Professional Documents
Culture Documents
CrossMark
View Export
Online Citation
A new nickel EAM potential for atomistic simulations of ablation, spallation, and shockwave phenomena
AIP Conference Proceedings (March 2012)
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
Chen Zhang, Yunpeng Zhang,a) Jialei Su, Tingkun Gu, and Ming Yang
a)
Author to whom correspondence should be addressed: zhangyp@sdu.edu.cn
ABSTRACT
The accurate characterization and prediction of current-voltage characteristics of photovoltaic (PV) modules under different operating con-
ditions is essential for solar power forecasting and ensuring grid stability. The traditional method based on the single-diode model is inconve-
nient and complex because the current-voltage equation is implicit. In this paper, a novel method combining an artificial neural network
(ANN) with an explicit analytical model (EAM) is proposed for predicting the I-V characteristics of PV modules under different operating
conditions. The EAM makes it efficient to obtain the I-V curves from the estimated model parameters due to its simplicity and explicit
expression. The ANN based on the EAM is composed of a three-layer feedforward neural network, in which the inputs are solar irradiation
and module temperature and the outputs are the four parameters in EAM. Once the ANN is built and trained by using the measured I-V
curves, the shape parameters and I-V curve are predicted by only reading solar irradiation and temperature without solving any nonlinear
implicit equations. The accuracy and capability of the proposed method are verified by the experimental data for different types of PV mod-
ules. Moreover, the dependence of shape parameters in the EAM on solar irradiation and temperature is investigated first.
Published under license by AIP Publishing. https://doi.org/10.1063/1.5131432
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-1
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
electric grids. Along with the rapid growth of PV application, charac- proven that the EAM is useful for characterizing I-V curves of a variety
terization and performance prediction of PV modules have become an of solar cells having different conduction phenomena (Karmalkar and
essential topic of research. The ability to precisely forecast the electrical Haneefa, 2008). The relationship between the parameters in the SDM
energy produced by PV systems is of great importance and has been and those in the EAM was studied, the extraction method of five phys-
identified as one of the key challenges for massive PV integration. As ical parameters using the EAM was given (Saleem and Karmalkar,
the performance of outdoor PV systems is significantly influenced by 2009), and an analytical derivation of this four-parameter model from
weather conditions such as solar irradiance and temperature, accurate a physical-based implicit I-V equation was demonstrated (Das and
prediction methods of photovoltaic cells/modules are required to ana- Karmalkar, 2011). A thorough validation of this EAM was presented
lyze the dependence of electrical characteristics on solar irradiance and (Karmalkar and Saleem, 2011). Additionally, a semiempirical
temperature. approach considered to determine the peak power voltage for any illu-
Many methods have been proposed to characterize and predict minated solar cell based on this model was proposed (Das, 2012). A
the I-V properties of PV modules under different operating conditions method was proposed to predict the I-V characteristics of a PV panel
based on equivalent circuit models of PV modules, including the by combining the SDM and the EAM (Zhang et al., 2017). The EAM
single-diode model (SDM) (Chenni et al., 2007; De Soto et al., 2006; has been validated to be more suitable for different types of PV mod-
Jain and Kapoor, 2004; Khezzar et al., 2009; Tian et al., 2012; Zhou ules with different physical phenomena. In these studies, the EAM is
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-2
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
in SDM based ANN methods, which is inconvenient for rapid predic- The fill factor “FF” is represented as
tion of I-V properties under different operating conditions. m
The aim of this research is to develop a novel ANN method, FF ¼ vm im ¼ vm ½1 ð1 mÞvm vm : (5)
combined with the EAM, for predicting the output performance of PV It can be clearly seen from the formulas listed above that the
modules under outdoor operating conditions in virtue of the simple EAM is much simpler to characterize I-V curves than the SDM as it
and fast calculation of the EAM as well as the accuracy of ANN pre- has only four parameters and do not have to use Lambert-W functions
diction. The measured experimental data of different types of PV to make the current-voltage relationship explicit, which could curtail
modules under different operating conditions are used to validate the calculation complexity drastically.
