Forecasting hourly global solar radiation with an Artificial Neural Network

Abstract

Many studies have been carried out to estimate the global solar radiation,
since the determination by manual or automatic equipment involves high
costs. In general, solar radiation prediction models are intended to predict
daily solar radiation and are developed based on temperature data. Studies
on hourly solar radiation are scarce. The determination of hourly solar
radiation can increase accuracy in some areas of research, such as precision
agriculture. In addition, the hourly variations of the solar radiation intrinsically
compute local attributes that interfere in the process of solar radiation, such
as topography, relief and atmospheric composition. The goal of this study
was to develop an Artificial Neural Network to predict hourly solar radiation
from temperature data and incident solar radiation at the top of the
atmosphere at noon. For this, monitoring data from the automatic station of
the National Institute of Meteorology, located in Campos dos Goytacazes, Rio
de Janeiro State, Brazil, at coordinates 41.35° W Longitude and 21.71° S
Latitude, were used. The adopted model was an Artificial Neural Network with
multiple layers (Multilayer Perceptron). The performance of the model was
evaluated by Mean Squared Error, Mean Absolute Error and R².