What is it about?
Wind and solar power depend strongly on weather, making their future output difficult to predict with certainty. For power system operators, knowing how reliable a forecast is can be just as important as knowing the predicted amount of electricity. This study develops a new method called Spatio-Temporal Adaptive Conformal Forecasting (STACF). It learns how renewable generation changes across both locations and time, while continuously adjusting the uncertainty around its forecasts as conditions change. Unlike many conventional approaches, STACF does not require renewable power uncertainty to follow a predefined statistical distribution. Instead, it uses recent forecasting errors and relationships among geographically distributed sites to produce prediction intervals that are both reliable and informative. Tests using multi-regional data covering 1,395 wind farms in the United States demonstrate that the method can provide well-calibrated uncertainty estimates while avoiding unnecessarily wide prediction intervals.
Featured Image
Photo by Nicholas Doherty on Unsplash
Why is it important?
As renewable energy supplies a growing share of electricity, inaccurate forecasts can increase operating risks, reserve requirements, and the difficulty of balancing power systems. Traditional uncertainty forecasting methods often struggle when weather patterns change or when renewable plants in different locations influence one another. This study addresses both challenges by jointly learning spatial and temporal relationships and continuously recalibrating forecast uncertainty. The resulting prediction intervals tell grid operators not only what renewable generation may be, but also how much confidence they can place in that forecast. Across multiple regional datasets, the proposed method reduced coverage deviation and produced more informative uncertainty intervals than leading alternatives. More reliable uncertainty information can ultimately support safer and more efficient decisions for renewable-rich power systems.
Perspectives
This work represents a shift from simply predicting renewable power toward understanding and managing the reliability of those predictions. By combining dynamic spatial learning with adaptive uncertainty calibration, the proposed framework can respond to changing relationships among renewable sites and evolving weather conditions. Importantly, its conformal approach provides uncertainty estimates without requiring restrictive assumptions about the underlying probability distribution. The framework also shows potential for operational applications. During a rapid wind-power ramping event, STACF maintained reliable coverage while producing substantially narrower prediction intervals than competing methods. Its lightweight online calibration also enables fast inference suitable for real-time decision-making. Future research could connect such uncertainty-aware forecasting directly with power system scheduling, reserve allocation, energy storage, and other operational decisions, allowing renewable uncertainty to be managed more intelligently throughout the decision-making process.
Chair, IEEE PES EICC Task Force on AI-Enabled Resilience of CPES|Clarivate HCR|AE: IEEE TSG/TSTE/TII Yang Li
Northeast Electric Power University
Read the Original
This page is a summary of: Spatio-Temporal Uncertainty Quantification for Renewable Power, IEEE Transactions on Sustainable Energy, January 2026, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/tste.2026.3727059.
You can read the full text:
Contributors
The following have contributed to this page
