Search

In this paper, a computational model for the prediction of direct solar radiation based on neural network analysis of atmospheric clearness is developed. It considers that the major portion of direct solar radiation reaching the earth's surface is governed by Sun-Earth geometry and atmospheric transmittance factors which are exactly calculable by clear day model. Additional variations are due to climate and weather phenomena characterized by relative humidity, mean duration of sunshine per hour, and rainfall, etc., in the atmosphere. These variations are taken into account with the help of a composite parameter referred to as atmospheric clearness index (CI) which is determined using artificial neural network analysis. The contour maps of CI as a function of latitude, time of the day, and month of the year are then prepared using the meteorological data of eleven stations. Model simulation and test results of the trained network for two typical locations (not used in training the network) are presented and compared with measured values. The deviations are well within acceptable error limits [percentage root mean square error values for Ahmedabad and Nagpur are 3.34 and 2.06, respectively]. The methodology, for predicting the direct solar radiation at an arbitrary location using the well known parameters such as altitude, latitude, time of the day, day of the year, and CI values derived from the contour maps, is discussed. The outcome of the present ANN model for four arbitrary locations is compared with NASA SSE data sets. [ABSTRACT FROM AUTHOR]