Analysing and Improving National Renewable Energy Forecasts: A Case Study in Belgium

The installed capacity for renewable energy capacity has grown significantly in many countries in recent years. In Belgium, solar and wind energy capacity rose from under 5 GW in 2016 to about 15 GW by 2024, with considerable further expansion foreseen in the coming years. Transmission system operators (TSOs) must manage generation data from renewable sources and provide accurate forecasts to balance electricity supply with demand. Poor data quality and large forecast errors can disrupt grid operations and lead to costly interventions to minimize imbalances and congestions etc. In this paper, we examine the energy generation data and forecasts publicly shared by Belgium's TSO, Elia as part of its transparency commitments. Our analysis finds some minor data quality issues, but a tripling in (absolute) forecast errors for solar and wind energy from 2017 to 2024 across various time scales. Our analysis also shows that while it plays a part, these errors cannot be attributed solely to capacity growth, highlighting a decline in forecasting accuracy. We also show that machine learning techniques can improve these existing forecasts with publicly available data, and consequently reduce balancing and system operation costs.