METHOD OFARTIFICIAL NEURAL NETWORKS WITH PARAMETER ESTIMATING WIND SPEEDS BETWEEN CITIES IN TURKEY
Numan S. Çetin1, Cem Emeksiz2,
1Ege University, Institute of Solar Energy 35100, Bornova / IZMIR / TURKEY
2Gaziosmanpaşa University, Turhal Vocational High School 60300, Turhal/TOKAT/ TURKEY
E-mail: ,
- ABSTRACT
Artificial neural networks, for solving problems which can not be expressed as a complex and complete offers an alternative way . Neural networks learn from examples, errors and incomplete of noisy data that are tolerated and can be used for solving nonlinear problems.In this study, using the data of wind speeds received by the General Directorate of State Meteorology Affairs, wind speed forecasting for 4 cities (Tokat, Amasya, Corum, Yozgat) was performed. By using the learning algorithms (Back Propagation and Levenberg-Marquardt) were performed for training. The data used for training in the year 2000 until 2005. Between cities wind speed forecasting was performed in 2005.
Keywords: Artificial Neural Networks, Renewable Energy, Estimating Wind Speed
- INTRODUCTION
Energy is a key factor in economic development and in providing vital services that improve quality of life. Energy is required for meeting all of the basic needs such as food and health, agriculture, education, information, and other infrastructure services and shows clear correlation with the Human Development Index HDI figure 1 [1].
FIG. 1.HDI and commercial energy consumption
Even so unlimited use of fossil fuels (coal, oil, and natural gas) brings serious influences to the environment and the future development of society. Due to increasing exploitation, fossil fuels are becoming fewer and fewer [2]. In addition, emission of greenhouse gases from fossil fuels causes global warming and declining air quality [3]. It is estimated that 16 million tones of CO2 are emitted into the atmosphere every 24 h worldwide. If all emissions were to stop today, the CO2 that has already been emitted will result in an enhanced GHG effect for the next 50 years [4].
Boyle concludes that, renewable energy is a kind of energy which can be obtained from the continuous or repetitive currents of energy recurring in the natural environment or energy flows which are replenished at the same rate as they are “used” [5].Compared with fossil fuels, renewable sources of energy are not substantially depleted by continued use and have less pollutant emissions. Consequently, renewable energy sources are becoming more and more attractive to all nations, especially for all countries.
Among renewable energy sources, wind energy is the one with the lowest cost of electricity production [6], but is feasible only as long as weather conditions allow. Surrounded by the Black Sea to the north, the Marmara and the Aegean Sea to the west and the Mediterranean Sea to the south, Turkey has huge potential for wind power generation [7]. A study carried out in 2002 concluded that Turkey has a theoretical wind energy potential of nearly 90,000MW [8]. So far only about 1000MW capacity wind farms are in operation in Turkey, generating less than 0.5% of total electricity consumed. There are a number of cities in Turkey with relatively high wind speeds [9]. To maintain economical power expedition of wind generated electricity, it is significancy to be able to make short term predictions of future wind speeds, which directly affects generation capacity. Without this ability, a wind farm operator is prone to allocate more generation units or supplemental energy reserves than necessary in order to ensure budgeted electricity outputs are met [6], with an end result of increased operating costs.
Generally, wind speed uses numerical weather prediction for longer time. For shorter time prediction, only the historical data are used. The methods for shorter time prediction are: Artificial Neural Network Model [10-11-12], Persistence Model[13-14-15], Lineer Regression Model, Auto Regression and Moving Average Model [16-17], Auto Regression Least Square Model, Self-Adapting Fuzzy Logic Model, Kalman Filtering Model [18], Fuzzy Expert System [19] etc.
Several studies are presented to propose some models for wind speed prediction. In an earlier study, Njau developed an electronic system for air temperature and wind speed prediction and found good agreement between the predicted and actual values of both the wind speed and temperature [20]. Rehman and Halawani used stochastic time series analysis for the prediction of hourly wind speed for nine Saudi Arabian cities and found a relatively good agreement between predicted and actual values [21]. Also, Mohandas and Halawani introduced support vector machines (SVM) to wind speed prediction and compared their performance with the multilayer perceptron (MLP) neural networks [22].
This paper reports an ANN model for wind speed prediction, which uses back propagation algorithm and multi-layer feed forward type of network . In this paper, an ANN modeled wind speed (WS) prediction program was developed to simulate wind speed of a city using that of another cities, for each pair of cities and each combination between them.
- ARTIFICIAL NEURAL NETWORKS
The term artificial neural network (ANN) has been conducted right after the recognition of the way the human brain computes. The human brain computes in an entirely different way from the conventional computer, the brain is a highly complex, nonlinear and parallel information processing system. A neural network is a machine that is designed to model the way in which the brain performs a particular task. The network is implemented by using electronic components or is simulated in software on a digital computer. A neural network is a massively parallel distributed processor made up of simple processing units, which has a natural propensity for storing experimental knowledge and making it available for use.