proposed method. The dependence of shape parameters c and m on
module temperature and solar irradiance is investigated first by the III. THE ANN BASED ON THE EAM FOR PV MODULES
proposed method. Due to its capability of arbitrary input-output mapping, the mul-
The rest of this paper is organized as follows. In Sec. II, the EAM tilayer perceptron (MLP) neural network is the most widely used type
and the SDM are reviewed. In Sec. III, the configuration of the ANN of supervised neural network for approximation tasks and forecasting
based on the EAM is elaborated. Section IV shows the performance of (Hornik et al., 1989). The known data including input vectors and the
the proposed method for different types of PV modules. Finally, in corresponding target vectors are used to train a network until it can
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-3
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
X
2
S1j ¼ wij xi h1j ; (6)
i¼1
where the factor j indicates the number of hidden layer neurons, wij is
the weight on the connection from the ith input layer neuron to jth
hidden layer neuron, and h1j for j ¼ 1, 2, … and 20 represent the bias.
A hyperbolic tangent sigmoid transfer function was used as the
activation function of the hidden layer in the proposed method,
f1 ðSj Þ ¼ tansig ðSj Þ ¼ 2=ð1 e2Sj Þ 1: (7)
The small size of four-output network in this paper is helpful to
reduce the computational effort. This is especially of great importance
in the testing phase of the network where fast response is required.
The training process of the ANN method mentioned above is
simulated in the MATLAB Neural Network Toolbox, and the back-
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-4
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
PV modules associated with meteorological data (Marion et al., 2014). well-trained ANN, I-V curves and maximum power points under dif-
To ensure the accuracy, the data were recorded within every fifteen ferent operating conditions are obtained using Eqs. (3) and (4) by the
minutes from sunrise to sunset, for approximately one-year period. proposed method. The RMSE, NRMSE, R2, and R are calculated for
The recorded solar irradiances were greater than 20 W/m2 and lower each PV module to better assess the proposed method.
than 1500 W/m2 , while the recorded module temperatures were
between 34 C and 75 C. The PV module back surface temperature A. Results for mSi and xSi PV modules
uncertainty is 1.9 C, and the average irradiance uncertainties for the
The results for mSi and xSi PV modules are shown in Fig. 3,
four PV modules are mentioned below 3.4704% (mSi), 3.225% (xSi),
where the solid lines represent the I-V curves predicted by the pro-
3.1802% (copper indium gallium selenide module, CIGS), and
posed method under different operating conditions shown in Table II,
2.9755% (aSiTandem). Quality Assessment (QA) methods were
and the measured I-V data are plotted in Fig. 3 by dots. For compari-
implemented to exclude data not meeting quality thresholds. The QA
son, the ones predicted by De-Soto’s method (De Soto et al., 2006) are
methods are based on those previously established to provide
shown as dashed lines. De-Soto’s method is a conventional analytical
International Organization for Standardization (ISO) 17025 accredited
data for PV modules installed on the PERT at NREL. They include method usually used for single/multicrystalline PV modules.
checks for the reasonableness of the I-V curves, irradiances, PV mod- The range of the irradiance under different operating conditions in
Table II is from 224.6 W/m2 to 1137.8 W/m2 and 230.3 W/m2 to
Series cells 36 36 72 38
Parallel cells 1 1 1 1
Module area (m2 Þ 0.3429 0.647 1.75 0.79
Short circuit current (Isc) (A) 2.65 994 4.975 6.06 558 1.10 594
Open circuit voltage (Voc) (V) 22.0341 21.9851 41.6701 59.9138
Operating current at maximum power (Im) (A) 2.44 501 4.43 791 4.74 149 0.887 437
Operating voltage at maximum power (Vm) (V) 17.9879 17.3352 29.3889 44.5723
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-5
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
mSi xSi
TABLE III. RMSE and maximum power computed by the proposed method and De-Soto’s method for mSi and xSi PV modules.
