to solar hour angle and the sunset hour
2pr 2r 2 angle
where ts is the true solar time in hours, and r is defined by dh
rd ¼ ¼ f ðx; xs Þ ð40Þ
1 Dh
r ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð31Þ
2pkTh ðt ¼ 12Þ and the proposed chart curves have been converted to the following
expression:
where rt (ts = 12) is the monthly mean hourly ratio in the monthly  
average per day of the global irradiation at mid-day true solar time. p cosðxÞ  cosðxs Þ
rd ¼ ð41Þ
To simplify the determination of r values, Jain [7] established a cor- 24 sinðxs Þ  px
180
s
cosðxs Þ
relation between r and theoretical day length SSo, and proposed the
following relation:
3. Used data
r ¼ 0:2SSo þ 0:378 ð32Þ
The retained formulas were tested for different Algerian
SSo can be calculated by the following expression: locations where Bouzareah was considered as a coastal site, and
Ghardaïa and Adrar were considered as dry sites and arid sites.
2
SSo ¼ cos1 ð tan gðuÞ tan gðdÞÞ ð33Þ Thus, the geographical co-ordinates of these sites are listed out
15 in Table 2.
u and d are, respectively, the latitude of the considered site and the For the three considered sites, monthly mean daily sunshine
solar declination. duration hours, relative humidity, minimum and maximum tem-
perature data were obtained from National Office of Meteorology.
2.2.1.2. Model of Collares-Pereira and Rabl [16]. Using monthly aver- The used monthly mean daily and hourly global and diffuse irradi-
age data of solar radiation measurement from five United States ation data were obtained from three recording data stations
stations and based on Whillier’s [18] and Hottel’s and Whiller’s installed, respectively, in the Renewable Energy Applications
[19] works, Liu and Jordan [4] have established a generalized chart Center at Ghardaia site, in Saharan Renewable Energy Research
of rt as function of day length and the hour angle, and presented Center at Adrar site and in the Center of Renewable Energies Devel-
the chart curves in the following form: opment at Bouzareah site. It should be noted that 5 min and 1 h
are, respectively, used as recording step time for radiometric
gh parameters at Ghardaïa site and Bouzaréah and Adrar sites.
rt ¼ ¼ f ðx; xs Þ ð34Þ
Gh However, all measured diffuse and total solar radiation data are
averaged to monthly mean hourly and daily values.
However, Collares-Pereira and Rabl [16] used the least squares
fit and presented the proposed chart curves in a semi empirical Table 2
form and gave the following expression: Geographical co-ordinates of the considered sites.
  Location Latitude in degrees Longitude in degrees Height in meters
p cosðxÞ  cosðxs Þ
rt ¼ a þ b cosðxÞ ð35Þ
24 sinðxs Þ  pxs
180
cosðxs Þ Bouzaréah 36.8N 3.08E 345
Adrar 27.82N 0.18W 263.9
Ghardaia 32.4N 3.80E 468.4
yields the coefficients given by
M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235 1231

4. Method of statistical comparison 4.4. The normalized mean bias error

There are numerous works in literature journals which deal


MBE
with the assessment and comparison of solar radiation estimation NMBE½% ¼ 1 Pn  100 ð45Þ
models [20–23]. The most popular statistical parameters indicators n i¼1 Gi;meas

are the mean bias error, MBE, and the root mean square error, This test provides information on long-term performance. A low
RMSE. In this study, to evaluate the accuracy of the calculated data, NMBE value is desired. A negative value gives the average amount
from the models described above, we used three statistical tests, of underestimation in the calculated value. So, one drawback of
the mean percentage error, MPE, the mean bias error, MBE, and these two mentioned tests is that overestimation of an individual
the root mean square error, RMSE. observation will cancel underestimation in a separate observation.

