Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
n
i1
d
2
i
n
i1
y
i;m
y
i;c
_ _
2
n
i1
y
i;m
a bx
i
_ _
2
8
By determining the values of a and b that minimise F, the best line can be found. The following method can be
used to determine these parameters. Firstly, the following are obtained:
oF
oa
2
n
i1
y
i;m
a bx
i
_ _
0 9
oF
ob
2
n
i1
y
i;m
a bx
i
_ _
x
i
0 10
The following equations are then formed depending on a and b:
an b
n
i1
x
i
n
i1
y
i;m
11
a
n
i1
x
i
b
n
i1
x
2
i
n
i1
x
i
y
i;m
12
m
H
c m
H H =
i,m
H
'
m
H
i,c
H
c i
H
,
c
H
d
i2
Fig. 2. The least squares regression for H
c
H
m
(calculated daily solar radiation measured daily solar radiation).
2444 M. Yorukoglu, A.N. Celik / Energy Conversion and Management 47 (2006) 24412450
Finally, the parameters are calculated as
a
n
i1
y
i;m
b
n
i1
x
i
n
y bx 13
b
n
n
i1
x
i
y
i;m
n
i1
x
i
n
i1
y
i;m
n
n
i1
x
2
i
n
i1
x
i
_ _
2
14
If it is a nonlinear regression, the unknown parameters can be calculated in a similar manner, forming as many
equations as the number of unknown parameters. The correlation coecient R is then dened by
R
r
2
x;y
r
x
r
y
XY
X
Y
X
2
X
2
_ _
n
Y
2
Y
2
_ _
_
X XY Y
_ _
X X
2
Y Y
2
_ 15
The correlation coecient R ranges between 1 and +1. The value of the correlation coecient is a measure
of the strength of the relationship. A correlation coecient of R = 0.50 indicates a stronger degree of linear
relationship than one of R = 0.40. Likewise, a correlation coecient of R = 0.50 shows a greater degree of
relationship than one of R = 0.40. Thus, a correlation coecient of zero (R = 0.0) indicates the absence of a
linear relationship and correlation coecients of R = +1.0 and 1.0 indicate a perfect linear relationship. The
coecient of determination is,
R
2
1
n
i1
y
i;m
y
i;c
n
i1
y
i;m
y
m
2
16
Applying Eq. (16) to Fig. 1, the following is obtained:
R
2
1
1
n
i1
H
i;m
H
i;o
H
i;c
H
i;o
_ _
2
n
i1
H
i;m
H
i;o
H
m
H
o
_ _ _ _
2
1
RMSE
2
1
r
2
1
17a
Applying Eq. (16) to Fig. 2, the following is obtained:
R
2
2
1
n
i1
H
i;m
H
i;c
2
n
i1
H
i;m
H
m
_ _
2
1
RMSE
2
2
r
2
2
17b
Expressing R
2
as in Eqs. (17a) and (17b) contributes to understanding of the relationships between the param-
eters; that is R
2
is a function of RMSE and r. A further deduction is that R
2
increases as RMSE decreases or
vice versa because r
2
is constant for a particular data set. Similar to Eqs. (17a) and (17b), applying the def-
initions of the parameters of d
i
, RMSE, MBE, MABE, MPE and MAPE, as described in Refs. [1113], to
Figs. 1 and 2, Table 1 is obtained. As can be seen in Table 1, MPE and MAPE remain the same, while RMSE,
MBE and MABE change with the data set used. The subscripts of 1 and 2 denote the data set from which the
parameters are derived.
4. Critical review
An index of the researchers and the parameters that they used for assessing the goodness of an estimation
model or the estimation of daily or monthly global solar radiation from sunshine duration are presented in
Table 2. It is noted in Table 2 that some of the researchers drew the statistical parameters from the
H/H
o
S/S
o
data set, for example, Elagib and Mansel [11] and Elagib et al. [21], while most of the researchers
cited in Table 2 determined the statistical parameters from the H
c
H
m
data set, mostly regardless of what
their objective was. Note in Table 2 that a question mark ? was used for the cases where the name of the
parameter was declared by the authors but the formula of the parameter was not given. It was, therefore,
impossible to deduce from the explanations whether they were from the H/H
o
S/S
o
or the H
c
H
m
data
sets, as in Almorox and Hontario [14], Lin and Lu [17] and Hussain et al. [20]. Amongst these, however,
M. Yorukoglu, A.N. Celik / Energy Conversion and Management 47 (2006) 24412450 2445
Table 1
Formulae of d
i
, RMSE, MBE, MABE, MPE and MAPE for two dierent data sets
Data set 1 (Fig. 1) Data set 2 (Fig. 2)
Formula Unit Equation Formula Unit Equation
d
i1
H
i;m
H
i;o
H
i;c
H
i;o
(18a) d
i2
H
i;m
H
i;c
(MJ m
2
day
1
) (18b)
RMSE
1
1
n
n
i1
H
i;m
H
i;o
H
i;c
H
i;o
_ _
2
_
(19a)
RMSE
2
1
n
n
i1
H
i;m
H
i;c
2
(MJ m
2
day
1
) (19b)
MBE
1
1
n
n
i1
H
i;m
H
i;o
H
i;c
H
i;o
_ _
(20a)
MBE
2
1
n
n
i1
H
i;m
H
i;c
(MJ m
2
day
1
) (20b)
MABE
1
1
n
n
i1
H
i;m
H
i;o
H
i;c
H
i;o
_ _
(21a)
MABE
2
1
n
n
i1
H
i;m
H
i;c
_ _ (MJ m
2
day
1
) (21b)
MPE
1
1
n
n
i1
H
i;m
H
i;c
H
i;m
_ _
100
(22) MPE
2
MPE
1
(22)
MAPE
1
1
n
n
i1
H
i;m
H
i;c
H
i;m
_ _
100
(23) MAPE
2
MAPE
1
(23)
2
4
4
6
M
.