Module Condition Proposed method De-Soto’s method Measured Proposed method Error (%) De-Soto’s method Error (%)
better applicability and accuracy than De-Soto’s method in predicting in Table IV, which are calculated based on the I-V curves recorded on
the I-V characteristics and power peak under different operating a day in Fig. 4. The values of R2 and R for the proposed method for
conditions. mSi and xSi are from 0.9983 to 0.9998 and from 0.9994 to 0.9999,
The performance indicators R2 and R in the proposed method respectively, which are higher than those in the literature (Ma et al.,
and De-Soto’s method for the two silicon-based PV modules are listed 2014) as well as than the value predicted by De-Soto’s method, which
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-6
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
are from 0.9952 to 0.9979 and from 0.9986 to 0.9995, respectively. The The predicted and measured maximum powers of CIGS and
higher value of R2 and R for the two PV modules indicates that the aSiTandem PV modules on a day are shown in Fig. 6, where the red
proposed method has better performance in the prediction of the I-V solid line depicts the prediction result of the proposed method and
characteristics and output power than De-Soto’s method. black dots are the measured peak power. It is clear that the predicted
peak power has good agreement with the measured data, which indi-
B. Results for CIGS and aSiTandem PV modules cates a high accuracy of the proposed method in a wide range of oper-
ating conditions.
The prediction results of the proposed method for CIGS and The performance indicators R2 and R in the proposed method
aSiTandem PV modules are obtained and shown in this section with- for CIGS and aSiTandem PV modules are shown in Table VII. The
out comparison with the result from De-Soto’s method. The reason is values of R2 and R calculated by the proposed method for the two
that the physical parameters at the reference condition, obtained by types of PV modules are both 0.9993 and 0.9998, respectively, which
De-Soto’s method, are negative without physical meaning. are close to the value of one. The value of R2 and R for CIGS and
The predicted results under different operating conditions for aSiTandem shows that the proposed method is also accurate in the
CIGS and aSiTandem PV modules are shown in Fig. 5. The details of prediction of the I-V characteristics and output power for CIGS and
operating conditions are listed in Table V. The range of the irradiance aSiTandem PV modules, which are not able to be modeled and pre-
under different operating conditions in Table V is from 218.2 W/m2 dicted by De-Soto’s method.
to 1090.6 W/m2 and 218.0 W/m2 to 1002.4 W/m2 for CIGS and The average training times for the four PV modules are shown in
aSiTandem PV modules, respectively. It can be seen that the predicted Table VIII. The numbers of I-V curves calculated in the training phase
I-V curves show good consistency with the measured data, and the are 190 (mSi), 193(xSi), 202 (CIGS), and 196 (aSiTandem), respec-
specific data are provided in Table VI for analysis. In Table VI, the tively. The increased RAM are 106 MB (mSi), 113MB (xSi), 120MB
maximum values of RMSE for CIGS and aSiTandem are 0.0474A and (CIGS), and 118MB (aSiTandem), respectively.
0.0093A, respectively, and the errors of maximum power for CIGS The average computing times of the proposed method for the
and aSiTandem are less than 1%, except for condition 5 with lower four PV modules under each operating condition are also listed in
solar irradiation. Because of the low ratio of data under low solar irra- Table VIII, as well as the average computing time of De-Soto’s method
diation in the training data, the errors between the predicted and mea- for mSi and xSi PV modules. As inputting the solar irradiance and
sured power peak increase in the low solar irradiation region. module temperature of each operating condition, the average
TABLE IV. Performance statistics of the proposed method and De-Soto’s method on predicted power for xSi and mSi PV modules.