4.1. The mean percentage error 4.5. The normalized root mean square error

n  
1X Gi;calc  Gi;meas RMSE
MPE ¼  100 ð42Þ NRMSE½% ¼ 1 Pn  100 ð46Þ
n 1 Gi;meas i¼1 Gi;meas
n
where Gi,meas, Gi,cal and n are, respectively, the ith measured values
and ith calculated values of global or diffuse solar radiation and
5. Results and discussion
the number of values.
5.1. Monthly mean daily models
4.2. The mean bias error
The performances of the 10 global and of the nine diffuse solar
1Xn
radiation models were evaluated using the three statistical indica-
MBE ¼ ðGi;calc  Gi;meas Þ ð43Þ
n 1 tors, i.e MPE, MBE and the RMSE, discussed early. The obtained re-
sults are presented in Tables 3 and 4. Furthermore, as presented in
the corresponding literature, relations (7), (11), (13), and (14) are
4.3. The root mean square error expressed in British units system, and in our work the results are
" #12 converted into Wh/m2 (international unit system).
1X n
RMSE ¼ ðGi;calc  Gi;meas Þ2 ð44Þ 5.1.1. Total solar radiation
n i¼1
From results presented in Table 3, we note that generally Rietveld
[1], Garg and Garg [3], Glover and McCulloch [2] have used more or
So, to evaluate the performance of the hourly models we used less comparable levels of different statistical indicators in this work.
the MPE and the normalized mean bias error, NMBE, and root So, their correlations yield, respectively, the best results for Adrar,
mean square error, NRMSE. Ghardaia and Bouzareah. Therefore, relations (3), (15) and (4) are

Table 3
MPE (%), MBE (kWh/m2) and RMSE (kWh/m2) values between measured and estimated total solar irradiations.

Adrar Ghardaia Bouzareah


MPE (%) MBE (kWh/m2) RMSE (kWh/m2) MPT (%) MBE (kWh/m2) RMSE (%) MPE (%) MBE (kWh/m2) RMSE (kWh/m2)
Rietveld’s Eq. (3) 0.57 0.005 0.212 3.22 0.162 0.346 4.15 0.177 0.270
McCulloch’s Eq. (4) 0.058 0.035 0.215 5.27 0.277 0.394 2.47 0.085 0.225
Hay’s Eq. (5) 0.41 0.020 0.237 2.94 0.160 0.337 5.47 0.225 0.303
Sayigh’s Eq. (11) 8.35 0.473 0.837 10.3 0.556 .974 7.4 0.219 1.072
Reddy’s Eq. (7) 27.01 2.892 3.800 15.6 0.969 1.443 12.4 0.619 1.292
Swartman’s Eq. (13) 21.26 1.529 1.607 13.6 0.814 1.196 14.1 0.637 1.420
Swartman’s Eq. (14) 19.25 1.330 1.477 14.7 0.896 1.200 10.7 0.444 1.352
Garg’s Eq. (15) 3.89 0.212 0.242 0.4 0.020 0.218 8.4 0.358 0.411
Hussain’s Eq. (17) 1.3 0.111 0.215 4.8 0.266 0.349 5.3 0.210 0.294
Sambo’s Eq. (18) 17.8 0.817 .864 22.5 0.952 1.012 9.7 0.462 0.562

Table 4
MPE (%), MBE (kWh/m2) and RMSE (kWh/m2) values between measured and estimated diffuse solar irradiations.

Bouzareah Ghardaia Adrar


MPE (%) MBE (kWh/m2) RMSE (kWh/m2) MPE (%) MBE (kWh/m2) RMSE (kWh/m2) MPE (%) MBE (kWh/m2) RMSE (kWh/m2)
Iqbal’s Eq. (19) 10.88 0.195 0.249 14.18 0.240 0.293 12.21 0.232 0.285
Iqbal’s Eq. (29) 3.12 0.008 0.151 23.05 0.246 0.347 32.94 0.315 0.336
Hay’s Eq. (23) 4.30 0.005 0.137 14.94 0.259 0.289 15.36 0.281 0.313
Hussain’s Eq. (2) 12.4 0.252 0.294 15.4 0.269 0.303 13.6 0.246 0.276
Iqbal’s Eq. (20) 17.1 0.328 0.355 11.7 0.198 0.221 16.7 0.294 0.311
Page’s Eq. (26) 8.4 0.139 0.193 11.8 0.133 0.147 3.8 0.048 0.125
Collares P.’s Eq. (27) 26.1 0.277 0.353 22.0 0.441 0.509 22.3 0.476 0.562
Liu’s & Jordan’s Eq. (25) 4.4 0.075 0.147 10.8 0.125 0.130 4.9 0.057 0.099
Erbs et al.’s Eqs. (28.a & b) 7.6 0.141 0.185 0.9 0.05 0.112 6.8 0.123 0.172
1232 M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235

