Escolar Documentos
Profissional Documentos
Cultura Documentos
1. Introduction
The key issue in the world lately has been the damage of the climate conditions by greenhouse gases.
The emission of greenhouse gases is related to various factors around the world. One of the main
reasons is the burning of the conventional fuels like coal, oil and natural gas [1]. In order to reduce the
pollution concerns and to meet the swiftly increasing energy demand, government is paying huge
attention towards the non-conventional sources such as solar, wind and hydro [2]. Pakistan is a
developing country which is unable to meet its energy demand with acute energy shortfalls reaching
5500 MW in 2015. The major reason of shortfall is heavy reliance on the thermal energy rather
renewable energy [3]. The oil and gas are the major fuels in Pakistan to produce electricity with 80%
and 6% of the whole generation respectively. However, the renewable energy has not been exploited
fully until now. Pakistan is a favorable country for the solar energy exploitation which receives 15.5
×1014 kWh/year. The solar energy potential can be used to produce 1600 GW per year [4]. Quetta is
one of the major cities of Pakistan with considerably high level of irradiation up to 5.54 kWh/m2/day
as compared to other main populated cities of Pakistan [5]. The similar country which is undergoing
an enormous surge in exploiting the solar energy is Chile. In March 2016, the installed capacity of PV
was 1.1 GW but many new projects have been carried out. Most of these projects are based in
Atacama Desert. This place has been known as one of the places with the highest surface solar
irradiation. The main characteristics of Atacama Desert are high altitude and clear sky days [6]. The
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
mean clearness index of Atacama Desert is 3% with global horizontal irradiation reaching up to 2500
kWh/m2 and maximum temperature of 38oC [7]. On the other side, more than 40,000 villages have no
access to the nationalpower grid in Pakistan. The southern province of Pakistan i.e. Sindh and
Baluchistan are suitable for solar energy with maximum daily sun hours of 7-8 [8]. Chile, similarly, is
in its way to maximize the proportion of its solar energy percentage in national energy market. The
20% energy is intended to be produced by the renewable energy resources until 2020. However, there
are lot of hindrances in the way of achieving this target such as lack of solar database and utilization
mechanism [9]. The 50% Chile’s economy is based on mining industry (mainly coal industry) and
80% of energy demand is needed in the Atacama Desert [10]. Many countries are now adopting the off
grid photovoltaic systems in public policies to ensure the easy access of energy by installation on
remote places [11]. The standalone photovoltaic systems can be regarded as the best choice for rural
areas because of simple installation system topology and control with a load demand of 1-100 KW
[12]. Two geographical location are selected to be used in this study for power output potential
production of off-grid photovoltaic system; Quetta, Pakistan and Copiapo-Chamonate (CC), Chile.
The simulation software used here is PVSYST which is developed by EPFL for the reliable and
efficient evaluation of various configurations of both off grid and on grid photovoltaic systems [13].
The fundamental purpose of this paper is to determine the potential of Quetta in comparison to CC
for generating power from off grid photovoltaic system for a standard household consumer. This paper
is structured in five sections. The first section is for the general description and introduction of the
problem considered. The second section is for determining the geographical parameters including
albedo and tilt angle. The third section is for designing the electrical parameters including electrical
components and loads. In the fourth section the discussion of simulation results is presented and, in the
section 5, the conclusions are drawn.
Figure 1. Global Irradiation (kWh/m2) for Quetta Figure 2. Monthly Average Temperature (oC) for
and CC. Quetta and CC.
The sites chosen are Quetta and Copiapo-Chamonate (CC). The geographical coordinates of
CapiapoChamonate are 27.30 S latitude and 70.42 W longitude. The elevation of the site is 301 m.
The time zone is UT -4. The latitude and longitude of Quetta site is 30.1 N and 66.95 E respectively
with an elevation of 1702 m with time zone of UT +5. Since, CC is located in the Southern
Hemisphere and Quetta is situated in the Northern Hemisphere, the winter and summer months are
different for these two places. In CC, the winters are in the months from May to August and the
summer lasts from September to December. July is the month with lowest average temperature of
11.8oC and January with the highest temperature of 22.4oC. The temperature variance between
summer and winter at CC is not significant. Quetta has a significant trend of variation in climate
2
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
between summers and winters. The summer season is from March to October with 30.5 oC as the
highest average temperature and the winter season is from November to February with average lowest
temperature of 7.3oC.
The global irradiation GI (kWh/m2) and average month temperature of the two locations is shown
in figures 1 and 2, respectively.
