You are on page 1of 6

IPASJ International Journal of Computer Science (IIJCS)

A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

A comparative study of ANN and angstrom Prescott model in the context of solar radiation analysis
JUHI JOSHI1 , VINIT KUMAR2
2 1 M.Tech, SGVU, Jaipur, India Assistant. Professor, SGVU, Jaipur, India

ABSTRACT
Since the people are moving toward the Green Energy and the solar energy is the most vernacular source of green energy .so it is necessary to have knowledge about solar radiation to install solar system at their location .The four cities Ahmadabad (lat 23.07, long 72.63), Bangalore (lat 12.95 long 77.63), dehradoon (lat 30.31 long 78.03) and Kolkata (lat 22.65 long 88.45) are selected across the India to estimate the monthly average global solar radiation. The global solar radiation data were collected courtesy of NASA government site. The objective of this study is to estimate the monthly average global solar radiation using Angstrom Prescott model and to predict the solar radiation using ANN. Prediction is done by using artificial neural network fitting tool. In ANN model, four parameter namely sunshine hour, latitude, longitude and altitude were used to predict monthly average global solar radiation on a horizontal surface for the location. Perhaps in Angstrom Prescott model only sunshine duration were used estimate monthly average global solar radiation. Levenberg-Marquard algorithm was used for analysis. The results of ANN model and Angstrom Prescott model are compared on the basis of mean square error (MSE) and regression value ( R 2). The MSE and R2 values for Angstrom Prescott model are - Ahmadabad (0.4225,4.3965) , Bangalore(0.0059,0.0149) ,Dehradun(0.1024,1.404) and Kolkata(0.0625,0.0498). and The MSE and R2 values for ANN model are- Ahmadabad(0.002,0.99),Bangalore(0.006,0.98),dehradun(0.01,0.90) and Kolkata(0.006,0.99). the comparison between ANN and Angstrom Prescott model have shown the vantage of the proposed ANN prediction model . KEYWORDS solar radiation, artificial neural network, multilayer Perceptron, Angstrom Prescott model,

1. INTRODUCTION
The amount of solar radiation, received at the earth is capable to converting from the form of heat and light into thermal energy .which could be used in photovoltaic system solar thermal power to generate electricity .therefore it is must to have knowledge about solar radiation for the solar energy system. In practical studies it is logical to consider that the solar radiation is directly proportional to the sunshine duration. [1]. in places where no measured values are available, a common application has been to determine this parameter by appropriate correlations which are empirically established using the measured data. [2].Utilization of the two most usual renewable energy sources, wind and solar energy, is more challenging than formal energy sources. Neural network is used because of many of the artificial intelligence techniques have the potential for making better, quicker and more practical predictions than any of the traditional methods. In this study analysis is based on past history data is therefore likely to be better understood and appreciated by designers than other theoretical and empirical methods. Due to rapid increase of solar power generation, the predictions of incoming solar energy are gaining more importance. Solar radiation is an essential parameter in solar energy application due to generation from photovoltaic (PV) is directly related to this parameter. Generally, solar radiation varies nonlinearly because of atmospheric effects such as cloudy weather, rain, humidity etc. Hence estimation of solar radiation is an fetching issue in solar energy field. 1.1 Angstrom Prescott model Prescott modified angstrom model in 1940 and it is called Angstrom- Prescott model: (1) Where 0 is the monthly average daily extraterrestrial radiation. The ratio of solar radiation at the surface of the Earth (H) to extraterrestrial radiation ( 0), that is, / 0, is called the Clearness Index. Values of the monthly average daily extraterrestrial radiation ( 0) are calculated from the following 2 (2) 2 Where sc = solar constant with a value of 1367W/m , = day of the year from January 1 to December 31 taking January 1st as 1, = latitude of the location, = declination angle, = sunset hour angle, and 3) Declination angle ( ) can be obtained by the equation given by Cooper in 1969 (4)

Volume 2 Issue 3 March 2014

Page 1

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992


(5)

The maximum possible sunshine duration N (hours) for a horizontal surface is given by a and b are the coefficient of angstrom Prescott model

Thus, coefficients of the Angstrom-Prescott model for India can be estimated using these coefficients even if only sunshine data is available. Where i is date of the month. 1.2 About Artificial Neural Network In multilayer feed forward neural network there is a layers input units, hidden units, and one output layer of units. A neural network without hidden units is called a Perceptron. In this study, the four cities are picked out across the India (east, west, north, and south) to estimate the monthly average global solar radiation. The four cities are Ahmadabad (lat 23.07, long 72.63), Bangalore (lat 12.95 long 77.63), dehradoon (lat 30.31 long 78.03) and Kolkata (lat 22.65 long 88.45).The monthly average global solar radiation value on horizontal surface, H, were collected from NASA government website. The 4 parameters are altitude, longitude, altitude and sunshine duration. these parameter are divided into three sample : 8 are used for training , 2 are used for validation and 2 are used for testing .this process follow is followed by all the cities .In angstrom Prescott model, global solar radiation is calculate for each day of 12 month (January to December) for 4 cities. Then monthly average of estimated global solar radiation is calculated. Obtained solar radiation values from were converted from MJ/M2 to Kwh/m2/day. Four neurons in the input layer corresponding with their input parameters and units are discussed below. 20 neuron are present in hidden layer and 1 neuron in output layer.