recommended respectively to be use for calculating the monthly perform well at the considered sites, and with their universal
mean horizontal global solar radiation, for Adrar, Ghardaia and applicability, their use is recommended for estimating the monthly
Bouzareah and for all sites included in the same climatic zones. average per day of diffuse irradiation, respectively, for Bouzareah
Hay’s [8] relation is preferred best. Furthermore, monthly val- and Ghardaia.
ues of ground and atmospheric albedo are needed. These values Page’s [6] relations derivation was based, respectively, on data
are based on Canadian atmospheric conditions. So, they appear from locations within latitude u = ±40°. So it is the next best model,
to perform well at the considered Algerian stations. especially for Adrar, but the correlation results in larger errors than
Hussain’s [12] models result in larger errors than Hay’s [8] rela- Liu’s and Jordan’s [4] and Erbs et al.’s [5] relations for Bouzareah
tion for Adrar and Bouzareah sites, but are better for the Ghardaia and Ghardaia where the MPEs and MBEs are positive and negative,
site. respectively, for coastal sites and arid sites. Its simplicity and uni-
The above five models results can be explained by the fact that versal applicability should be emphasized upon. However, this
these models are in an empirical form and are based on the main model is recommended to reconstitute the monthly mean daily
weather parameters such as relative humidity, dry temperature diffuse irradiation for Adrar location and is applied for all sites in
and monthly average daily of bright sunshine and on some astro- the same climatic zone.
nomical parameters such as solar constant, earth sun distance var- Iqbal’s [17] relation (29) and Hay’s [15] equation produce re-
iation, the monthly mean daily extraterrestrial solar energy and sults very close to each other. They have more or less comparable
theoretical bright sunshine duration. Thus, the latitude and sea- levels of MBEs and RMSEs. They present the best results for Bouz-
sonal effects are simplified by the use of monthly average of the areah location, but produce the largest RMSEs, MBEs and MPEs for
various astronomical and geographical parameters, and on the the two Saharian sites. Furthermore, these correlations were estab-
other hand, the climatic zones effects are represented by the mea- lished by using data from Canadian site, and to apply Hay’s [15]
sured and used weather parameters in these two models. equation the monthly values of ground and atmospheric albedo
The model proposed by Sayigh [10] is the next best one with the are needed. However, they appear to perform very well at the
smallest range of the MPE, MBE and RMSE values for Adrar and coastal site, but not at the arid sites. This is due to the fact that
Ghardaia considered as dry and arid locations. The negative values the ground and the atmospheric albedo and the climatic parame-
of MPE and MBE show that there is a constant underestimation. So, ters are totally different and the diffuse radiation is directly related
the magnitude of the three statistic indicators is generally larger to these parameters. So this was not observed in the relation (5) re-
for Bouzareah, which is considered as a coastal site. sults where it was mentioned that Hay’s [8] relation performs well
Reddy’s [9] equation and Swartman’s and Ogunlade’s [11] two at all the considered sites. Thus, the atmospheric diffusion and the
models result in larger errors than the previous correlations, and multiple reflections have an inverse effect on each of the two dif-
the MPEs are positive, showing that there is constant overestima- fuse and direct solar radiation components. Furthermore, these
tion. As mentioned previously, it is clearly observed that these two phenomena increase and decreases, respectively, the diffuse
models overestimate the total solar radiation throughout the year, and the direct components, and their effect cannot appear clearly
and it is more apparent for result of Reddy’s [9] model, especially on the global radiation component.
for Ghardaia and Adrar locations.
Sayigh’s [10], Swartman’s and Ogunlade’s [11] and Reddy’s [9] 5.2. Monthly mean hourly models
relation results can be explained by the fact that the derivation
of these models is based on data from Indian locations character- 5.2.1. Global irradiation models
ized principally by the humidity factor wij and the number of rainy To make a comparison between the models, the calculated and
days which depend strongly on the considered sites climate and measured data were compared for each month, and a statistical
can be different from those of the three considered Algerian sites, summary of the monthly and overall performance of the combina-
especially those of coastal site. tion of the different test indicators discussed previously as MPE,
NMBE and the NRMSE is, respectively, presented in Tables 5–7
5.1.2. Diffuse irradiation models and 8–10 for the hourly global and diffuse irradiations.
The performance of the nine diffuse solar radiation correlations For the hourly global irradiation, the results presented in Tables
was evaluated following the same procedure. The results of 5–7 show that Jain’s [7] model generally leads to the best results.
the comparison of the different retained models are presented in For the three considered sites, the NRMSE values obtained by using
Table 4. this model do not exceed 12% except for October in Adrar location
From the results presented in Table 4, we note that Liu’s and where it is equal to 16.42% (see Table 6). Even though its derivation
Jordan’s [4] correlation yields the best results in term of MBEs was based on data from Canada locations, this equation appears to
and RMSEs. perform well at the considered Algerian stations.
Erbs et al.’s [5] model is the next best one. The low MBEs are Collares-Pereira’s and Rabl’s [16] model resulted in largest
particularly remarkable for the three considered sites. Even though NRMSE with the highest value close, respectively, to 18.58% and
the Liu and Jordan [4] and Erbs et al. [5] correlations were estab- 18.18% for April at Ghardaia location (see Table 5) and for October
lished by using data from the United State sites, they appear to at Adrar location (see Table 6). Furthermore, the monthly values of