R2 (% ) R (% )
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-7
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
CIGS aSiTandem
TABLE VI. RMSE and maximum power computed by the proposed method and measured data for CIGS and aSiTandem PV modules.
computing times of the proposed method for each I-V curve are less the higher calculation speed of the proposed method is it takes advan-
than 0.01 s. tage of the EAM that contains less parameters and does not have to
The average computing times of De-Soto’s method for mSi and use the Lambert-W function or iterative calculations.
xSi are 0.6511 s and 0.6756 s, respectively, which are much higher The uncertainty analysis of the proposed method is investigated
than the sets of times of the proposed method. The possible reason for using NRMSE and massive measured data for the four PV modules,
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-8
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
shown as Fig. 7. The numbers of I-V curves used for calculating open-circuit voltage [V/(W/m2 )], the relative temperature coefficient
NRMSE are 36165 (mSi), 11929 (xSi), 34775 (CIGS), and 39186 of open-circuit voltage (V/K), and the relative temperature coefficient
(aSiTandem). The mean of NRMSE calculated by the proposed of short-circuit current (A/K), respectively.
method and measured data for mSi, xSi, CIGS, and aSiTandem are The dependence of shape parameters c and m on irradiation and
2.44%, 2.40%, 1.99%, and 0.60%, respectively. It is obvious that the temperature has not been reported. Since the shape parameters c and
maximum NRMSE for all the four PV modules is lower than 8.1%, m lack physical meaning, it is difficult to get the analytical relationship
while the one for aSiTandem is generally lower (less than 2%) and between shape parameters and operating conditions. Using the pro-
more concentrated, indicating that the proposed method is more accu- posed method, shape parameters at different solar irradiation and tem-
rate implementing for this module. peratures are obtained and investigated. The actual values of Voc and
Isc were obtained from the measured I-V curves directly, and the shape
V. DEPENDENCE OF PARAMETERS ON IRRADIATION parameters c and m are obtained by fitting the measured I-V curve
AND TEMPERATURE with Eq. (3) using the lsqnonlin instruction in MATLAB. In the pro-
In this section, the dependence of parameters in the EAM on cess of curve fitting, the calculation precision is set as 1 106 to
ensure the reliability of the calculated results. For comparison, the
irradiation and temperature is discussed. The irradiation and tempera-
dependence of the four parameters on irradiance and temperature is
ture dependence of Voc and Isc has been discussed in the literature
also descripted analytically, using Eqs. (12) and (13) for Voc and Isc and
(Singer et al., 1984) and given by
using the relationship between shape parameters c and m and physical
G parameters Rsh , Rs , and g in SDM (Das and Karmalkar, 2011).
Isc ¼ Isc;ref 1 þ kI ðT Tref Þ ; (12)
Gref
Voc ¼ Voc;ref ln e þ kðG Gref Þ ½1 þ kU ðT Tref Þ; (13)
TABLE VIII. Average training time for each PV module and computing time of two
where G and T represent the solar radiance and cell temperature and methods.
k, kU , and kI are constants that can be obtained directly from the
manufacturer and represent the relative irradiation coefficient of Average computing time
for each I-V curve (s)
TABLE VII. Performance statistics of the proposed method for CIGS and aSiTandem Average training Proposed De-Soto’s
PV modules. Module time (s) method method
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-9
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-10
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
represent the results calculated by the analytical method, and the dots Figure 9(c) shows how the open circuit voltage of different PV
represent the actual value. For each PV module, the relationship modules changes with the module temperature. For the four PV mod-
between parameters and temperature is shown at G ¼ 800 W/m2 . ules, the dependences of Voc on solar irradiance and module tempera-
In Fig. 9, with increasing module temperature, the value of ture are similar where the open circuit voltage ascends slightly with
parameter m decreases lineally for mSi, xSi, and CIGS, while increases the increase in solar irradiance and decline linearly with the increase
lineally slightly for aSiTandem. The value of m decreases in the order in module temperature. Figure 9(d) shows that the short circuit cur-
of mSi, xSi, aSiTandem, and CIGS under certain irradiation. rent of different PV modules almost remains constant with increasing
The parameter c varies lineally with increasing module tempera- module temperature. The value of Isc decreases slightly for the consid-
ture. For mSi and xSi, the values of c are near 1 with small variation, ered PV modules except xSi where its value increases with the increase
while the value of c varies around 0.9 with relatively large variation. in module temperature.
