Geostationary Satellite Image-based Deep Learning Techniques For Solar Radiation Forecasting

With growing global awareness regarding the environmental consequences associated with non-renewable fossil fuels, the demand for clean and renewable energy sources has surged. Solar photovoltaic (PV) technology has emerged as a promising alternative, but accurate solar irradiance forecasting remains a challenge due to the variable nature of solar radiation. This thesis explores the potential of deep learning techniques, specifically employing a 2D-Convolutional Long-Short-Term Memory (ConvLSTM2D) architecture, to develop real-time and future solar irradiance forecasting models. The study focuses on Global Horizontal Irradiance (GHI) forecasting, utilizing multi-resolution satellite images from the Meteosat family of geostationary satellites managed by EUMETSAT as exogenous inputs. The impact of high-resolution visible (HRV) satellite imagery on the model's performance is investigated through a comparative study. The proposed methodology is evaluated on a dataset from Carpentras, France, and its performance is compared with benchmark models, showcasing improved accuracy and either outperforming or matching other state-of-the-art solutions. Additionally, the forecasting horizon is extended from the standard 6 hours to 10 hours, revealing promising performance and contributing to the advancement of intra-day GHI prediction. Furthermore, a generalization study is conducted, evaluating the model's ability to make predictions in unseen locations across Europe using satellite data from 12 different locations. The results demonstrate the model's capability to generalize without fine-tuning, emphasizing the importance of considering spatio-temporal information for accurate GHI forecasts in diverse geographical and environmental contexts. The findings of this research, utilizing multi-resolution satellite images and HRV channel comparison, provide valuable insights for efficient energy management and planning in solar energy systems, facilitating the wider adoption of clean and renewable energy sources.