2. METHODOLOGY
India lies in sunny belt between 6 N and 32 N latitudes, its geographical position favors the development and utilization of solar energy. [3]
Define Input and output Split the data into training and testing data Design ANN structure Initial weight, bias and transfer function are selected Train the network with neural network fitting tool Get predicted values Evaluate the MSE and R 2 error Select the best output (least error)

Figure 1: ANN process used in the study 2.1 Analyzing the data using NFTOOL Multilayer Perceptron neural network was used in this study, with four input Parameter latitude, longitude, altitude and sunshine duration. There is following procedure to develop artificial neural network model.

Figure 2 MLP structure used in the study

Volume 2 Issue 3 March 2014

Page 2

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

Step 1: Create excel files for input and target data Step 2: Convert this excel file into .csv format. Step 3: Open the input and target .csv file in MATLAB software by using following command; uiopen (path\filename) uiopen(D:\report on neural network\matlab\initial.csv) uiopen(D:\report on neural network\matlab\target.csv) Where initial and target are respectively input and target data file.

Start

Open input and target .csv file in matlab software

Open neural network fitting tool (nftool)

nftool design neural network

Neural network train by using levenberg marquardt back propagation algorithm

Calculate MSE and R2

If MSE <= 0 and R2 =1 ;

Network is fit,

Stop

Figure 3: flow chart for ANN Step 4: open neural network fitting tool (nftool) in MATLAB. It divides the data into 3 samples data: training, validation and testing. Step 5: now neural network is designed having 20 neuron in hidden layer. And select the weight connections, bias and transfer function. Basically t bias and weight connection select randomly. By default the activation function for hidden layer is tansig and for output layer is purelin. Step 6: levenberg marquardt back propagation (trainlm) algorithm was used to train the network. Step 7: calculate the Mean square error and regression values Step 8: train the network until target reaches its maximum value. Table 1 Summary of annual average of measured and calculated values of monthly global solar radiation by artificial neural network.
cities Ahmadabad Bangalore Dehradun Kolkata Hm(Kwh/m2 /day) 5.32 5.25 5.32 4.67 Hc( Kwh/m2/day) 5.16 5.33 5.43 4.79 ERROR(%) 3 -1 -2 -2

Volume 2 Issue 3 March 2014

Page 3

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

2.2 Estimation of monthly global solar radiation using Angstrom Prescott By using equation 1, monthly average global solar radiation (H) is calculated. Monthly average daily extraterrestrial radiation ( 0) is calculated from the equation 2. These values are calculated for the period of 2005. There are following table which shows the Summary of monthly mean average of sunshine duration, extraterrestrial solar radiation, measured and calculated values measured and calculated clearness index and relative percentage error for each city. Table 2 Summary of annual average of sunshine hours, extraterrestrial solar radiation measured and calculated values, measured and calculated clearness indexes and relative percentage error
Error Cities Ahmadabad Dehradun Kolkata Bangalore N 8. 7. 5. 6. Hm 5.3 1 5.3 2 4.6 7 5.2 5 Hc 4.6 7 5.2 4 4.3 4 4.9 9 H0 9.0 8 8.6 5 9.1 9.5 7 Hm/H0 0.59 0.62 0.52 0.55 HC/H0 0.53 0.56 0.48 0.53 12 1.5 7.0 4.9

4. RESULT
This tables show the values of two model, which are used in this study. ANN model results are more precise than Angstrom Prescott model. Table 3 Comparison between annual average monthly global solar radiation of Measured, Angstrom Prescott and ANN model
model Ahmadabad Bangalore Dehradun Kolkata Measured (Kwh/m2/day) 5.3 5.2 5.3 4.6 ANN((Kwh/m2/day) 5.2 5.3 5.4 4.7 Angstrom Prescott (Kwh/m2/day) 4.7 4.9 5.2 4.3