This indicates the absolute value of the slope on the I-V curve around The value of m calculated by the analytical method decreases
Isc for mSi and xSi modules are lower than CIGS and aSiTandem. with the increasing temperature for the four PV modules, which agrees
Compared to parameter m, the variation of c with changing tempera- with the results of mSi, xSi, and CIGS. While the actual value of m for
ture could be negligible because of the small range of variation. aSiTandem increases with the increase in temperature, there is a large
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-11
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
deviation from the actual value for CIGS. The value of c calculated by under different operating conditions. The proposed method consists
the analytical method increases slightly with the increasing tempera- of a three-layer feedforward neural network, in which the inputs are
ture and is close to the actual value for mSi, xSi, and aSiTandem except solar irradiation and module temperature and the outputs are the four
for CIGS. Besides, as shown in Figs. 9(c) and 9(d), the analytical parameters in the EAM. Once the neural network is trained by using
method using Eqs. (12) and (13) could capture the relationship the measured I-V curves, the four parameters of the EAM and the cor-
between Voc , Isc , and temperature in an acceptable accuracy. responding I-V characteristics are estimated by only reading solar irra-
To sum up, it can be seen in Figs. 8 and 9 that the proposed diation and temperature without solving any nonlinear implicit
method has a better agreement with the measurements than the ana- equations. The proposed method is verified by the experimental data
lytical method for the four PV modules, indicating a higher accuracy for four different types of PV modules, indicating the wider range of
and a wider range of applications for the proposed method. applications of the proposed method for I-V prediction of different
types of PV modules. The results show a good agreement between the
VI. CONCLUSION estimated and actual I-V curves under different operating conditions,
In this paper, a novel method combining the ANN with the as well as higher calculation speed of the proposed method. Using the
EAM is proposed for estimating the I-V characteristics of PV modules proposed method, the dependence of parameters in the EAM on solar
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-12
Published under license by AIP Publishing
Journal of Renewable ARTICLES scitation.org/journal/rse
and Sustainable Energy
irradiation and temperature is also obtained and discussed. The Karatepe, E., Boztepe, M., and Colak, M., “Neural network based solar cell mod-
dependence of Voc and Isc on operating weather conditions accords el,” Energy Convers. Manage. 47(9-10), 1159–1178 (2006).
Karmalkar, S. and Haneefa, S., “A physically based explicit J-V model of a solar
with the results in the literature. The dependence of shape parameters
cell for simple design calculations,” IEEE Electron Device Lett. 29(5), 449–451
c and m on operating weather conditions is investigated for the first (2008).
time. A good agreement between the estimated and actual values of Karmalkar, S. and Saleem, H., “The power law J-V model of an illuminated solar
parameters validates the accuracy of the proposed method under dif- cell,” Sol. Energy Mater. Sol. Cells 95(4), 1076–1084 (2011).
ferent operating conditions. Khezzar, R., Zereg, M., and Khezzar, A., “Comparative study of mathematical
methods for parameters calculation of current-voltage characteristic of photo-
voltaic module,” in 2009 International conference on Electrical and Electronics
ACKNOWLEDGMENTS Engineering-ELECO (2009), pp. I24–I28.
Laudani, A., Fulginei, F. R., and Salvini, A., “High performing extraction proce-
We would like to thank NREL and especially Mr. Bill Marion dure for the one-diode model of a photovoltaic panel from experimental I-V
for providing the experimental data. curves by using reduced forms,” Sol. Energy 103, 316–326 (2014a).
This work was supported by the National Key Research and Laudani, A., Fulginei, F. R., and Salvini, A., “Identification of the one-diode
Development Program of China (Grant No. 2018YFB0904200). model for photovoltaic modules from datasheet values,” Sol. Energy 108,
432–446 (2014b).
Lun, S. X., Wang, S., Yang, G. H., and Guo, T. T., “A new explicit double-diode
REFERENCES
J. Renewable Sustainable Energy 12, 013501 (2020); doi: 10.1063/1.5131432 12, 013501-13
Published under license by AIP Publishing