Tsa = T S
(1)
NS
TWa = T W
(2)
NW
Where Tsa,Twa, Ts, Tw, Ns and Nw are the summer average tilt angle, winter average tilt angle,
individual summer months tilt angle, individual winter months tilt angle, number of summer months
and number of winter months respectively. The individual tilt angles for every month can be obtained
from [13]. The tilt angles for Quetta in summer and winter months are 16o and 53o respectively and
13o and 47o for Copiapo- Chamonate, respectively.
2.3. Albedo
The albedo is the characteristics of a surface linked with its proportion of reflectance or absorption.
The usual range of albedo is 0.1-0.5 [14]. The dark black surface is considered to be at 0 while the
pure white surface is termed to be 1. The albedo of the two sites considered can be obtained from [15].
The albedo of the two geographical sites is given in the table 1.
3. Electrical system
3
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
4
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
to meet the actual user demand and provided only 88.9% of the total load. The Quetta lacked only in
January with the provision of 95% of the load need. The solar fraction for both sides is given in figure
5. The performance ratio and solar fraction can be calculated from the equations (5) and (6)
respectively.
Yf
PR = (5)
Yr
Eu
SF = (6)
El
Where PR, SF, Yf, Yr, Eu, El are performance ratio, solar fraction, system yield, nominal system
yield, energy supplied to the user and energy need of the user respectively. The performance ratio PR
is the ratio of actual power produced by the PV array and the nominal power expected to be produced
by PV array at STC conditions. The Eu is the actual power supplied to the load and El is the actual
load demand. The solar fraction SF is the ratio of energy provided from the array to the load versus the
actual demand of the load.
Figure 4 Solar Fraction (SF) at Quetta and CC. Figure 5 Comparison between Energy supplied to
the load Eu and Energy demanded by the load El.
The missing energy at CC has shown a downfall curve in the winters especially in June. The Euat
Quetta performed at capacity despite in January while, in winters, CC was unable to supply the load
demand with 57.6 kWh energy shortfall in June. The energy supplied to the load Euand energy
demand of the load El for both sites is given in figure 6.
The missed energy is 183.66 kWh at CC with maximum of 57.65 kWh in June. The Quetta site
performed well in terms of missing energy that is 17.93 kWh only in January. The total unused energy
for Quetta is 534 kWh with maximum in the month of June and considerable average of 21.625 kWh
in the winters. Whereas, the excess energy at CC is in the months of Summer with an average of
65kWh with the maximum value of 118.4 kWh in December. The various losses of the off grid
photovoltaic system at CC and Quetta is given in figure 7. The most considerable losses at both sides
is in the working of the system at MPP (maximum power point) and temperature loss. Temperature
loss at Quetta is 6.6% and 5.5% at CC. The Quetta site has more temperature loss because of high
temperatures as compared to CC. However, the performance of CC at MPP is very low and 1872 kWh
energy is lost.
5
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
Figure 6. Available Energy Eav (kWh) at CC Figure 7. Major Losses at Quetta and CC.
and Quetta.
5. Conclusion
This paper simulates and compares the two geographical sites of Quetta and CC for an off grid
photovoltaic system. CC, Atacama is considered to be one of the best receiving solar irradiation place.
PVSYST model results show that Quetta performs well in comparison to CC. The following results
are obtained from the studies done in this paper.
⚫ Solar Fraction at Quetta site was 100% the whole year, however, lacking only in January with
missing energy of 17.93kWh
⚫ The energy used by the load at Quetta is 3.807% more as compared to CC
⚫ Performance ratio PR at Quetta is 0.7% more than at CC
⚫ The Quetta site has produced 3.96% excess annual energy as compared to CC
This paper can provide the further research on modeling the off grid photovoltaic systems in
Pakistan for domestic consumers. According to the author, special attention should be paid in reducing
the temperature losses, tilt angle adjustment and MPP running losses. The photovoltaic roofs with low
reflectance and less albedo are recommended. The tilt angles should be modified according to the
location temperature to increase performance ratio PR.
References
[1] Danish, Wang ZH 2017 Role of renewable energy and non-renewable energy consumption on
EKC: Evidence from Pakistan J. Clean. Prod. 156 855-64.
[2] Kamran M 2018 Current status and future success of renewable energy in Pakistan Renew.
Sustain. Energ. Rev. 82 609-17.
[3] Qureshi T M, Ullah K and Arentsen M J 2017 Factors responsible for solar PV adoption at
household level: A case of Lahore Pakistan Renew. Sustain. Energ. Rev.78 754-63.
[4] Zafar U, Rashid T U, Khosa A A, Khalil M S and Rahid M 2013 An overview of implemented r
enewable energy policy of Pakistan Renew. Sustain. Energ. Rev. 82 654-65.
