Forecasting Meteorological Solar Irradiation Using Machine Learning and N-BEATS Architecture

Md. Tawhid Anwar, Md. Farhadul Islam, Md. Golam Rabiul Alam

Abstract: For developing nations like Bangladesh, the Calamity of Energy, one of the most important warnings experienced in the modern world, is a significant problem. Solar energy is a great solution for the future. When solar panels are used to create electricity, no greenhouse gas emissions are produced. Solar energy is essential to the shift to the production of clean energy because the sun generates more electricity than people could possibly need. In this study, for better planning and decision making for solar energy consumption, we propose a forecasting model based on Bangladeshi data. We collected the data from NSRDB(National Solar Radiation Database). Using meteorological data of 4 Regions of Bangladesh - Chittagong(CTG), Khulna(KHU), Sylhet(SYL), Rajshahi(Raj). Firstly, we find the most important feature for radiation prediction and we conduct a regression analysis based on the selected feature, then for time series analysis, we use state-of-the-art N-BEATS architecture, which gives us impressive results with very low computational cost and time. N-BEATS outperformed other popular models like LSTM and SARIMA.