Power Output Forecasting of a Solar House by Considering Different Cell

Temperature Methods

1
¨
Ozge ˘ ¨
Ayvazogluyuksel ¨
, Ümmuhan Başaran Filik2

Department of Electrical and Electronics Engineering,

Anadolu University, Eskişehir
1
ozgeayvazogluyuksel@anadolu.edu.tr, 2 ubasaran@anadolu.edu.tr

Abstract Undoubtedly, it is assumed that solar panels operate under

ideal conditions during the manufacturing process of PV panels.
Forecasting of power generation is needed for accurate de- However, this situation is not valid when dynamic change of the
sign and performance evaluation of solar energy systems outdoor parameters are considered. In addition, measurement of
to associate demand and source side dynamics efficiently. the PV cell temperature is not accessible in many systems. Due
Since the power output values of solar energy systems are to these reasons, in literature, cell temperature estimation meth-
significantly affected by the cell temperature, estimation of ods are presented to forecast power output values of the solar
cell temperature has gathered wide interest in recent years. energy systems. Standard model which considers only global
In this study, cell temperature values of the on-grid photo- solar radiation and ambient temperature is developed in [5]. In
voltaic panels of a solar house placed in Anadolu University [4], Mattei models are presented to obtain cell temperature with
¨ Campus are estimated by using six different mod-
İki Eylul defining two different parametrizations of heat exchange coeffi-
els. In addition, power output values of the system are fore- cient. Skoplaki models are obtained in [6] by integrating wind
casted with three different models by using the estimated data in standard model on the basis of two different descriptions
cell temperature values, measured outdoor parameters and of wind convection coefficient. In [7], Kurtz model is defined,
panel specifications. Therefore, the most accurate models which does not consider PV specifications. The model only
for cell temperature estimation and power forecasting are considers outdoor parameters as global solar radiation, ambi-
determined according to the results of statistical test analy- ent temperature and wind speed. A simple emprical model pro-
sis methods. posed in [8] is used and some constants are described depending
on PV technologies with Koehl model in [9]. In [10], Muzathik
1. Introduction model is developed as a function of wind speed, global solar
radiation and ambient temperature.
With increasing concern about dependence on fossil fuels Accurate information on power generation of a PV system
and environmental issues, alternative energy solutions have come is essential for planning and projecting of a solar system in dif-
into prominence in recent years [1]. Since energy is one of the ferent environmental conditions. With the help of forecasting
basic factors of economic and social developments, it is crucial power generation, an efficient energy analysis can be achieved
to meet required energy demand efficiently to contribute to the to meet the required energy demand. Hence, power generation
development of the countries. At that point, fossil fuels are not forecasting methods have gathered wide interest in literature.
sufficient due to their environmental disadvantages and limited These methods are divided into three main groups as physical,
lifetime [2]. In place of fossil fuels, solar energy is mostly statistical and hybrid methods [11]. In physical models, output
recommended to be used in electricity generation with its power values are defined as a function of global solar radiation,
durability, abundance and cleanliness. Associated with the use ambient temperature and some other outdoor parameters. In
of the solar energy in power system applications, photovoltaic addition to the meteorological parameters, solar cell properties
systems (PV) are placed in appropriate regions to benefit from such as cell temperature plays an important role on the perfor-
the sun in the most efficient ways. mance evaluation of a PV system by using physical models.
PV cell temperature is an important parameter that directly As a second group, statistical models are based on the concept
affects the performance of solar cells. This temperature depends of persistence or stochastic time series. Within the scope of
on many parameters such as outdoor conditions, climatic location these models, artificial neural network (ANN) have been used.
of PV modules, type of PV cells and properties of materials used In these methods, historical data about weather estimation and
for PV systems. Increase in PV cell temperature causes open- environmental conditions are needed to train ANN and predict
circuit voltage to decrease significantly and short-circuit current power generation values of solar energy systems [12]. Finally,
to increase slightly [3]. It is known that PV cell temperature is hybrid models are described which combine two or more mod-
the same with PV module temperature [4]. Therefore, PV cell els to prevent disadvantages of a single model [13].
temperature can be used as an efficient input to forecast power Among power generation forecasting models, physical mo-
output value of a PV module on the basis of a single PV cell. dels are commonly used with the increasing technological de-
velopments in measurement of outdoor parameters and analysis
of solar panel specifications. Therefore, outdoor conditions and
solar cell properties should be defined clearly to forecast power
generation values. This detailed solar energy analysis helps us