[5] Khalid A and Junaidi H 2013 Study of economic viability of photovoltaic electric power for
Quetta-Pakistan Renew. Energ. 50 253-8.
[6] Ferrada P, et al 2017.Potential for photogenerated current for silicon based photovoltaic
modules in the Atacama Desert Sol. Energ.144 580-93.
[7] Olivares D, Ferrada P, de Matos C, Marzo A, Cabrera E, Portillo C and Llanos J
2017.Characterization of soiling on PV modules in the Atacama Desert Energy Procedia 124
547-53.
[8] Sheikh M A 2010 Energy and renewable energy scenario of Pakistan Renew. Sustain. Energ.
Rev. 14 354-63.
[9] Escobar R A, Ortega A, Cortés C, Pinot A, Pereira E B, Martins F R and Boland J 2014. Solar
energy resource assessment in Chile: satellite estimation and ground station measurement
Energy Procedia 57 1257-65.
6
NEFES 2018 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 188 (2018)
1234567890 ‘’“” 012008 doi:10.1088/1755-1315/188/1/012008
7
IOP Conference Series: Earth and Environmental Science
E-mail: cksolar@163.com
Abstract. In order to select effective samples in the large number of data of PV power
generation years and improve the accuracy of PV power generation forecasting model, this
paper studies the application of clustering analysis in this field and establishes forecasting
model based on neural network. Based on three different types of weather on sunny, cloudy
and rainy days, this research screens samples of historical data by the clustering analysis
method. After screening, it establishes BP neural network prediction models using screened
data as training data. Then, compare the six types of photovoltaic power generation
prediction models before and after the data screening. Results show that the prediction
model combining with clustering analysis and BP neural networks is an effective method to
improve the precision of photovoltaic power generation.
1. Introduction
With the expansion of installed capacity of photovoltaic power generation, its proportion in the power
grid is also increasing these days. The random fluctuation characteristics of its output will cause the
considerable impact on the power grid system, which indirectly affects the operation of the security
and stability of the power system. Therefore, it is extremely important to predict the photovoltaic
power generation effectively to ensure the security and stability of the power system. At present, some
forecasting models for photovoltaic power generation are SVM model [1,2], ARMA model [3], gray
theoretical model [4] and neural network model, etc. [5,6]. However, the neural network model
generally has better performance among them [7]. Therefore, this study uses this method to predict the
photovoltaic power generation.
In the process of BP neural network prediction, the quality of the input samples will directly affect
the accuracy of the prediction model. Practically, when the photovoltaic power plant collects and
stores the running data, the errors occur which form the abnormal samples. These anomaly samples
deviate from the factual basis and the corresponding weather patterns are not correctly described
according to its established model. Therefore, it is necessary to screen out these anomalous samples.
Manually, it is very difficult to screen out anomalous samples because of a large amount of data
recorded in the past years. Therefore, most of the research papers use raw data collected from the
untreated power plants [4-8], and only a few papers deal with filtered data [9]. However, these papers
do not focus on the specific screening method. There are some papers, which use the wavelet analysis
method, the mean fill method and other methods to deal with the original data [10,11]. However; there
is no specified method to verify the accuracy and effectiveness of the power generation forecasting
model.
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
Therefore, based on the statistical data of different kinds of weather, this paper cleans the historical
data using the clustering analysis. Then, based on the BP neural network, 6 different kinds of
photovoltaic power generation forecasting models are established and verified in the sunny, cloudy
and rainy days in MATLAB software [12,13]. The results show the validity and accuracy of the
clustering analysis of historical data.
Cluster
Forecasting analysis (data
model based on cleaning)
BP neural
network Forecasting
model based on
BP neural
network
Input simulation data Input simulation data
Firstly, 3 typical weather types of sunny, cloudy and rainy days are taken as examples to
analyze the theoretical radiation, actual radiation and power generation. The outliers are
filtered out according to the cluster pedigree.
Secondly, the historical data of 2012.01~2014.01 is used as the training sample. The
theoretical radiation and the highest and the lowest temperatures are used as the input of BP
neural network. The historical data collected by the data collector of PV power plant are used
as the output of BP neural network. The neural network model is trained to establish the
prediction models of 6 kinds of photovoltaic power generation
Thirdly, the simulation data of 2014.02 ~ 2015.01 is put into the prediction model of
photovoltaic power generation. The purpose is to make use of the BP neural network's
2
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
nonlinear learning ability to predict the daily power generation under different weather types
Finally, in order to verify the need of data screening, the MAPE (the mean absolute percentage
error) value is calculated for the predicted power generation and the actual generation and the
performance is evaluated.
