Enhancing TSO’s grid security with probabilistic forecasting

Summary

In this case study, you will discover details regarding:

1. Reasons why Tennet is facing more challenges in predicting day-ahead congestion.
2. The innovative probabilistic approach developed by N-SIDE, which utilizes artificial intelligence and dynamic reliability margins to enhance the accuracy of risk-conscious grid security analysis. 
3. The advantages of implementing such tools including minimizing false alarms and avoiding “missed congestion” events, along with supporting decision-making through an operator dashboard. 
4. Finally, the conclusions drawn from the collaboration with TenneT throughout this project.

Project Introduction

The main objective of the project was to develop and demonstrate a novel probabilistic congestion forecasting approach. This method enhances grid security analysis for TSOs by taking significant sources of uncertainty in grid operations into account.

This collaboration between N-SIDE and TenneT combines business and analytical expertise to enable more reliable, risk-aware decision-making in TenneT’s control room. This was done by leveraging advanced forecasting tools for operational use.

The Challenge

TenneT faces increasing difficulties in forecasting grid congestion due to the higher incorporation of renewable energy sources (RES), larger short-term market fluctuations, and the growing complexity of the network. At TenneT, grid security analysis currently depends on deterministic approaches, offering operators a single forecasted thermal loading value for each monitored asset, commonly referred to as a point forecast.

However, traditional point forecasting methods are inadequate for modern power systems because they cannot capture the stochastic and uncertain characteristics of these highly dynamic environments. As sources of uncertainty continue to develop, reliance on these point-forecast methods can lead to higher congestion risks, costly actions, and challenges in maintaining grid security. 

This sets the path for innovative tools to assist operators, such as TenneT, in addressing these challenges, including enhanced forecasting methods that quantify these uncertainties.

The Approach

To address the growing uncertainty in Day-Ahead (DA) congestion forecasting, a new approach was developed specifically for TenneT. This method is called Probabilistic Congestion Forecasting (PCF). The methodology begins with forecasting nodal power uncertainties linked to the most impactful uncertainty drivers: cross-border flows, renewable generation, and electricity demand.

Then, grid nodal scenarios are generated using a sound statistical method to account for nodal correlations, accurately reflecting plausible real-world grid conditions. For each probabilistic scenario and Critical Network Element under Contingency (CNEC), an accelerated AC load-flow analysis is performed to determine congestion. 

By aggregating the resulting forecast loading over all scenarios, a probabilistic congestion profile can be reconstructed, providing insights into the probability and severity of congestion. A dynamic user interface dashboard presents these profiles and percentiles, enabling operators to evaluate risks and make informed decisions.

Benefits of the collaboration

A comparison between a point forecast model calibrated* with a fixed margin and a PCF model calibrated with a dynamic margin over a six-month evaluation period (considering all voltages of TenneT’s grid) demonstrates that the probabilistic approach significantly enhances accuracy in predicting congestion status compared to the point forecast method:

• Improved accuracy: Reduces missed congestion cases by 63% at a fixed false alarm rate (precision of 0.6) compared to point forecasting, enhancing congestion prediction reliability. 
• Reduced false alarms: Reduces false alarms by 47% when calibrated at a fixed security level (10% missed congestion cases), thereby minimizing unnecessary costly redispatch actions. 
• Dynamic Reliability Margin: A margin that adapts based on predicted uncertainty levels associated with weather or market conditions, enabling better anticipation of congestion. 
• Enhanced decision support: Provides a dynamic dashboard detailing congestion profiles and percentiles, enabling risk monitoring and more informed decision making. 

*Calibration of the forecasts has been performed using a reliability margin, i.e. a quantity that is added to the point forecast (PF) or a percentile (e.g., P50) to account for the uncertainties in the prediction. A static margin is a fixed quantity that is added on top of all predictions, whereas a dynamic margin varies with the width of the predicted probability distribution, capturing the predicted level of uncertainty specific to a CNE and a time unit.

[Note: An important caveat is that, during this evaluation, the IDCF grid models used for TenneT’s point forecast were based on intraday files with their prognoses imported from the day-ahead process. As a result, these files did not include updated intra-day forecast information, potentially affecting the accuracy of the point forecast and partially explaining the observed performance gap. This point of attention should be considered when interpreting the comparative results.]

Conclusion

Concluded in September 2025, this project successfully demonstrated the value of moving from a deterministic to a probabilistic congestion forecasting approach.

By quantifying nodal uncertainties and propagating them into the loading of Critical Network Elements under Contingency (CNEC’s), the PCF model allows to support more informed, risk-based decisions (less risks). 

In simple terms, operators can now reduce the risk of missing congestion, better leveraging the existing infrastructure.

Possible future steps include adapting the use-case for Week-Ahead forecasting, incorporating more uncertainty sources, and transitioning to common grid models (CGMs).