A neuron is an information processing unit that is fundamental to the operation of a neural network. The three basic elements of the neuron model are:
1. A set of weights, each of which is characterized by a strength of its own. A signal “xj” connected to neuron “k” is multiplied by the weight “wkj”. The weight of an artificial neuron may lie in a range that includes negative as well as positive values.
2. An adder for summing the input signals, weighted by the respective weights of the neuron.
3. An activation function for limiting the amplitude of the output of a neuron. It is also referred to as squashing function which squashes the amplitude range of the output signal to some finite value figure 2.
FIG. 2.Model of ANN [23]
A neural network has 3 layers.The first layer is always the input layer and the last layer is always the output layer. The layers placed between the first and the last layers are the hidden layers.
In this study; multi-layer feed forward type of network is used. Feed forward type of network where computations proceed along the forward direction only Figure 3. Multiple layer feed forward network has been applied successfully to solve some difficult diverse problems by training them in a supervised manner with a highly popular algorithm known as the back-propagation algorithm. This algorithm is based on the error-correction learning rule. Most important reason for this; the capacity of learning is high and algorithm is simple [25].
FIG. 3. Typical feed forward network [24]
- MODEL OF ANN
In this study, we combined the wind speed data of four cities (Tokat, Amasya, Corum,Yozgat). We have estimated wind speeds of the fourth city by using wind speeds of the other three cities with the network that is created. Therefore four different types of training was performed. These training processes are also performed on the same network.
- TRANING PROCEDURES
In this study, wind speed data of four cities (Tokat, Amasya, Corum, Yozgat), collected by General Directorate of State Meteorology Affairs between 2000-2005, were used for wind speed prediction using ANN.
The wind speed data were normalized between 0-1. Normalization procedure was initiated before the start of the training of the selected data. Nonlinear approach that makes a significant feature of ANN that the data kept in a normalization of the course. Method chosen for the normalization of the data, it directly affects the performance of ANN. Because of normalization, the input data can be transferred in the active region the function to be passed to. Data normalization, the data provides cumulative totals constitute a negative block [26].
The network that is used for traning has an input layer with 3 neurons, a hidden layer with 10 neurons and a output layer with one neuron figure 4.
FIG. 4. Neural Network Diagram
The data for 180 months during 2000-2004 were used for each training purpose and the 12 months’data from 2005 as testing data. The testing data were not used in training the neural networks. The mean squared error (MSE) for these data was found for each traning. Mean square error defined by Gauss often just "average error" is also. Gives the most accurate idea about the degree of accuracy of measurements. The mean square error is the most commonly used in criteria for the degree of accuracy. Because the squares of errors taken, the effect of large error is more and small errors and large errors don’t deal equally.
- RESULTS
The result of education, the info window in figures 5-6-7-8 Regression values (R) of considering coefficients for four cities were evaluated. R determines the level of the relationship between the desired output of the network output. Take the values of R close to 1 indicates that the accuracy of the relationship between the desired output of the network output [27].
Regression graph of between the network output and desired output shown in figure 5 for Tokat city. As a result of the first training process graph obtained from the regression analysis; R = 0.90223 for the test data, R = 0.9218 for the training data and R = 0.96712 for the validation data were as. The mean-square error was calculated for Tokat city as 0.0164.
Regression graph of between the network output and desired output shown in figure 6 for Corum city. As a result of the first training process graph obtained from the regression analysis; R = 0.95078 for the test data, R = 0.92821 for the training data and R = 0.96388 for the validation data were as. The mean-square error was calculated for Corum city as 0.0487.
Regression graph of between the network output and desired output shown in Figure 7 for Yozgat city. As a result of the first training process graph obtained from the regression analysis; R = 0.99999 for the test data, R = 0.90325 for the training data and R = 0.99873 for the validation data were as. The mean-square error was calculated for Yozgat city as 0.0532.
Regression graph of between the network output and desired output shown in Figure 8 for Amasya city. As a result of the first training process graph obtained from the regression analysis; R = 0.97425 for the test data, R = 0.92038 for the training data and R = 0.95347 for the validation data were as. The mean-square error was calculated for Amasya city as 0.1025. And also for graps that obtained for measured and estimated wind speeds of 4 cities were shown in figures 9-10-11-12.
FIG. 5. Regression graphic for Tokat City
FIG. 6. Regression graphic for Corum City
FIG. 7. Regression graphic for Yozgat City
FIG. 8. Regression graphic for Amasya City
FIG. 9. Estimated and measured wind speeds of Tokat
FIG. 10. Estimated and measured wind speeds of Corum
FIG. 11. Estimated and measured wind speeds of Yozgat
FIG. 12. Estimated and measured wind speeds of Amasya
- CONCLUSION
In this study, forecasting was performed by using wind speeds of 4 different cities ın Turkey. Mean-square error was the lowest for wind speed prediction of Tokat city. Wind speed data for 4 different cities in Turkey between 2000 and 2004 were used for training a feed forward ANN using back propagation algorithm. Data for 12 months of the year 2005 were used to test the performance of the ANN system. The results obtained in this study show that while articial neural network investigated have the potential for short term wind speed prediction. The applicability of this technique to systems that have the potential for renewable energy sources. Parameters necessary to calculate the annual energy retrieval of system can easily be estimated with ANN. As a result, the system is reduced before the establishment of the necessary financial resources allocated to the taping and shortens the process of goal achievement. Future work forecasting by using different neural network is considered.