Y
o
r
u
k
o
g
l
u
,
A
.
N
.
C
e
l
i
k
/
E
n
e
r
g
y
C
o
n
v
e
r
s
i
o
n
a
n
d
M
a
n
a
g
e
m
e
n
t
4
7
(
2
0
0
6
)
2
4
4
1
2
4
5
0
deducing from the R
2
values presented, it is highly likely that Almorox and Hontario [14] drew the statistical
parameters from the data set of H/H
o
S/S
o
. The researchers that use the H/H
o
S/S
o
data set calculate the
statistical parameters that are the measure of the goodness of the estimation model. The researchers that use
the H
c
H
m
data set calculate the statistical parameters that do not represent the goodness of the model used,
but that of H
0
m
. Because these two sets of parameters are completely dierent, it is meaningless to compare
them. However, it is quite common in the literature to refer to the numerical values of a particular set of
parameters as the other. For example, Almorox and Hontoria [14] refer to the coecients calculated from
the H
c
H
m
data set in Refs. [19,20], while they themselves deal with the statistical parameters of the data
set of H/H
o
S/S
o
. This is because these two dierent sets of parameters are not distinguished properly.
As noted in Table 2, it is more common to calculate the statistical parameters based on the H
c
H
m
data
set in the literature. Elagib and Mansel [11] and Elagib et al. [21] correctly drew the statistical parameters from
the H/H
o
S/S
o
data set, as their objective was to study the estimation model itself (i.e. the relationship
between clearness index (H/H
o
) and relative sunshine duration (S/S
o
)). However, Ertekin and Yaldiz [13],
Chen et al. [16], Nguyen and Pryor [22] and Almorox et al. [23] drew the statistical parameters from the
H
c
H
m
data set, while their objective, which is deduced from the title of their articles, was to determine
the coecients of the Angstro mPrescott equation.
Further inconsistencies are encountered in the literature, as presented in Table 2. Firstly, a particular
parameter was named dierently by dierent researchers. For example, Chen et al. [16] named R
2
2
as NSE
(NashSutclie Equation). Almorox and Hontaria [14] named what are usually known as RMSE and MABE
as SEE and MAE, respectively. Tadros [18] denoted MAPE as e, dening it as the absolute percentage of the
deviation of the estimated measured values of global solar radiation. Secondly, the parameters were dened
incorrectly as in the case of Ampratwum and Dorvlo [12], who confused MPE and MAPE, and as in the case
of Togrul and Onat [15], who confused MBE and MPE. Another inconsistency is that a particular parameter
was given by a dierent formula that is dierent from the usually accepted denition. For example, Togrul and
Onat [15] described RMSE as
RMSE
3
1
n
n
i1
H
i;c
H
i;m
H
i;m
_ _
2
_
24
Table 2
M. Yorukoglu, A.N. Celik / Energy Conversion and Management 47 (2006) 24412450 2447
Nguyen and Pryor [22] described RMSE and MBE as
RMSE
4
% 100
1
H
m
n
i1
H
i;c
H
i;m
2
n
_ _
_
25
MBE
3
% 100
1
H
m
n
i1
H
i;c
H
i;m
n
26
Finally, some researchers used completely dierent parameters than the commonly used parameters, for exam-
ple, Ertekin and Yaldiz [13] and Almorox et al. [23]. Ertekin and Yaldiz [13] described a parameter called the
percentage error as follows to assess the goodness of the global irradiation prediction,
e
H
i;m
H
i;c
H
i;m
_ _ _ _
100 27
Almorox et al. [23] described a parameter called the t-statistic dened as
t n 1 MBE
2
= RMSE
2
MBE
2
_ _ _
0:5
28
Another kind of inconsistency was realised in the article by Ampratwum and Dorvlo [12], who compared six
dierent models (linear, quadratic, linearlogarithmic, logarithmic, power and power-trigonometric) for a to-
tal of ve stations in Oman using 10 years long monthly average daily solar radiation data. They analysed each
of the stations separately and concluded that the power-trigonometric model was the best amongst the models.
They assessed the goodness of the solar radiation estimations in terms of R
2
2
and RMSE
2
. It is, however, inter-
esting to note that they calculated R
2
2
0:7193 and RMSE
2
= 0.9936 MJ/m
2
for the linear model and
R
2
2
0:7666 and RMSE
2
= 1.0005 MJ/m
2
for the power-trigonometric model for the Marmul station. How-
ever, as was pointed out earlier in Eq. (17b), this is not logical, because R
2
2
cannot increase unless RMSE
2
de-
creases or vice versa.
5. A case study
The Angstro mPrescott equation has been widely used by researchers in solar radiation research since Pres-
cott developed it in 1940 [4]. Alternative models to the Angstro mPrescott equation have since been proposed.
In the present article, ve of the models, which were previously used by researchers commonly, are used to
estimate the daily global solar radiation from the measured duration of sunshine hours. The models used are
1. Linear model (Angstro mPrescott model) as given by Eq. (2).
2. Quadratic model (O
gelman H, Ecevit A, Tas demiroglu E. A new method for estimating solar radiation from bright sunshine data. Solar Energy
1984;33(6):61925.
[25] Samuel TDMA. Estimation of global radiation for Sri Lanka. Solar Energy 1991;47(5):3337.
2450 M. Yorukoglu, A.N. Celik / Energy Conversion and Management 47 (2006) 24412450