GIS-based solar irradiation forecasting using support vector regression and investigations of technical constraints for PV deployment in Bangladesh

Bangladesh, one the fast developing countries in the world has the huge potential to harness the solar energy to mitigate its ever increasing power demand. However, identifying the solar irradiation on different locations for practical use is a challenging task because of its unpredictable nature. Recent advancements in machine learning techniques are found to be a very effective tool to automatically configure a nearly accurate forecast model. This prediction is very useful in properly designing the solar systems which include PV panels, solar collectors etc which are directly influenced by the fluctuations in radiations. This paper demonstrates a support vector regression model to predict the monthly and daily solar potential across the six divisions of Bangladesh. The SVM (Support Vector Machine) model discussed here is comprised of six weather variables considered as inputs which have been derived from GIS (Geographic Information Systems) based on solar mapping tools. The result from this work indicates that the prediction model closely follows the regression line and a good data fitting to the statistical model with less MSE (mean squared error) proves the validity of this model.