Professional Documents
Culture Documents
I. INTRODUCTION
The most promising method for PV forecasting is the artificial neural network method [10][11], especially Back Propagation(BP) neural network. BP neural network has the advantages of complex nonlinear systems simulation, good approximation performance, strong learning ability and fault data tolerance. However, it also has some inherent defects, such as
slow convergence rate, easy to fall into local minimum value
and cannot get the global optimal solution, some improvements
have been made on BP network to enhance better convergence[12] -[15].
While predicting PV power outputs, few researches have
taken aerosols into consideration, which can be induced by
desert dust, biomass burning, volcano smoke, power plant
emission, and so on. In fact, there is a strong correlation between total solar radiation and aerosol index [16]-[18]. Typically, the higher the aerosol optical depth is, the greater the
attenuation of the solar power reaching the surface of the earth
is. Therefore, aerosols influence the output of the PV system
indirectly.
In this paper, a novel PV forecasting model is proposed,
based on multiple meteorological data including aerosol index
values. Firstly BP neural network is trained by historical meteorological data and hourly average output power data, and
then weather data of the objective day are used as the input
variables of BP neural network to predict hourly power outputs of PV systems. The efficiency of the proposed method is
validated by analyzing the mean absolute percentage error
between predicted values and measured values.
II. FORECASTING MODEL FOR PV OUTPUT POWER
Since there are various factors affecting PV power outputs, it is difficult to figure out the power outputs with a fixed
mathematical model. The unstable and nonlinear relationship
between power outputs and influencing factors implies that
different weather types will lead to significant variance in PV
x1 x12
A simple linear correlation analysis is given for a twomonth period of data. As shown in Fig. 1, the daily total PV
generated energy has a correlation coefficient of -0.321 with
AI.
x13
Maximum humidity
x14
Minimum humidity
x15
Average humidity
x16
Maximum temperature
x17
Minimum temperature
x18
Average temperature
x19
x20
Aerosol index
Y =
[y , y
1
With the weather reports of the objective day, the PV forecasting model is able to select related data to predict the
power outputs. Specifically, the forecasting model will find
similar days with the same weather type in the same time period (15 days separately before and after the objective day) of
last three years as training samples. The solar radiation of these
days is supposed to be closest, so PV output curve will also be
similar in these days.
The input variables used in our forecasting model are described in TABLE I, including historical meteorological data
and PV output power data from similar days. Besides, AI data
are also chosen as an input variable. Comparisons will be made
with the average forecast errors of the forecasting model which
does not take aerosols into account.
TABLE I
unit y k is W = [ w1 , w2 , wn ] .
For output layer,
yk = f ( nk ) , k = 1, 2,..., l
T
nk = w jk h j + b, k = 1, 2,..., l
(2)
(3)
j =1
h j = f ( n j ) , j = 1, 2, ..., m
(4)
nj =
v x
ij
+ b, j = 1, 2,..., m
(5)
j =1
The transfer function of the neurons is the sigmoid function, which can be expressed as follows:
f ( x) =
1
1 + e x
(6)
A. Sunny Weather
Usually there is not a universal method to decide the number of hidden layer neurons. It is chosen to be 9 neurons in this
study, after repeatedly testing of the forecasting performance of
the neural network. The gradient decent method is applied in
the training process, which has a better convergence rate. Parameters of the network are shown in TABLE II.
TABLE II
PARAMETERS OF BP NEURAL NETWORK
BP Variables
Numbers
19 (without AI)
Number of input-layer neurons
20 (with AI)
Number of hidden-layer neurons
12
(a)
xi =
xi xmin
xmax xmin
(7)
(b)
Fig. 4. Sunny Weather Without AI : (a) Output Power (b)Percentage Error
Where, xi is the ith component in the original input data vector; xmin and xmax are the minimum and maximum input data
separately?
In order to evaluate the accuracy of the estimated results,
Mean Absolute Prediction Error (MAPE) is introduced. MAPE
represents the variation of predicted values around measurement data and indicates the performance of the short-term forecasting.
MAPE =
100
N
i =1
| Pi f Pi a |
Pi a
(a)
(8)
(b)
Fig. 5. Sunny Weather With AI: (a) Output Power (b)Percentage Error
B. Cloudy Weather
PV active power outputs on cloudy days are smaller than
those on sunny days and the MAPE is larger in Fig. 6(a) and
Fig. 7(a).