1253
 Nomenclature

I Global solar radiation on PV module (W/m2 ) hw Wind convection coefficient (W/(m2 K))
Tc Cell/module temperature (◦ C) Vt Thermal voltage (V )
Ta Ambient temperature (◦ C) Voc Open circuit voltage (V )
vw Local wind speed close to the module (m/s) Isc Short circuit current (A)
uP V Heat exchange coefficient for the total surface of u0 Coefficient describing the effect of the radiation on
module (W ◦ C −1 m−2 ) the module temperature (W ◦ C −1 m−2 )
dVoc /dTc Voltage temperature coefficient (◦ C −1 ) dIsc /dTc Current temperature coefficient (◦ C −1 )
β Temperature coefficient of maximal power of the so- Pm Maximum power of a PV module (W )
lar cells (◦ C −1 ) Pm,cell Maximum power of a PV cell (W )
η Efficiency of the solar cells (unitless) Vm Maximum voltage(V )
u1 Cooling by the wind (W s ◦ C −1 m−3 ) α Absorption coefficient of the solar cells (unitless)
Im Maximum current (A) rs Normalized resistance (unitless)
τ Transmittance of the cover system (unitless) voc Normalized voltage (unitless)
Rs Series resistance (Ω) γ Cell maximum power temperature coefficient (◦ C −1 )

to optimize system size and dynamics of a house. Along with where TN OCT is the nominal operating cell temperature con-
the estimation of cell temperature and forecasting of power gen- sidered under nominal operating conditions of IN OCT = 800
eration may improve the accurate application of the PV systems W/m2 , Ta,N OCT = 20◦ C and wind speed of 1 m/s [16].
in future’s world. The value of TN OCT is accepted as 45◦ C depending on our PV
module’s datasheet.
2. Cell Temperature Estimation
2.2. Koehl Model
There are many correlations that describe cell temperature
as a function of outdoor parameters as well as solar cell char- The model is developed by considering the energy balance
acteristics defined by the manufacturers [14]. Cell temperature of a solar thermal collector defined in [8]. The cell temperature
significantly depends on global solar radiation on the surface estimation model is given as [9]:
of solar panels. In addition, it is affected by many outdoor I
T c = Ta + (2)
parameters such as wind speed, wind direction and ambient u 0 + u1 vw
temperature [15]. In this part of the study, cell temperature values
where the constants of u0 and u1 are the coefficients describ-
of the on-grid PV system of a solar house placed in Anadolu
ing, respectively, the effect of global solar radiation on module
University ˙Iki¨ Eylul Campus are estimated for five months by
temperature and cooling by the wind. These parameters are se-
using selected six different methods. The solar house built on
lected according to the specifications defined in [9] depending
campus is shown in Fig. 1.
on PV technologies.

2.3. Mattei Model

The model is developed by confirming an energy balance on
PV module, which neglects the temperature difference between
PV cells and the cover.
The effect of temperature on PV cell efficiency (η) can be
described in many ways. One of the most important models that
shows the effect of temperature on η is defined as:
 
η = ηST C 1 − β (Tc − TST C ) (3)
where ηST C is the reference module efficiency and β is the tem-
perature coefficient of maximal power for the PV module. In
addition, the energy balance can be described as:
α · τ · I = η · I + uP V (Tc − Ta ) . (4)
If the expression in (3) is included in (4), the proposed model is
Fig. 1. Solar house placed in Anadolu University İki Eylul
¨
obtained as:
Campus
uP V (vw ) Ta + I · (τ · α − ηST C (1 + βST C · TST C ))
Tc =
uP V (vw ) − βST C · ηST C · I
2.1. Standard Model (5)
where the expression of the heat exchange coefficient for the
The model only considers global solar radiation and am-
total surface of the module (uP V ) is defined as:
bient temperature. In this model, wind speed is not included
unlike the other considered models. The proposed model is: uP V (vw ) = 24.1 + 2.9 vw . (6)

I In (5), βST C is defined as temperature coefficient of maximal

Tc = Ta + (TN OCT − Ta,N OCT ) (1) power under standard test conditions of IST C = 1000 W/m2 ,
IN OCT