REFERENCES
[1]Rehling U, Karcher H, Pradhan M., ‘‘ Capacity building in developing countries – bringing renewable energy to the people’’, Science Forum 2004:6. Available online at: 04.pdf.
[2]V. Quaschning, ‘‘Understanding Renewable Energy Systems’’, London: Earthscan, 2005, p. 5.
[3]J. L.Seitz, Global Issues: An Introduction, 2nd ed., Blackwell, UK, 2002, pp. 118-128, 139.
[4]Kamaruzzaman Sopian, Baharuddin Ali, Nilofar Asim; ‘Strategies for renewable energy applications in the organization of Islamic conference (OIC) countries’ Renewable and Sustainable Energy Reviews 15 (2011) 4706– 4725.
[5]G. Boyle, Renewable Energy, 2nd ed., New York: Oxford, 2005, pp. 11- 12, 108, 207.
[6]A. Fabbri, T. Roman, 1. Abbad, V. Quezada, "Assessment of the Cost Associated With Wind Generation Prediction Errors in a Liberalized Electricity Market", IEEE Transactions on Power Systems, August 2005, pp. 1440-1446.
[7]I. Yuksel , K. Kaygusuz, ‘Renewable energy sources for clean and sustainable energy policies in Turkey’ Renewable and Sustainable Energy Reviews 15 (2011) 4132–4144.
[8]K. Kaygusuz, A. Kaygusuz, Renewable energy and sustainable development in Turkey. Renewable Energy 2002;25:431–53.
[9]Geothermal energy in Turkey. [accessed 20.06.08].
[10]Kariniotakis G., Stavrakis G., Nogaret E., Wind power forcasting using advanced neural network models [J]. IEEE Trans Energy Conversion, 1996, 11(4): 762-767.
[11]Mohamed A.Mohandes, Shariqur Rehman, Talal O.Halawani., A Neural Networks Approach for Wind Speed Prediction [J]. Renewable Energy, 1998, 13(3): 345-354.
[12]A. Sfetsos, A Novel Approach fort he Forecasting of Mean Hourly Wind Speed Time Series [J]. Renewable Energy, 2002, 27: 163-174.
[13]L. Landberg, Short-term prediction of the power production from wind farms [J]. Journal of Wind Engineering and Industrial Aerodynamics 1999,80: 207-220.
[14]L. Landberg, Short-term prediction of local wind conditions [J], Journal of Wind Engineering and Industrial Aerodynamics 2001,89: 235-245.
[15]Alexiadis M., Dokopoulos P., Sahsamanoglou H., et.al.,Short term forecasting of wind speed and related electrical power [J]. Solar Energy, 1998, 63(1): 61-68.
[16]M. Milligan, M. Schwartz, Y. Wan, Statistical wind power forecasting models: Results for U.S wind farms. Preprint: to be presented at WINDPOWER 2003 Austin, Texas, May 18-21, 2003.
[17]J. L. Torres, A. Garcia, M. De Blas, et al., Forecast of hourly average wind speed with ARMA models in Navarre (Spain) [J]. Solar Energy, 2005, 79: 65-77.
[18]Bossanyi E.A., Short-term wind prediction using Kalman filters [J]., Wind Engineering, 1985, 9(1): 1-8.
[19]I. G. Damousis, P. Dokopoulos, A fuzzy expert system fort he forecasting of wind speed and power generation in wind farms. European Union Project CARE: Advanced Control Device for Power Systems with Large Scale Integration of Rnewable Energy Sources.
[20] E.C. Njau ,"An electronic system for predicting air temperature and wind speed patterns". Renewable Energy 1994; 4(7):793-805.
[21] S. Rehman , T.O. Halawani ,"Statistical characteristics of wind in Saudi Arabia". Renewable Energy 1994; 4(8): 949-956.
[22] M. Mohandes ,"Support vector machines for wind speed prediction". Renewable Energy 2004; 29 : 939.947
[23] Kumar, M. (2009). SHORT‐TERM LOAD FORECASTING USING ANN TECHNIQUE. Rourkela: National Institute of Technology
[24] More A., Deo M.C., ‟Forecasting wind with neural networks’’, Marine Structures, 16, 35-49, 2003.
[25] Elmas, Ç., ‟Artificial Intelligence Applications”, Seçkin Publishing, Ankara, 2007.
[26] Saraç, T., ‟Artificial Neural Networks”, Gazi University Department of Industrial Engineering, Seminar Project, June 2004, ANKARA.
[27] Yaprak, H., Karacı, A.,‟After High Temperature The Polypropylene Fiber Reinforced Concrete Pressure Strength Prediction Using Artificial Neural Networks’’, Int.J.Eng.Research & Development,Vol.1,No.2, June, 2009.