2.2.1. Clustering method. According to the different clustering methods, clustering analysis can be
divided into system clustering method, adding method, decomposition method and dynamic
classification method, etc. [14]. In this paper, the purpose of clustering analysis is to deal with the
outlier data (abnormal samples) from the original data to improve the quality of the original data.
Therefore, this paper uses the system clustering method. The main algorithm steps are as follow:
To convert the objects in the sample set;
The processed objects are divided into n classifications and every class contains an object;
Calculating the distance between each of the two classes;
Merging the nearest two classes as a new class;
Calculating the distance between the new class and the current class. If the distance is close
enough, then merge into a class. This process continues until all classes (or samples) are
combined. If this step works well then we will move further to step 6, otherwise, to step 4.
Drawing the clustering pedigree of the sample set, as shown in figure 2;
According to the number of classes to be classified, the corresponding classification results are
obtained.
2
class a
4
Y
class b
1
class c
5
Figure 2 shows the data clustering results for randomly selected eight groups on rainy day. The
sample can be clustered into two, three, four or five categories. As shown in the figure, when the class
3
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
distance is 5, the samples can be clustered into three classes. The samples 6, 7, 2, 3 and 4 are class a,
sample 1 and sample 8 are b, and sample 5 is c.
2.2.2. Select the metric. The metric directly affects the quality of the cluster, so the statistical analysis
of the historical data of the database is carried out to determine the metric. In the figures 3-5 of cloudy
weather, some samples are found away from the main data sets which are outliers. While, the data
distribution exhibits a distinct oblique strip.
4000 30
3000
20
2000
10
1000
0 0
0 5 10 15 20 25 30 5 10 15 20 25 30
Actual daily radiation/ MJ·m-2 Theoretical daily radiation/ MJ·m-2
Figure 3. Scatter plots of actual radiation and Figure 4. Scatter plots of theoretical and actual
power generation. radiation.
3000
io n /k W h
2000
P o w er g en erat
1000
30
0
/㎡
25
The 0
oret 20
on/MJ
ical 1 0 iati
radi
atio 15
al rad
u
n/M 20
J/㎡ 10 Act
30
Figure 5. Spatial scatter plots of actual radiation, theoretical radiation and power generation.
According to the previous study results [15], the data covered by the oblique range belong to a
class. Selecting the cosine coefficient as the criteria, the data on the oblique bar is classified as valid
data and other data is classified as the outlier which is far from the oblique bar. After the outlier
deletion, the discrete degree of the sample is optimized. The cosine distance is measured according to
4
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
formula (1):
0
X ik X jk
k 1
dij cos(ij ) (1)
n n
X ik2 X 2jk
k 1 k 1
In the above formula, d ij -the distance between simples X ik and X jk ; X ik - the k-the variable of
the i sample; X jk - the k-the variable of the j sample; n - number of samples.
i j k
X1
Input X2
mode
…
Xm
wij w jk
m q L
input layer hidden layer output layer
Figure 6. Structure of BP neural network.
Where wij is the connection weight between the input layer node and the hidden layer node, and w jk is
the connection weight between the hidden layer node and the output layer node. The input of the
hidden layer and the output layer node is the weighted sum of the output of the previous node [16]. In
this paper, a three-layer neural network is used which contains only one hidden layer.
Evaluation of photovoltaic power generation forecasting model. In this paper, a two-layer forward
neural network based on Levenberg-Marquardt (LM) algorithm is adopted. The neural network
toolbox of Matlab 2010b is used to build and train the network and obtain the simulation results. The
training data and test data are randomly selected by neural network as 90% and 10% respectively of
the total data. In order to ensure the credibility of the results, test data is not used in the training part.
The simulation results use the average absolute percentage error (MAPE) to evaluate the entire
system, which is calculated as (2):
N P i Pi
100
MAPE
N i 1
f
Pa i
a
% (2)
In the above formula, N - the total number of data; Pf - the actual power generation; Pa - predicted
power generation; i - the data serial number. The lower MAPE value ensures the better predictive
effects.
5
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
30
Actual daily radiation/MJ·m-2
20
10
0
5 10 15 20 25 30
Theoretical daily radiation/ MJ·m-2
Figure 7. The theoretical and actual daily radiation scatter plots after screening in cloudy.
The clustering results of the various data samples are shown in table 1. In the sunny days, the
atmospheric state is relatively stable and the reverse radiation of the atmosphere is weak. Therefore,
the theoretical daily radiation is close to the actual amount of radiation in the whole day, and the
degree of data discrimination is small and the screening ratio is 10.7%. In the cloudy days will have
large random variation in terms of clouds and the similarity of the data is low, the screening ratio is
13.1%. In the rainy days, the atmospheric state fluctuates at a higher degree and the screening ratio is
also large which is 21.3%.
