Incorporating Weather Uncertainty in Energy Forecasts: Using Ensembles for Intraday and Day‑Ahead Congestion Management

The electricity grid of the Netherlands is nearing its limits, making short-term load forecasts central to active congestion management. At the same time, the rapid rise of variable renewables has increased the grid’s sensitivity to the weather. Weather forecasts inherently carry some degree of uncertainty, which can be incorporated in energy forecasts in multiple ways. We explore how ensemble weather forecasts can improve probabilistic day-ahead and intraday energy predictions by coupling data-driven load forecasting models with physical Numerical Weather Prediction (NWP) ensembles.

Using the open-source OpenSTEF framework, we train asset-specific forecasting models that predict grid load from calendar, weather, historical load, and market price features. Our approach replaces deterministic meteorological inputs with ensemble quantiles during inference. Tests on real grid assets show improved accuracy, calibration, peak detection and forecast stability. We also identify two key operational challenges: managing dependencies between weather variables and combining different types of specialized weather forecasts with ensembles.

Propagating weather uncertainty into energy forecasts improves the efficiency of grid operation during the energy transition. We invite discussion on hybrid modelling strategies, calibration techniques and validation, and practical aspects such as optimal resolution. We look forward to exchanging ideas and experiences that advance robust and open probabilistic forecasting practices.