1254
TST C = 25◦ C and AM = 1.5. The values fo ηST C and is found as 0.0069 Ω according to the panel specifications given
βST C are obtained from the panel’s datasheet. In addition, τ · α by solar manufacturers. The maximum power point is defined
is accepted as 0.81 as in [9]. as:
Pm,cell = Vm Im (13)
2.4. Skoplaki Model where Vm and Im values are found:
In addition to global solar radiation and ambient tempera-   
b
ture, the proposed model considers wind speed and solar cell Vm = Voc 1 − ln a − rs 1 − a−b , (14)
νoc
properties such as efficiency, temperature coefficient of maxi-  
mal power, transmittance of the cover system and absorption Im = Isc 1 − a−b . (15)
coefficient of the cells [6]. The developed model is defined as:
 
Ta + I
IN OCT
h
(TN OCT − Ta,N OCT ) w,N OCT
h (v)
· 1 − ηST
τ ·α
C
(1 + βST C · TST C ) In (14) and (15), a and b coefficients are defined by the follow-
Tc =   w 
1−
βST C · ηST C I hw,N OCT
(TN OCT − Ta,N OCT )
ing relationships:
τ ·α IN OCT hw (v) a
(7) a = νoc + 1 − 2voc rs , b = (16)
1+a
where ηST C and βST C are defined as in Mattei model. Also,
hw,N OCT is the wind convection coefficient of wind speed un- where voc = Voc /Vt and rs = Rs / (Voc /Isc ). After finding
der normal operating conditions. The τ · α value in (7) accepted Pm,cell , Pm of a single PV module is found by considering cell
as 0.9 as in [6]. The wind convection coefficient (hw ) is defined number of the PV module, which is 60.
as:
hw = 5.7 + 3.8 vw (8) 3.2. Model 2
where vw is the local wind speed close to the module.
The model in [18] is performed by considering PV-Trombe
2.5. Muzathik Model wall (PV-TW) assisted with DC fan. The proposed power fore-
casting model is defined as:
The proposed model derives PV cell temperature as a func-  
tion of global solar radiation, ambient temperature and wind Pm = ηST C · A · I 1 − 0.0045 (Tc − 25) (17)
speed. However, the model does not consider the PV technol-
where A is the surface area of the PV module exposed to the
ogy of the considered solar panels. Hence, cell temperature is
interlayer and ηST C is the reference module efficiency. The
defined as [10]:
value of ηST C is accepted as 0.1598 according to the panel’s
Tc = 0.943 · Ta + 0.0195 · I − 1.528 · vw + 0.3529 . (9) datasheet.

2.6. Kurtz Model 3.3. Model 3

Similar to (9), the proposed model does not include material The model calculates the maximum power on the basis of a
characteristics of the PV panels. Instead, it is a parametrization single cell and, it is described as [19]:
of global solar radiation, ambient temperature and wind speed, I
∗
which is defined as [7]: Pm,cell = Pm,cell · ∗ · (1 + γ (Tc − 25)) (18)
I
Tc = Ta + I · exp (−3.473 − 0.0594 vw ) . (10) ∗
where Pm,cell is the cell maximum reference power and γ is
3. Power Output Forecasting the cell maximum power temperature coefficient. The value
of γ ranges from −0.005 to 0.003◦ C −1 in crystalline silicon.
Solar panel manufacturers provide maximum power values Since the parameter is not provided routinely by the AIL cer-
of solar panels in datasheets by considering ideal conditions. tificate of calibration of the module, the value of γ is accepted
However, these conditions are not stable in real time as −0.0035◦ C −1 as in [20]. Similar to Model 1, Pm,cell is
applications. Therefore, solar panel specifications should be multiplied by 60 to calculate Pm .
defined depending on unsteady outdoor parameters and, actual
power generation values must be analyzed in detail. In this part 4. Simulation and Results
of the study, three different power output forecasting models
are performed by using the estimated cell temperature values. Cell temperature values which are estimated by using six
These values are gathered from the model that has the highest different models are compared with the measured cell temperature
accuracy according to the measured values. values as shown in Fig. 2. In addition, statistical analysis methods
of Root Mean Square Error (RMSE), Mean Bias Error (MBE)
3.1. Model 1 and Mean Absolute Bias Error (MABE) are used to evaluate
the performance of the considered cell temperature estimation
Since panel datasheets show theoretical short-circuit current methods, which are described as:
and open-circuit voltage parameters, these values are found as 
[17]:
   n
I∗ dI RMSE =
1 (ci − mi )2 , (19)
Isc = sc I 1 + (Tc − Tc∗ ) sc , (11) n i=1
I∗ dTc
 
