Forecasting Solar Irradiance Using Machine Learning

Abstract

Renewable energy is becoming a very popular source for power generation nowadays. In the context of Bangladesh, solar energy has become the most prospective renewable resource for which solar irradiance is a very important parameter. Being able to forecast the solar irradiance accurately can facilitate efficient design of any solar power plant. In this study we have used an Artificial Neural Network (ANN) which is essentially a Machine Learning (ML) approach. As it is time series-based forecasting, we have taken past 15 years' (20002015) daily data from the renewable energy community of NASA database. We have chosen a coastal area for this study case like Saintmartin near Teknaf which has a boundless role in Bangladesh. Here, a feed forward back propagation neural network has been used. Eight important parameters have been considered as independent input variables to forecast daily solar irradiance and the parameters are - air temperature, wind speed, precipitation, humidity, surface pressure, insolation clearness index, and earth skin temperature. The proposed model has provided prediction results with good accuracy and minimal error.