Table 1. The clustering result statistics for all kinds of sample data.
Weather type Sunny Cloudy Rainy Total
Before screening 150 168 61 379
After screening 134 146 48 328
Screening ratio 10.7% 13.1% 21.3% 13.5%
6
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
three kinds of weather before and after the clean-up process (the day number in the x axis is the data of
the forecast in a year). From the results, it can be seen that the difference between the predicted value
and the actual value of the model established by the weather type is relatively large before the data is
cleaned up. After data clearing, the difference between the predicted value and the actual value
become small. The predicted value of the model after the cluster analysis is closer to the actual value.
2000 2000
1000 1000
0
20 80 146 206 263 335 0
2 45 112 148 272 329
Day number Day number
Figure 8. The relationship between the actual Figure 9. The relationship between the actual
power generation and the prediction of power power generation and the prediction of power
generation in sunny weather. generation in cloudy weather.
1500
Actual power generation
Before screening
Power generation / kWh
After screening
1000
500
0
62 218 250 256 281
Day number
Figure 10. The relationship between the actual power generation and the prediction of power
generation in rainy weather.
Table 2 shows the difference between the predicted results of the six neural network models.
Compared with the MAPE mean value, it can be seen that the predicted value of the model after the
cluster analysis is closer to the actual value in each case. The average MAPE value of the model after
clustering analysis is 23.2% which is much smaller than the conventional models. Indicating that the
7
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
clustering analysis is effective, and the prediction model of photovoltaic generation is more accurate
base on the data after clustering analysis. Therefore, it is necessary to screen data.
In the case of sunny days, the prediction results of two models are closer, because the sunny
weather atmospheric state is relatively stable and the degree of data discrete is small, the MAPE value
of the prediction model that after screening is 11.7% lower than that before screening, the prediction
effect is better. In the case of cloudy days, the thickness and location of the clouds is difficult to
predict, the predicted results of the two models are relatively different, the MAPE value of the
prediction model that after screening is 23% lower than that before screening. In the case of rainy days,
the MAPE value of the prediction model that after screening is 51.3% lower than that before screening,
which difference is the biggest. It can be seen from the table 2, the prediction of rainy days is far less
accurate than sunny and cloudy days. The reason is the rain weather atmospheric changes are more
complex, in different rainy days, clouds, rain and vapor will be very different, and the absorption of
solar radiation will be very different, this will produce a large relative error, so it is difficult to achieve
the desired prediction effect in rainy days.
Based on overall assessment, the MAPE value of the predicted model is 23.6% which is less than
46.8% of the pre-screening prediction model. The comparison results show that photovoltaic power
generation models are more accurate when using the clustering analysis of the original data. Therefore,
it is necessary to carry out data cleanup.
5. Conclusion
In this paper, the prediction model is established by combining the clustering analysis and the BP
neural network. The abnormal samples in the historical monitoring data of the PV power plant are
screened out, and the prediction results are evaluated and analyzed. Research indicates:
Using the cosine coefficient as the evaluation criteria, the abnormal samples are removed
using clustering analysis and screening of historical data, which can effectively filter the
discrete operating data.
After the data is cleaned up as the training data, the BP neural network prediction models can
be established which can improve the accuracy of the prediction results.
The data cleaning is necessary for establishing the prediction model using the historical data.
The cluster analysis can be effectively applied to the photovoltaic power generation
forecasting system.
The above research results can provide an effective method for the forecasting technology of PV
power plants, which is beneficial to the popularization and application of photovoltaic power
generation technology.