1
n
∗ dVoc I
Voc = Voc + (Tc − Tc∗ ) + Vt ln (12) MBE = (ci − mi ) , (20)
dTc I∗ n i=1
where I ∗ and Tc∗ are, respectively, the reference global solar 1
n

radiation on solar panels and cell temperature values. Tc is esti- MABE = (|ci − mi |) , (21)
n i=1
mated by using considered methods. The series resistance, Rs

1255
 Fig. 2. Comparison of different cell temperature estimation models

1  |Pm,h − Pf,h |
N
where ci is the ith calculated cell temperature data, mi is the ith
measured cell temperature data and n is the number of data. The NMAE% = · 100 , (23)
N CN
h=1
accuracy of the cell temperature estimation methods is shown
in Table 1. The results show that Skoplaki model has the high- where Pm,h is the power measured in the hour, Pf,h is the
est accuracy with the lowest RMSE, MBE and MABE values. power forecasted in the hour, CN is the net capacity of the plant
From the point of accuracy, it is analyzed that Skoplaki model is and N is the number of daylight hours. Table 2 shows the ac-
followed by Koehl model in terms of RMSE and MABE values. curacy of the power output forecasting methods. According to
When MBE values are considered, Muzathik model is the sec- this table, Model 1 has the highest accuracy among the consid-
ond model that leads the highest accuracy after Skoplaki model. ered models because it has minimum NMAE% and WMAE%
In addition, Muzathik model is the only model that gives un- values. Unlike Model 1, Model 3 has the maximum NMAE%
derestimation with the result of negative MBE values among and WMAE% values, which results in the lowest accuracy.
models. Finally, Table 1 indicates that Standard model, which Table 2. Accuracy of the power output forecasting methods
is the only model that does not consider wind speed among the
considered models, has the lowest accuracy. NMAE% WMAE%
Model 1 4.6394 12.3645
Table 1. Accuracy of the cell temperature estimation methods Model 2 4.7363 12.6229
Model 3 4.8816 13.0102
Model Name RMSE MBE MABE
Standard Model 4.5730 3.4378 3.4448
Koehl Model 2.3829 1.9520 2.0086
Mattei Model 2.5647 2.1248 2.1616 5. Conclusion
Skoplaki Model 2.2349 1.0943 1.8188 In this study, cell temperature values of a solar house placed
Muzathik Model 3.0926 -1.4297 2.3718 in Anadolu University ˙Iki Eylul¨ Campus are estimated by
Kurzt Model 3.3401 2.7028 2.7164 using selected methods. The results show that Skoplaki model
has the highest accuracy. Estimated cell temperature values of
Since Skoplaki model has the highest accuracy among the Skoplaki model are used in three different power forecasting
considered cell temperature methods, the cell temperature val- methods with global solar radiation, ambient temperature and
ues estimated by this model are used to forecast power output panel specifications. The statistical analysis methods indicate that
values. For this purpose, three different models are selected to Model 1 gives the best results among the considered models.
perform for four months of 2017. These forecasted power out- Therefore, it is concluded that these models are recommended to
put values are compared with the measured power output values be performed in any location that has similar climatic conditions
of the PV system in Fig 3. with the considered region.
The forecasted power output values based on the cell tem-
perature, global solar radiation and the specifications of the PV 6. Acknowledgement
modules are evaluated by using two statistical analysis methods
as Normalized Mean Absolute Error (WMAE% ) and Weighted This study is supported in part by the Scientific Research
Mean Absolute Error (WMAE% ). These analysis methods are Projects Commission of Anadolu University under the grants
described as: of 1505F512 and 1705F291.
N
|Pm,h − Pf,h | 7. References
h=1
WMAE% = · 100 (22)

N [1] A. Dolara, S. Leva, and G. Manzolini, “Comparison of
Pm,h different physical models for PV power output predic-
h=1