Table 5
Monthly and overall performance of hourly global irradiation models (Ghardaia site).

January February March April May June July August September October November December
MPE (%) Eq. (35) 0.4 5.9 10.1 9.0 7.0 12.1 9.6 4.3 8.8 6.5 16.0 16.6
Eq. (30) 3.2 4.1 1.4 0.7 2.2 1.8 0.5 0.5 1.6 0.5 2.8 1.02
NMBE (%) Eq. (35) .77 1.88 6.02 11.89 7.00 8.82 7.67 4.69 5.70 7.55 10.51 11.32
Eq. (30) .04 4.30 0.91 1.19 3.03 2.03 1.00 1.31 2.44 0.09 2.03 0.65
NMRSE Eq. (35) 6.88 8.63 7.39 18.58 13.55 6.95 8.98 10.04 6.50 11.82 12.31 12.52
Eq. (30) 8.21 5.95 4.70 8.20 6.17 6.40 8.13 8.27 4.50 3.75 4.84 7.73
M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235 1233

Table 6
Monthly and overall performance of hourly global irradiation models (Adrar site).

January February March April May June July August September October November December
MPE (%) Eq. (35) 2.9 5.3 7.3 4.1 1.3 4.4 13.2 10.9 8.3 14.2 5.4 10.9
Eq. (30) 10.5 6.5 16.0 7.2 4.6 1.9 1.1 1.8 4.2 1.4 0.3 4.5
NMBE (%) Eq. (35) 2.27 3.50 5.83 0.87 1.70 0.55 9.17 9.15 3.78 8.57 4.35 8.07
Eq. (30) 5.97 5.41 2.76 2.73 4.24 0.66 3.04 1.89 2.79 3.49 0.73 0.32
NMRSE (%) Eq. (35) 7.05 8.28 7.34 6.47 6.34 12.13 14.24 13.36 5.50 18.18 9.73 10.23
Eq. (30) 9.47 8.82 7.15 6.71 7.44 4.28 9.93 8.76 3.72 16.42 6.87 7.69

Table 7
Monthly and overall performance of hourly global irradiation models (Bouzareah site).

January February March April May June July August September October November December
MPE (%) Eq. (35) 0.3 2.7 2.6 9.2 0.6 0.9 6.2 6.5 0.2 21.0 4.82 10.1
Eq. (30) 10.1 9.8 9.7 0..9 0.8 4.5 1.9 13.7 1.6 0.5 4.1 3.6
NMBE (%) Eq. (35) 0.99 3.63 3.62 .52 8.67 0.52 2.60 2.50 0.21 11.21 3.76 4.12
Eq. (30) 7.86 6.63 6.62 1.10 0.65 2.16 1.28 1.81 0.30 1.40 3.39 2.28
NMRSE (%) Eq. (35) 6.32 6.50 6.44 4.87 10.36 11.80 5.84 7.11 7.75 12.71 12.53 7.84
Eq. (30) 9.03 11.78 11.88 2.29 4.36 6.02 4.09 5.62 7.03 5.08 9.03 5.14

Table 8
Monthly and overall performance of hourly diffuse radiation models (Ghardaia site).