References
[1] Li R and Li G M 2008 Photovoltaic power generation output forecasting based on support
vector machine regression technique Electr Pow 41(2) 74-8
[2] Shi J, Lee W J, Liu Y Q, et al 2012 Forecasting power output of photovoltaic systems based on
weather classification and support vector machines IEEE Trans Ind Appl 48(3) 1064-9
[3] Lan H, Liao Z M and Zhao Y 2011 ARMA model of the solar power station based on output
prediction Electr Meas Instrum 48(54) 31-4
[4] Wang S X and Zhang N 2012 Short-term output power forecast of photovoltaic based on a grey
and neral network hybrid model Autom Electr Pow Syst 36(19) 1-5
[5] Christophe P, Cyril V, Marc M et al 2010 Forecasting of preprocessed daily solar radiation time
series using neural networks Sol Energy 84(12) 2146-60
[6] Zhang Y X and Zhao J 2011 Application of recurrent neural networks to generated power
forecasting for photovoltaic system Pow Syst Prot Control 39(15) 96-101
[7] Şenkal O and Kuleli T 2009 Estimation of solar radiation over Turkey using artificial neural
network and satellite data Appl Energ 86(7-8) 1222-8
[8] Wang F, Mi Z Q, Zhen Z et al 2013 A classified forecasting approach of power generation for
8
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012024 doi:10.1088/1755-1315/93/1/012024
photovoltaic plants based on weather condition pattern recognition Proc CSEE 33(34) 75-82
[9] Lu J, Zhai H Q and Liu C 2010 Study on statistical method for predicting photovoltaic
generation power East China Electr Pow 38(4) 563-7
[10] Lang Y and Zhang W T 2016 Research on the application of MATLAB in the data processing
of photovoltaic power generation Electr Technol Softw Eng 2016 87
[11] Zhang H N, Zhang J T, Yang L B et al 2016 Date processing method for PV power station data
acquisition system Electr Energ ManageTechnol 2016 9-13
[12] Chen C S, Duan S X, Cai T et al 2011 Online 24-h solar power forecasting based on weather
type classification using artificial neural network Sol Energy 85(11) 2856-70
[13] Dai Q, Duan S X, Cai T et al 2011 Short-term PV generation system forecasting model without
irradiation based on weather type clusting Proc CSEE 31(34) 28-35
[14] Wu S, Pan F M et al 2014 SPSS Statistical Analysis (Tsinghua University Press) pp 315-30
[15] Mohamed A and Badrul C 2015 Solar power forecasting using artificial neural networks Proc of
The 47th Annual North American Power Symp (USA: Charlotte, North Carolina)
[16] Chen C S, Duan S X and Yin J J 2009 Design of photovoltaic array power forecasting model
based on neutral network Trans China Electrotech Soc 24(9) 153-8
9
IOP Conference Series: Earth and Environmental Science
E-mail: cksolar@163.com
Abstract. Solar irradiation and ambient temperature are characterized by region, season and
time-domain, which directly affects the performance of solar energy based car system. In this
paper, the model of solar electric cars used was based in Xi’an. Firstly, the meteorological data
are modelled to simulate the change of solar irradiation and ambient temperature, and then the
temperature change of solar cell is calculated using the thermal equilibrium relation. The above
work is based on the driving resistance and solar cell power generation model, which is
simulated under the varying radiation conditions in a day. The daily power generation and solar
electric car cruise mileage can be predicted by calculating solar cell efficiency and power. The
above theoretical approach and research results can be used in the future for solar electric car
program design and optimization for the future developments.
1. Introduction
Solar electric cars rely on the photovoltaic effect to produce electricity. In these types of cars, the
controller and battery are used to drive the motor by overcoming the driving resistance. The solar
energy characteristics are clean, pollution-free and inexhaustible. The application of solar energy in
the renewable based cars has broad prospects and has emerged to be a hot topic [1,2]. In 1984, the
solar electric car of Japan’s Tokai University won the competition named as World Solar Challenge by
using an average speed of 100 km/h [3]. Domestically, the research institutions of the universities
including Tsinghua University, Shanghai Jiao Tong University and Jilin University have also carried
out relevant research, and successfully produced a number of solar electric cars [4].
The core components of the solar electric car energy system are solar cells. The performance of the
battery is affected by solar radiation and temperature, and the power generation capacity changes with
time. Domestic scholars have done some research in proposing the corresponding solution to improve
the matching of the power system and the energy efficiency [5-7]. The prediction of energy balance
and utilization is still very complex because of many contributing factors such as change in solar
irradiation, temperature and environmental factors. These above-mentioned factors change in the form
of season, time domain, regional and domestic variations. Therefore, the theoretical based
performance prediction of the solar electric car energy system is carried out by using driving
resistance, solar cell power generation model, metrological data and thermal variations.
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
resistance Rr (N), air resistance Ra (N), acceleration resistance Rac (N) and slope resistance Rc (N),
the formulae are given in preceding sections [8-10]:
In the above formula, r - rolling resistance coefficient and W - total car weight (kg).
2
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
The principle of solar cell power generation is based on the semiconductor photovoltaic effect of solar
radiation directly converted to electricity. The maximum power generation Pmpp of the solar cell is
obtained under standard test conditions where the atmospheric thickness is AM 1.5, the temperature is
25°C, and the solar irradiance is 1000W/m2. Since, the actual performance of the solar cell is affected
by solar radiation and temperature, it is necessary to use the maximum power temperature coefficient
to correct the power at the actual temperature and obtain the final actual power according to the
approximate relationship with the solar irradiance [11].
In the above formula, ' - the maximum power temperature coefficient is generally around - 0.5%
/°C.
In the above formula, G on - the solar irradiance received by the solar cell (kWh/m2), its value can
be obtained according to the hourly meteorological data.
