International JournalOf Advances in Engineering and Management(IJAEM)) Volume 1, Issue 2, August - 2014. IJAEM © 2014 www.ijaem.org Forecasting of solar radiation by using artificial neural networks and regression methodsfor Manisa region in Turkey Bekir Cirak* *Siirt University, Faculty of Engineeringand Architecture, Department of MechanicalEngineering, Kezercampus, Siirt / TURKEY -------------------------------- ● -------------------------------- Abstract: In this study, the methods of Artificial Neural Networks (ANN) and Regression Analysis (RA) were used in estimatingmonthly average daily solar radiation arriving in Manisa.The back-propagation ANN is used to establish the nonlinear multivariate relationships between the solar radiation parameters and responses.Meteorological and region data like monthly average value of radiation, monthly average daily hours of bright sunshine, day length, relative humidity, wind speed, temperature and declination angle. Mean bias error (MBE), root mean square error (RMSE) and t-statistic methods were used to evaluate performance of the estimation.A comparison between the result of the proposed ANN and that of a (RA) is conducted in this study. The comparison shows that the performance of the (RA) is more effective than the ANN in finding the optimal solar radiation parameters. It was seen at the end of the study that the equation obtained through (RA) method yielded better performance than that of obtained through ANN method. Keywords: Artificial Neural Networks, Regression Analysis, Solar Radiation. -------------------------------- ● -------------------------------- 1. Introduction Total daily solar radiation is considered as the most important parameter in the performance prediction of renewable energy systems, particularly in sizing photovoltaic (PV) power systems, agriculture and building design applications [1].In spite of the importance of solar radiation measurements, this information is not readily available due to the cost and maintenance and calibration requirements of the measuring equipment [2]. Therefore, using meteorological and geographic data in predicting monthly average daily global solar radiation arriving on horizontal surface is a practical method. Literature contains various models constituted through both of ANN and regression methods in estimation of solar radiation by using meteorological and geographic parameters. Şen [3] used Angström-type equation and the third degree non-linear equation developed by by him to estimate solar radiation in his study, which was conducted in eight stations in Turkey, and determined, as a result of the study, that the non-linear equation developed by him has very lower error rate. Rehman and Mohandes [4] conducted a study by using temperature and relative humidity values, which had been measured through ANN, to estimate global solar radiation and found that temperature and relative humidity measurements may be used effectively in predicting magnitude of solar radiation. Deniz and Atik [5] conducted a study by using some of the meteorological data obtained in Zonguldak in a 10-year time period between 1995 and 2004 to estimate average magnitude of solar radiation through ANN and regression analysis methods and ANN method produced R 2 value as 0.9986 while it was calculated as 0.9999 for the model constructed through regression method. It was reported at the end of this study that relative error rates occur at lower levels in regression analysis. Mubiru and Banda [6] compared the model, which was established in their study on estimation of monthly average daily global solar radiation by using ANN method, with empirical expressions in the literature and found that the expression, which was obtained through ANN, may be used effectively in estimation of solar radiation. Mubiru [7] used ANN method in another study, which was conducted on estimation of monthly average daily solar radiation arriving on horizontal surface in Uganda and he found correlation value as 0.997 at the end of the study. Ülgen and Hepbaşlı [8] tested different models in determining diffused radiation amount for metropolises of Turkey (Istanbul, İzmir and Ankara).Güreland Ergün studied Estimation of global solar radiation on horizontal surfaceusing meteorological datain Rize [9]. The present study was conducted to investigate usability of artificial neural network and regression analysis methods, which are among the methods used in calculating magnitude of solar radiation, in calculation of actual data. Thus, meteorological and geographic parameters measured inManisa City were provided and their means were calculated to estimate magnitude of solar radiation for the year of 2013 by using two different methods and the obtained results were compared. *Corresponding author:B. Cirak, Tel: +90-484-216-4008,E-mail: bekircirak@mynet.com