1256
 Fig. 3. Comparison of different power forecasting models

tion,” Solar Energy, vol. 119, pp. 83 – 99, 2015. [11] R. Ulbricht, U. Fischer, W. Lehner, and H. Donker, “First
[2] Ö. Ayvazoğluyüksel, Ü. Başaran Filik, and T. Filik, “Re- steps towards a systematical optimized strategy for solar
motely monitoring and modelling of renewable energy of energy supply forecasting,” in Proceedings of the Joint
a controlled home placed in A.U. İki Eylül Campus,” in ECML/PKDD 2013 Workshops, 2013.
8th International Ege Energy Symposium and Exhibition, [12] F. O. Hocaoğlu, Ö. N. Gerek, and M. Kurban, “Hourly
(Afyon), pp. 499 – 505, 2016. solar radiation forecasting using optimal coefficient 2-D
[3] D. M. Tobnaghi, R. Madatov, and D. Naderi, “The ef- linear filters and feed-forward neural networks,” Solar En-
fect of temperature on electrical parameters of solar cells,” ergy, vol. 82, no. 8, pp. 714 – 726, 2008.
International Journal of Advanced Research in Electri- [13] E. Ogliari, F. Grimaccia, S. Leva, and M. Mussetta, “Hy-
cal, Electronics and Instrumentation Engineering, vol. 2, brid predictive models for accurate forecasting in PV sys-
no. 12, pp. 6404 – 6407, 2013. tems,” Energies, vol. 6, no. 4, pp. 1918 – 1929, 2013.
[4] M. Mattei, G. Notton, C. Cristofari, M. Muselli, and [14] Ö. Ayvazoğluyüksel and Ü. Başaran Filik, “Estimation of
P. Poggi, “Calculation of the polycrystalline PV module monthly average hourly global solar radiation from the
temperature using a simple method of energy balance,” daily value in Çanakkale, Turkey,” Journal of Clean En-
Renewable Energy, vol. 31, no. 4, pp. 553 – 567, 2006. ergy Technologies, vol. 5, no. 5, 2017.
[5] T. Markvart, Solar Electricity, vol. 4 of 10. John Wiley & [15] J. S. Griffith, M. S. Rathod, and J. Paslaski, “Some tests
Sons, 3 ed., 7 2000. of flat plate photovoltaic module cell temperatures in sim-
[6] E. Skoplaki, A. G. Boudouvis, and J. A. Palyvos, “A sim- ulated field conditions,” in 15th IEEE Photovoltaic Spe-
ple correlation for the operating temperature of photo- cialists Conference, pp. 822–830, 1981.
voltaic modules of arbitrary mounting,” Solar Energy Ma- [16] E. Skoplaki and J. Palyvos, “Operating temperature of
terials and Solar Cells, vol. 92, no. 11, pp. 1393 – 1402, photovoltaic modules: A survey of pertinent correlations,”
2008. Renewable Energy, vol. 34, no. 1, pp. 23 – 29, 2009.
[7] S. Kurtz, K. Whitfield, D. Miller, J. Joyce, J. Wohlgemuth, [17] O. Elma and U. S. Selamoğulları, “A comparative sizing
M. Kempe, and et al., “Evaluation of high-temperature ex- analysis of a renewable energy supplied stand-alone house
posure of rack-mounted photovoltaic modules,” in 34th considering both demand side and source side dynamics,”
IEEE Photovoltaic Specialists Conference, pp. 2399– Applied Energy, vol. 96, pp. 400 – 408, 2012.
2404, 2009. [18] J. Jie, Y. Hua, P. Gang, J. Bin, and H. Wei, “Study of PV-
[8] D. Faiman, “Assessing the outdoor operating temperature Trombe wall assisted with DC fan,” Building and Envi-
of photovoltaic modules,” Progress in Photovoltaics Re- ronment, vol. 42, no. 10, pp. 3529 – 3539, 2007.
search and Applications, vol. 16, no. 4, pp. 307 – 315, [19] M. Fuentes, G. Nofuentes, J. Aguilera, D. Talavera, and
2008. M. Castro, “Application and validation of algebraic meth-
[9] M. Koehl, M. Heck, S. Wiesmeier, and J. Wirth, “Mod- ods to predict the behaviour of crystalline silicon PV mod-
eling of the nominal operating cell temperature based on ules in Mediterranean climates,” Solar Energy, vol. 81,
outdoor weathering,” Solar Energy Materials and Solar no. 11, pp. 1396 – 1408, 2007.
Cells, vol. 95, no. 7, pp. 1638 – 1646, 2011. [20] A. Luque and S. Hegedus, Handbook of photovoltaic sci-
[10] A. M. Muzathik, “Photovoltaic modules operating temper- ence and engineering. John Wiley & Sons, 2003.
ature estimation using a simple correlation,” International
Journal of Energy Engineering, vol. 4, no. 4, pp. 151 –
158, 2014.

1257