The increasing penetration of variable renewable energy sources (vRES) and electrification-driven demand is challenging the real-time power system balance, particularly under conditions of high variability in renewable generation and/or demand. Moreover, as illustrated by the 28 April 2025 Iberian blackout, it is crucial to equip power system operators with additional early-warning tools to support their decisions, enabling them to anticipate critical periods and enhance situational awareness sufficiently ahead of real-time operation.

This work presents an early warning forecasting framework to anticipate critical operating conditions in a power system – applied to the Portuguese case - by defining five levels of system awareness, ranging from normal operation to critical states, characterised by the magnitude and temporal patterns of balancing reserve activation. First, relevant features influencing the activation of balancing reserves are identified using a regression tree-based model. To support this step, historical vRES and demand forecasts obtained from the ENTSO-E Transparency Platform, together with operational data from the Portuguese system, are used. Second, inspired by methodologies used in atmospheric sciences for extreme event detection, a holistic forecasting framework driven by numerical weather prediction (NWP) outputs and operational data is developed to predict system awareness levels using Long Short-Term Memory (LSTM) neural networks. The framework is evaluated across different forecast time horizons to assess the trade-off between lead time and accuracy.

Preliminary results for the Portuguese power system indicate the proposed methodology can effectively complement existing deterministic operational tools, enabling a more comprehensive assessment of system risk and supporting the timely commitment of additional reserves to mitigate operational constraints, thereby contributing to an adaptive reserve allocation. Results also show that the framework helps mitigate amplitude and phase errors commonly associated with single-point forecasts by incorporating a more holistic representation of system drivers.

As in other domains such as meteorology, the results of this work may therefore extend beyond system operators to a broader set of stakeholders, including market participants and consumers, enabling more effective management of critical situations, which is particularly relevant in future ~100% vRES power systems.

This research was funded by CETPartnership, the Clean Energy Transition Partnership under the 2022 joint call for research proposals, co-funded by the European Commission (GA N°101069750) and with the funding organizations detailed on https://cetpartnership.eu/funding-agencies-and-call-modules, in specific, the FCT - Fundação para a Ciência e a Tecnologia (CETP/0001/2022).