Estimation of Monthly Mean Daily Global Solar Radiation over Bangkok, Thailand using Artificial Neural Networks

Thailand is located in an equatorial belt that receives abundant solar energy. In order to achieve the optimum utilization of solar energy available, it is necessary to evaluate the incident solar radiation over the region of interest. Solar radiation can be assessed by means of measurements or mathematical modeling. Accurate measurements, using sophisticated and costly equipment, is available and has indeed been used extensively to assess solar radiation. This paper concentrates on an alternative approach to assess global solar radiation (GSR) by using Artificial Neural Networks (ANNs) together with classical observed meteorological data. The model is applied to the region of Bangkok, Thailand, using meteorological data, along with solar radiation measurements, for the period 2001-2010 from the Thai Meteorological Department (TMD). More precisely, three combinations of observed monthly mean meteorological data, i.e. maximum, minimum, and mean temperatures; relative humidity; rainfall amount; and sunshine hours were used with 3, 5 and 6 parameters as the model input for the ANN training to predict the solar radiation over the territory. A feed-forward back-propagation ANNs were trained based on three algorithms, i.e. the Quasi-Newton, the conjugate gradient with Polak-Ribiere updates and the Bayesian regularization. The root mean square error (RMSE) and the mean bias error (MBE) between the observed and the predicted solar radiations in 2011-2012 were computed in order to investigate the performance of the ANNs. Results showed that, for monthly mean number of sunshine hours in the range of 3.58 to 9.55 hr/day, the monthly mean GSR above the atmosphere of Bangkok was in the range of 5.64 to 22.53 MJ/m(2)/day. The RMSE and the MBE were 0.0031 - 0.3632 and -0.0203 -0.003, respectively, thus indicating that the ANN modeling has sufficient performance to predict the monthly mean GSR over an area where classical meteorological data are measured. (C) 2014 Published by Elsevier Ltd.