GraphCast vs. IFS: a comparative analysis of weather forecasting models for optimized renewable energy trading

Renewable electricity generation is set to grow significantly as the transition to sustainable energy accelerates in response to climate change. While the environmental benefits of this shift are clear, integrating intermittent renewable sources into the power grid presents challenges. The energy output of solar panels and wind turbines, the most widespread renewable technologies, depends entirely on weather conditions, which are highly variable and difficult to predict. This uncertainty complicates renewable energy forecasting, making it harder for suppliers to optimize market bids. As a result, grid stability becomes difficult to maintain, and suppliers risk financial penalties for failing to meet commitments. Currently, Numerical Weather Prediction (NWP) models, particularly the Integrated Forecasting System (IFS) operated by the European Centre for Medium-Range Weather Forecasts (ECMWF), set the standard for weather simulation. However, Machine Learning Weather Prediction (MLWP) has recently emerged as a promising alternative, offering the potential for improved prediction accuracy and efficiency. This thesis evaluates GraphCast, a machine learning-based forecasting tool developed by Google DeepMind, against ECMWF’s IFS. The study focuses on predicting wind speed and temperature—key factors for wind and solar energy generation—using a publicly available pre-trained version of GraphCast. The models’ performance is assessed by comparing six months of forecasts with actual measurements from six SYNOP stations across Italy. Results show that GraphCast achieves accuracy comparable to IFS and demonstrates greater stability over time, with less degradation in predictive performance for extended forecasts. As weather prediction is crucial for optimizing renewable energy trading, advancements in ML-based forecasting systems like GraphCast will be key to a cleaner, more resilient energy future aligned with global climate goals.