The coefficient of determination or regression value (R2), root mean square error (RMSE) were influenced and compared. A new approach is projected for the prediction of the future observations of the monthly global solar radiation without applying other meteorological parameters. The MLP neural networks and the Angstrom Prescott model is then compared. The results of MSE and R2 of Angstrom Prescott models are : Ahmadabad(0.4225,4.3965),Bangalore(0.0059,0.0149),dehradun(0.1024,1.404) and kolkata(0.0625,0.0498). and The MSE and R2 values for ANN model are Ahmadabad(0.002,0.99),Bangalore(0.006,0.98),dehradun(0.01,0.90) and Kolkata(0.006,0.99). Table 4 error analysis
model cities Ahmadabad Dehradun Kolkata Bangalore angstrom Prescott MSE 0.422 5 0.005 9 0.102 4 0.062 5 R2 4.396 5 0.014 9 1.404 0.049 8 ANN MSE 0.00 2 0.01 0.00 6 0.00 6 R2 0.9 9 0.9 0 0.9 9 0.9 8

The results show that the MLP neural networks improve the accuracy of the prediction. The study depicts that selected ANN neural model has lower RMSE value than empirical model. And regression value is closer to 1 in ANN model, which show the good agreement. The sunshine duration and monthly average global solar radiation data for 12 month were used for training, testing and validation purpose. First, a multilayer Perceptron model was trained to predict the global solar radiation based on the monthly average sunshine duration, latitude, longitude and altitude. After several experiments it was found that a network with four inputs, 20 hidden neurons in one layer, and one output unit was sufficient for such an application. Regression plot show that target values are enough close to fit line

Volume 2 Issue 3 March 2014

Page 4

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

4 (a) Ahmadabad

4 (b) Bangalore

4(c) Dehradun 4 (d) Kolkata Figure 4 Best lines fit between actual and predicted solar radiation for different cities

5 (a)

5(b)

5(c)

5(d)

Figure 5 Comparison graph between monthly average global solar radiation of Measured, Angstrom Prescott and ANN model

6(a)

6(b)

Figure 6 Comparison of MSE and R2 between Angstrom Prescott and ANN model result

REFERENCES
[1] R. C. Srivastava and Harsha Pandey, Estimating Angstrom-Prescott Coefficients for India and Developing a Correlation between Sunshine Hours and Global Solar Radiation for India, Hindawi Publishing Corporation ISRN Renewable Energy Volume 2013, Article ID 4037427, [2] A.M. Muzathik1, W.B.W. Nik1, M.Z. Ibrahim, K.B. Samo1, K. Sopian and M.A. Alghoul. DAILY GLOBAL SOLAR RADIATION ESTIMATE BASED ON SUNSHINE HOURS. International Journal of Mechanical and Materials Engineering (IJMME), Vol.6 (2011), No.1, 75-80

Volume 2 Issue 3 March 2014

Page 5

IPASJ International Journal of Computer Science (IIJCS)


A Publisher for Research Motivation ........

Volume 2, Issue 3, March 2014

Web Site: http://www.ipasj.org/IIJCS/IIJCS.htm Email: editoriijcs@ipasj.org ISSN 2321-5992

[3] T. Krishnaiah, S. Srinivasa Rao, K. Madhumurth y, K.S. Reddy, Neural Network Approach for Modelling Global Solar Radiation, Journal of Applied Sciences Research, 3(10): 1105-1111, 2007] [4] Prediction of Global Solar Radiation in Abu Dhabi Ali Assi,Mohammed Jama, andMaitha Al-Shamisi, International Scholarly Research Network ISRN Renewable Energy , Volume 2012, Article ID 328237, 10 pages doi:10.5402/2012/328237 [5] Notes on Multilayer, Feedforward Neural Networks CS494/594: Projects in Machine Learning Spring 2006 Prepared by: Lynne E. Parker [6] Using MATLAB to Develop Artificial Neural Network Models for Predicting Global Solar Radiation in Al Ain City UAE Maitha H. Al Shamisi, Ali H. Assi and Hassan A. N. Hejase United Arab Emirates University United Arab Emirates [7] T. V. Ramachandra, Solar energy potential assessment using GIS, Energy Education Science and Technology 2007 Volume (issue) 18(2): 101-114. [8] Emmanuel A. Sarsah, Felix A. Uba Monthly-Specific Daily Global Solar Radiation Estimates Based On Sunshine Hours In Wa, Ghana, INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 2, ISSUE 8, AUGUST 2013, 2013 ISSN 2277-8616 246 [9] Auburn University. Levenberg-Marquardt Training.[Online]. [10] Neural Network Toolbox for Use with MATLAB Howard Demuth Mark Beale, Users Guide. [11] Anuradha Chug and Ankita Sawhney , The Comparative Study of Forecasting Analysis Based on Backpropogation (NFTOOL and NTSTOOL) and Adaptive nuero fuzzy interference system ANFIS Software Maintenance , International Journal of Advancements in Research & Technology, Volume 2, Issue 5, M ay2013 [12] Biological Neurons and Neural Networks, Artificial Neurons .Neural Computation: Lecture 2 John A. Bullinaria, 2013.

Volume 2 Issue 3 March 2014

Page 6

You might also like