January February March April May June July August September October November December
MPE (%) Eq. (41) 9.4 10.9 0.6 9.3 7.6 3.5 4.3 15.3 3.1 8.6 6.5 5.8
Eq. (38) 17.0 11.6 5.9 7.5 10.8 2.8 0.6 17.8 10.1 11.6 14.4 13.3
NMBE (%) Eq. (41) 11.10 11.66 0.73 8.97 10.41 2.15 2.05 12.64 7.40 9.26 9.24 5.30
Eq. (38) 16.50 15.43 6.01 12.44 14.63 2.99 2.97 16.38 12.48 13.56 14.77 11.01
NMRE (%) Eq. (41) 15.52 14.97 7.86 20.30 12.63 11.16 7.73 14.83 11.05 12.45 13.38 8.64
Eq. (38) 19.15 19.43 12.2 24.45 18.33 13.36 13.88 19.26 16.76 17.35 18.78 13.48

Table 9
Monthly and overall performance of hourly diffuse radiation models (Adrar site).

January February March April May June July August September October November December
MPE (%) Eq. (41) 7.0 5.3 1.5 0.4 0.5 1.0 6.7 1.1 0.7 22.8 9.1 0.8
Eq. (38) 1.5 2.3 5.8 7.4 3.5 5.0 3.3 2.1 6.7 26.2 0.3 7.6
NMBE (%) Eq. (41) 6.46 5.84 3.14 3.02 1.51 8.22 3.73 6.49 1.48 18.23 9.39 0.24
Eq. (38) 0.01 0.99 8.31 8.42 5.90 12.64 1.14 10.27 6.87 12.43 2.78 5.86
NMRSE (%) Eq. (41) 12.04 12.60 10.27 7.58 12.33 17.74 7.29 10.69 7.15 22.55 11.76 4.06
Eq. (38) 15.50 15.46 12.15 12.87 13.87 22.59 11.11 15.81 12.23 23.13 12.24 9.09

Table 10
Monthly and overall performance of hourly diffuse radiation models (Bouzareah site).

January February March April May June July August September October November December
MPE (%) Eq. (41) 0.8 3.3 0.3 5.5 2.0 1.3 2.4 2.6 1.9 2.8 0.1 0.7
Eq. (38) 6.7 5.2 6.8 6.4 3.2 7.6 3.8 5.1 9.1 1.1 7.7 6.2
NMBE (%) Eq. (41) 2.64 2.45 0.23 6.48 1.95 1.06 2.65 4.19 1.87 2.79 0.27 0.13
Eq. (38) 3.49 3.73 5.07 2.20 3.06 6.20 2.58 8.52 7.27 1.86 6.31 5.92
NMRSE (%) Eq. (41) 6.19 5.66 7.46 11.92 5.92 6.21 5.89 6.77 5.98 7.42 8.41 4.92
Eq. (38) 11.11 10.83 10.34 14.90 11.34 11.83 9.95 13.31 11.96 10.00 13.95 10.65

the a (Eq. (36.a)) and b (Eq. (36.b)) are based on Hump Mountain low NMBEs are particularly remarkable. So, the negative NMBEs
location in United States of America, and probably cannot perform values presented in Table 5 show that there is an underestimation
very well for the Algerian stations. for Bouzareah location (see Table 7) during the period January–
In addition, the NMBEs show the long-term performance of the May and October–November and during May, August and October
two models. Jain’s [7] relation generally yields the best results. The for Adrar location (see Table 6).
1234 M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235

Fig. 1. Matlab-Simulink program description.