(a)
VI. REFERENCES
[1]
[2]
[3]
(b)
Fig. 6. Cloudy Weather Without AI : (a) Output Power (b)Percentage Error
[4]
[5]
[6]
[7]
(a)
[8]
[9]
[10]
(b)
Fig. 7. Cloudy Weather With AI : (a) Output Power (b)Percentage Error
TABLE III
AVERAGE MAPE VALUES UNDER DIFFERENT WEATHER CONDITIONS
Without AI
With AI
Sunny
Cloudy
5.47%
10.33%
[11]
[12]
3.96%
6.57%
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
Lorenz E, Hurka J, Heinemann D, et al. "Irradiance forecasting for thepower prediction of grid-connected photovoltaic systems." IEEE Journal
of Selected Topics in Applied Earth Observations and Remote Sensing,
2009, 2 (1): 2-10
Kudo M, Nozaki Y, Endo H, et al. "Forecasting electric power generation in a photovoltaic power system for an energy network." Electrical
Engineering in Japan, 2009, 167 (4): 16-23
H. K. Elminir, Y. A. Azzam and F. I. Younes, "Prediction of hourly and
daily diffuse fraction using neural network, as compared to linear regression models," Energy, 2007, vol.32, no.8, pp.1513-1523.
Sulaiman S. I., Rahman T. K. A., Musirin I., et al. "Artificial neural
network versus linear regression for predicting Grid-Connected Photovoltaic system output. " Cyber Technology in Automation, Control, and
Intelligent Systems, 2012 IEEE International Conference on: 170-174.
Li Y, Niu J. "Forecast of power generation for grid-connected photovoltaic system based on Markov chain." Power and Energy Engineering
Conference, APPEEC 2009. Asia-Pacific. IEEE, 2009: 1-4.
Ying-zi L, Ru L, Jin-cang N. "Forecast of power generation for gridconnected photovoltaic system based on grey model and Markov chain."
Industrial Electronics and Applications, ICIEA 2008. 3rd IEEE Conference on. IEEE, 2008: 1729-1733.
Junseok S, Krishnamurthy V, Kwasinski A., et al. "Development of a
Markov-Chain-Based Energy Storage Model for Power Supply Availability Assessment of Photovoltaic Generation Plants." IEEE Transactions on Sustainable Energy, 2012,PP(99):1-10.
Li Ran, Li Guang-min. "Photovoltaic power generation output forecasting based on support vector machine regression technique." Electric
Power, 2008, 41(2): 74-78.
Shi J, Lee W J, Liu Y, et al. "Forecasting Power Output of Photovoltaic
Systems Based on Weather Classification and Support Vector Machines." IEEE Transactions on Industry Applications, 2012, 48(3):
1064-1069.
F. Wang, Z. Mi, S. Su and C. Zhang, "A practical model for single-step
power prediction of grid-connected PV plant using artificial neural network," in Innovative Smart Grid Technologies Asia (ISGT), 2011 IEEE
PES, pp.1-4.
Y. Ting-Chung and C. Hsiao-Tse, "The forecast of the electrical energy
generated by photovoltaic systems using neural network method," in
Electric Information and Control Engineering (ICEICE), 2011 International Conference on,pp.2758-2761.
F. Bizzarri, M. Bongiorno, A. Brambilla, G. Gruosso and G. S. Gajani,
"Model of Photovoltaic Power Plants for Performance Analysis and
Production Forecast," IEEE Transactions on Sustainable Energy, vol.,
no.99, pp.1-8.
C. Chen, S. Duan, T. Cai and B. Liu, "Online 24-h solar power forecasting based on weather type classification using artificial neural network,"
Solar Energy, vol.85, no.11, pp.2856-2870.
E. zgi, A. ztopal, B. Yerli, M. K. Kaymak and A. D. ahin, "Short
mid-term solar power prediction by using artificial neural networks," Solar Energy, vol.86, no.2, pp.725-733.
A. Yona, T. Senjyu, A. Y. Saber, T. Funabashi, H. Sekine and K. ChulHwan, "Application of neural network to 24-hour-ahead generating
power forecasting for PV system," in Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st
Century, 2008 IEEE, pp.1-6.
D. Calinoiu, M. Paulescu, I. Ionel, N. Stefu, N. Pop, R. Boata, A. Pacurar, P. Gravila, E. Paulescu, and G. Trif-Tordai, "Influence of aerosols
pollution on the amount of collectable solar energy," Energy Conversion
and Management, 2013, vol. 70, pp. 76-82.
S. Mekhilef, R. Saidur and M. Kamalisarvestani, "Effect of dust, humidity and air velocity on efficiency of photovoltaic cells," Renewable
and Sustainable Energy Reviews, 2012, vol. 16, pp. 2920-2925.
M. De Graaf and P. Stammes, "Global observations and spectral characteristics of desert dust and biomass burning aerosols," in Remote Sensing
of Atmospheric Aerosols, IEEE Workshop on, 2005, pp. 60-67.
Goddard Earth Sciences Data and Information Services Center [Online].
Available: http://disc.sci.gsfc.nasa.gov.
Solar Radiation Monitoring Laboratory, University of Oregon [Online].
Available: http://solardat.uoregon.edu.