Binning Based Data Driven Machine Learning Models for Solar Radiation Forecasting in India

Energy is the primary driving force in improvement of the human life cycle. All the activities for the betterment of human life are dependent on some form of energy. Conventional energy sources rely on fossil fuels which have limited reserves and we are bound to exhaust them soon. On the other hand, non-conventional/renewable energy sources are produced on a regular basis and are clean without any polluting emissions. These sources include solar, wind, hydraulic, biomass/bio gas, geothermal, tidal, etc. Solar energy is one of the primary sources in countries like India, but it does have drawbacks like high initial cost, dependency on weather, expensive storage, space requirement, etc. It is therefore imperative to create accurate solar radiation forecasting models to identify and address these issues. Forecasting models are created based on daily or hourly data and are location specific. In this work, binning based machine learning models are proposed for accurately forecasting hourly solar radiation. These models are data driven clustering based models. The clusters are identified based on geographic locations. The proposed approach also helps reduce the number of required models without compromising the high accuracy. In this work, global and diffuse solar radiation data, gathered from five geographically distinct stations from India, is analyzed. Validation of these models demonstrate increased performance. The number models required are also significantly smaller compared to the daily or hourly models.