Collares-Pereira’s and Rabl’s [16] equation presents NMBEs val- radiation. Furthermore, in terms of MPE indicators, generally Liu’s
ues higher than those obtained by Jain’s [7] model. Even though and Jordan’s [4] model leads to better performances.
without exceeding 10%, these are sometimes positive and some-
times negative over the whole year and for the three considered 6. Simulink program
sites.
Also, since the MPE indicator provides information on the size Using the obtained results, a Simulink program was developed
of the discrepancy in the measurements and quantifies the system- following the methodology described above:
atic component of the normalized difference for an individual
observation, the results show that generally the magnitude of 1. using the theoretical data, as:
MPE presented by Jain [7] is the smallest for the three considered – the considered site latitude;
sites, especially for Ghardaia location. – the representative day number for each month;
– all required theoretical parameters are calculated;
5.2.2. Diffuse irradiation models 2. using sunshine hours duration as input in relation3, 4, and 15
The two monthly mean hourly diffuse relations performances the monthly mean daily global irradiation is calculated for
were evaluated following the same procedure. The resulting MPEs, Adrar, Bouzareah or Ghardaia;
NMBEs and NRMSEs are presented in Table 8–10. 3. the calculated global irradiation are used, respectively, in
The NMBEs results show the long-term performance of the two relation (26), (25) and (28) to obtain the monthly mean daily
models. Jain’s [7] model presents an underestimation for almost all diffuse irradiation for Adrar, Bouzareah and Ghardaia;
the considered sites and months, except for February, July and 4. the monthly mean daily diffuse and global irradiation are
October–November for Adrar (see Table 9), and for April for Bouz- used as input, respectively, in relations (30) and (41) to calcu-
areah (see Table 10). On the other hand, for the three locations (see late the monthly mean hourly global and diffuse irradiation.
Tables 8–10), Liu’s and Jordan’s [4] relation results are the best, and
the low NMBEs are particulary remarkable for Bouzareah, which do The obtained Simulink program is presented in Fig. 1.
not exceed 7% over the whole year.
The NRMSE provide information on the short-term performance 7. Conclusions
of a model by allowing a term by term comparison of actual differ-
ence between the estimated and measured values and generally First, we can affirm that for any given site, the direct use of a
the lower NMRSE value correspond to the better model0 s perfor- model suggested in the literature can lead to erroneous values,
mance. However, from the results presented in Tables 8–10 it will and consequently can influence the dimensioning of the solar en-
be noted that Liu’s and Jordan’s [4] equation yields the best results. ergy conversion systems considerably. However, the choice of the
So, a larger magnitude of the NMRSE was obtained over almost the models strongly depends on the climatic characteristics of the con-
whole year for Ghardaia and Adrar considered as arid sites than sidered site compared to those on which its application is being
those obtained for Bouzareah considered as a coastal location. This considered. This was observed from results obtained by Eqs. (7),
can be explained by the fact that Ghardaïa and Adrar sites are sit- (11), (13), (14), and (18) for the monthly mean global solar radia-
uated, respectively, in the north and in the center of Sahara, where tion, from those obtained by Eqs. (20), (24), and (27) for the
the spring and autumn seasons are characterized by the wind sand monthly mean daily diffuse irradiation and by Eqs. (34) and (38)
phenomena which considerably increase the diffusion of the solar for the monthly average data per hour.
M. Koussa et al. / Energy Conversion and Management 50 (2009) 1227–1235 1235