The average daily sun irradiation is the daily average of the total radiation for the month, and the
3
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
sunny day total solar irradiation is the amount of radiation in the sunny day, and the following formula
is used [13]:
rt /24 (a b cos )(cos cos s )(sin s cos s )1 (8)
In the above formula, rt - the ratio of total radiation in hour to total radiation, s - sunset angle, –
hour angle; the coefficient a, b are calculated as follows:
a 0.409 0.5016 sin(s / 3) (9)
The solar irradiance at different time can be obtained by calculating rt of different time and the
daily total radiation.
In June and December, for example, the daily total solar irradiation and sunny total solar irradiation
distribution with time is shown in figure 1:
1000 30
Jun.daily Jun.
900 25 Dec.
Jun.sunny
800 Dec.daily 20
700
Irradiance /w/m2
Temperature /℃
Dec.sunny
600 15
500 10
400
5
300
200 0
100 -5
0 -10
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time /h Time / h
Figure 1. Hourly solar irradiance. Figure 2. Hourly ambient temperature.
The data show that [14,15], the ambient temperature in the day can be approximated with a cosine
function. Moreover, the maximum temperature generally occurs at 14:00, and the lowest temperature
appears in the 1 hour before the sunrise. According to the daily average temperature, temperature
differences and the change in sunrise time in the months of June and December are shown in figure 2.
4
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
solar cell
The surface of solar electric car is equipped with 432 pieces of HH125M165-S2 monocrystalline
silicon solar cells. In addition, 144 pieces of these cells constitute a PV module and a total of three PV
modules are used. Under standard conditions, the power comes out to be 1223 Wp. Under the standard
condition, the efficiency of HD125M165-S2 monocrystalline silicon solar cell is 18.2%, peak
power Pmpp is 2.833Wp, and the maximum power temperature coefficient is -0.46%/°C.
The four NCR18650A lithium ion battery stacks are used in series to form a power storage unit. In
addition, sub string is made by joining 10 groups of these power storage units and 45 groups of sub-
string are then connected in parallel. The total capacity of the battery pack is 2.09 kWh, the maximum
discharge current is 1395 A and the total weight is 81.9 kg.
The information relevant to the solar electric model car include temporary cruise speed of 30 km/h
(8.33 m/s), rolling resistance coefficient of 0.015, the wind area of 0.5 m2, air resistance coefficient
0.35, and air density is 1.205 kg/m3.
According to the formulas (1)-(6), the rolling resistance is 51.5 N and the air resistance is 7.3 N.
Therefore, the total resistance is 58.8N and the required power is 489.8 W. The different efficiencies
come out to be 95%, 90%, 95% and 75% for motor, mechanical transmission, MPPT solar controller
and battery respectively. In addition, the required solar power supply is 804 W.
The formula on the left side is the heat energy absorbed by the solar cell and the right side shows
the release of heat which includes the radiant heat to the sky and the convective heat transferring with
the environment. In the above formula, - solar cell heat radiation emissivity is set to 0.95; G on - solar
irradiance obtained by solar cells. According to meteorological data obtained by time; - solar cell
photoelectric conversion efficiency and - Stephen-Boltzmann Constant, 5.76×10-8W/m2∙K4. Tsky -
Sky temperature, calculated as follows:
Tsky 0.0552Ta1.5 (12)
Ta is the ambient temperature according to the meteorological data obtained by time. While, h - air
convection heat transfer coefficient is calculated by the following formula:
h 2.8 3.0vw (13)
Where vw is the surface wind speed which is approximately equal to the cruising speed of 8.33 m/s.
In June and December, for example, by using the formula (11) the changes in the solar cell
5
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
50
Jun.daily
45
Jun.sunny
40
Dec.daily
35
Dec.sunny
Temperature /℃ 30
25
20
15
10
5
0
-5
-10
0 2 4 6 8 10 12 14 16 18 20 22 24
Time /h
Figure 4. Hourly solar cell temperature.
22 Jun.daily
3.0 Jun.daily
Jun.sunny
21 Jun.sunny
Dec.daily 2.5 Dec.daily
Dec.sunny
Peak Power /Wp
20 Dec.sunny
Efficiency /%
2.0
19
1.5
18 1.0
17 0.5
16 0.0
0 2 4 6 8 10 12 14 16 18 20 22 24 0 2 4 6 8 10 12 14 16 18 20 22 24
Time /h Time /h
Figure 5. Hourly solar cell efficiency. Figure 6. Hourly peak power of solar cell.
The time of 12 o’clock can be taken as symmetry for the solar power produced all around the day.
By the impact of ambient temperature hysteresis, the power of the morning session is slightly higher
than the afternoon. The average efficiency in June was 18.0% and 17.7% with average power of 0.960
W and 1.378 W respectively. On the other hand, the average efficiency in December was 20.1% and
19.9%, with the average power of 0.893 W and 1.164 W respectively.
The solar cell efficiency and power change with time from January to December. The total
theoretical solar power generation from January to December is shown in figure 7. Further, according
to the needs of the solar electric car photovoltaic power supply, from January to December, the solar
electric car daily cruise stroke is shown in figure 8.
In the whole year, the maximum daily generation and cruise travel took place in May and the
minimum occurred in December. In June, for example, the performance prediction results show that
the average daily generating capacity of 432 solar modules is 6.33 KWh and 9.10 KWh under normal
daily radiation and sunny day radiation respectively. According to the cruise speed of 30 km/h and the
required solar power supply capacity of 804.0 W, the maximum theoretical travel is 232.0 km and
6
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
333.15 km with cruising time of 5.5 hours and 7.9 hours respectively. It should be noted that the actual
cruise travel and time will be less than the above forecasted results due to acceleration and slope
resistance.
10 250
Daily Daily
9 Sunny Sunny
7
150
6
5
100
4
3 50
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Month Month
6. Conclusion
Based on the concept model of solar electric car, this paper studies the performance prediction of
energy system, and can get the following conclusions:
The model and method established in this paper can be used to predict the performance of
solar energy car energy system in different months. It can be used to verify and analyze the
solar electric car in the developing stage. It can provide base for the selection and
improvement of the scheme.
The performance of the solar electric car energy system changes in a day with the time of 12
o’clock as the center. The 12 o’clock can be considered as symmetrical point. However, the
lag of environmental temperature results into the morning value slightly greater than afternoon.
Affected by changes in solar energy and ambient temperature in different months, the working
capacity of solar energy vehicle energy system has nearly doubled the scope of the design.
And the applications of the solar electric car system will have greater impact in the future.
References
[1] Kelly 2002 Raycing with the sun: Solar cars thrive in summer heat JOM Journal of the
Minerals, Metals and Materials Society 54 14-7
[2] Peter P and Phil H 2002 Critical speed control of a solar car Optimization and Engineering 3
97-107
[3] Surhone L M, Tennoe M T, Henssonow S F, Racing S C and Highway S 2010 World Solar
Challenge (Saarbrücken: Betascript Publishing)
[4] Zhao Y F, Wan J, Zhu Z P, Chen J, Le Z G, Ma F W, Ding Y and Zhao F Q 2011 The
application of solar cells on vehicle Agricultural Equipment and Vehicle Engineering 237
39-42
[5] Shi W K, Lv D H and Liang T Y 2008 A foundation investigation of parameters matching for
power train smission of solar electric car Acta Energlae Solaris Sinica 29 959-64
[6] Wang X C, Zhang P S, Zhang L H and Jia W 2009 Research on power control system of solar
electric vehicle Machinery and Electronics 12 36-9
[7] Wang X Y and Kang L Y 2010 The research and development of maximum power point tracker
for solar vehicle Power Electronics 44 12-3
[8] Guerra-zubiaga D A, Ramon-Raygoza E D, Rios-Soltero E F, Tomovic M and Molina A A
7
NEFES 2017 IOP Publishing
IOP Conf. Series: Earth and Environmental Science1234567890
93 (2017) 012025 doi:10.1088/1755-1315/93/1/012025
2009 PLM tools taxonomy to support product realization process: A solar racing car case
study Product Realization 1 1-26
[9] Japan Solar Energy Association 2010 Mechanism and Production of Eco-electric Vehicle (Xi'an:
Xi'an Jiaotong University Press)
[10] Hiroyuki O, Sumio N and Dai H 1998 Development of aerodynamics for a solar race car JSAE
Review 19 343-9
[11] Luo Y J, He Z N and Wang C G 2005 Solar Energy Technology (Beijing: Chemical Industry
Press)
[12] NASA: Surface meteorology and Solar Energy [EB/OL] http://eosweb.larc.nasa.gov/cgi-
bin/sse/sse.cgi?na
[13] Zhang H F 2004 Solar Energy Heat Utilization Principle and Computer Simulatio (Xi'an:
North-western Polytechnical University Press)
[14] Akasaka H et al 2000 Development of expanded AMeDAS weather data for building energy
calculation in Japan ASHRAE Transactions 206 455-6
[15] Tsinghua University, Tongji University, Xi'an Institute of metallurgy and architecture and
Chongqing Institute of civil engineering and architecture 1986 Air Conditioning (Beijing:
China Architecture and Building Press)
[16] Yang S M and Tao W Q 2006 Heat Transfer (Beijing: Higher Education Press)