However, the 10 monthly mean daily global irradiation correla- [3] Garg HP, Garg SN. Prediction of global solar radiation from bright sunshine
hours and other meteorological data. Energy Convers Manage 1983;23:113–8.
tions were compared statistically in terms of mean percentage, the
[4] Liu BYH, Jordan RC. The interrelationship and characteristic distribution of,
mean bias and root mean square errors. Rietveld’s [1], Glover’s and direct, diffuse and total solar radiation. Solar Energy 1960;4:1–19.
McCulloch’s [2] and Garg’s and Garg’s [3] correlations are the most [5] Erbs DG, Klein SA, Duffie JA. Estimation of the diffuse radiation fraction for
accurate in general. Only the sunshine hours, dry air temperature hourly, daily and monthly-average global radiation. Solar Energy
1982;28:293–302.
and relative humidity data available in different meteorological [6] Page JK. The estimation of monthly mean values of daily total short wave
offices are, respectively, needed as input in the first two models radiation on vertical and inclined surfaces from Sunshine records for latitudes
and in the third model. Their use is recommended for estimating 40°N–40°S. In: Proceeding UN conference of new source of energy. Paper no-
S98, vol 4, 1961. p. 378–90.
the horizontal monthly average per day of the global radiation, [7] Jain PC. Estimation of monthly average hourly global and diffuse irradiation.
respectively, for Adrar, Bouzareah and Ghardaia and for all other Solar Wind Technol 1988;5:7–14.
sites in the same climatic zone. [8] Hay E. Calculation of monthly mean solar radiation for horizontal and inclined
surfaces. Solar Energy 1979;23:301–7.
The nine monthly mean daily diffuse correlations were compared [9] Reddy SJ. An empirical method for the estimation of total solar radiation. Solar
following the procedure as described above. Liu’s and Jordan’s [4], Energy 1971;13:289–94.
Erbs et al.’s [5] and Page’s [6] correlations are the most accurate in [10] Sayigh AAM. Solar energy engineering. London: Academic Press, INC LTD;
1977.
general and are simple to use, especially the Page [6] correlation. Only [11] Swartman RK, Ogunlade O. Solar radiation estimates from common
global solar radiation is needed as input data. These three correlations parameters. Solar Energy 1967;11:170–2.
are recommended for estimating the horizontal monthly average per [12] Hussain M. Estimation of global and diffuse irradiation from sunshine duration
and atmospheric water vapor content. Solar Energy 1984;33:217–20.
day diffuse irradiation, respectively, Bouzareh, Ghardaia, Adrar.
[13] Sambo AS. Empirical models for the correlation of global solar radiation with
On the other hand, the two hourly global radiation correlations meteorological data for northern Nigeria. Solar Wind Technol 1986;3:89–93.
and the two diffuse equations were compared statistically in terms [14] Iqbal M. Correlation of average diffuse and beam radiation with hours bright
of mean percentage, the normalized mean bias and the normalized sunshine. Solar Energy 1979;23:169–73.
[15] Hay JE. A revised method for determining the direct and diffuse components of
root mean square errors. However, relation (30) and relation (41) total short wave radiation. Atmosphere 1976;14:278–87.
are, respectively, the most accurate in general to estimate the [16] Collares-Pereira Manuel, Rabl Ari. The average distribution of solar radiation –
monthly mean daily of global and diffuse irradiations. Further- correlations between diffuse and hemispherical and between daily and hourly
insolation values. Solar Energy 1979;22:155–64.
more, if only the monthly mean of the main meteorological param- [17] Iqbal M. A study of Canadian diffuse and total solar radiation data, I. Monthly
eters is available, we can calculate the monthly daily and hourly average daily horizontal radiation. Solar Energy 1979;22:81–6.
global and diffuse irradiations on a horizontal surface with a good [18] Whiller A. Solar radiation graphs. Solar Energy 1965;9:164–5.
[19] Hottel HC, Whiller A. Evaluation of flat plate collector performance.
accuracy. In this way, a Simulink program was developed to calcu- Transaction of the conference on the use of solar energy, Part I., vol.
late the monthly mean daily and hourly global and diffuse data 2. University of Arizona Press; 1955. p. 74.
from only monthly mean daily sunshine hours duration, dry tem- [20] Bahel V, Srinivsan R, Bakhsh H. Statistical comparison of correlations for
estimating of global horizontal solar radiation. Energy 1987;12:1309–16.
perature and relative humidity. [21] Svozil K, Neufeld N, Zeroual A, Ankrim M, Wilkinson AJ. The diffuse-global
correlation: its application to estimating solar radiation on tilted surface in
References Marrakech, Morocco. Renew Energy 1996;7:1–13.
[22] Ma CCY, Iqbal M. Statistical comparison of solar radiation correlations,
monthly average global and diffuse radiation on horizontal surfaces. Solar
[1] Rietveld MR. A new method for estimating the regression coefficients in the
Energy 1984;33:143–8.
formula relating solar radiation to sunshine. Agricult Meteorol 1978;19:243–52.
[23] Stone RJ. Improved statistical procedure for the evaluation of solar radiation
[2] Glover J, McCulloch JS. The empirical relation between solar radiation and
estimation models. Solar Energy 1993;50:247–58.
hours of sunshine. J Roy Me, Soc 1958;84:172–5.

Вам также может понравиться