Tackling Grid Congestion with AI: ML6 Wins Elia’s Hack the Grid 2025

Executive Summary
Grid congestion is one of the biggest bottlenecks for Europe’s energy transition. At Elia’s Hack the Grid 2025, ML6 developed an AI-powered model that cut transformer hotspot prediction errors by 65% and secured first place in the hackathon. Combined with our strategy tool, Optimus, we demonstrated how AI can help Transmission System Operators (TSOs) safely extend existing infrastructure, increase transformer lifetime, and bridge the 8-year gap in hardware delivery.

The Challenge: Grid Congestion Meets an 8-Year Wait

Europe’s energy transition is accelerating fast. Electricity demand is expected to grow by up to 7% annually until 2030, while the grid infrastructure struggles to keep pace. Transmission System Operators (TSOs) like Elia face three major challenges:

• Electrification boom: EVs, heat pumps, and industry all require more power.
• Aging infrastructure: Key assets like transformers are nearing end-of-life.
• Supply chain crisis: Lead times for new transformers have exploded from months to up to 8 years.

This means grid operators must do more with what they already have — or risk bottlenecks that slow down the energy transition.

Transformers: The Unsung Heroes of the Grid

At the heart of this problem lie power transformers. They manage voltage and keep electricity flowing efficiently. But their true limit isn’t capacity — it’s heat.

The “hotspot” temperature inside the insulation determines transformer health. Every +6°C doubles insulation aging, slashing the lifetime dramatically. Traditional physical models often underestimate heat stress, forcing operators to play it safe and underutilize assets.

Hack the Grid 2025: A Real-World Energy Hackathon

To tackle this, Elia launched Hack the Grid 2025 in Brussels, bringing together innovators to answer one key question:

“How can we safely push substations beyond their current limits and delay costly upgrades?”

ML6’s Energy Transition team joined with a mission: build an AI-powered model that is more accurate than existing physical methods, and turn it into actionable strategies for operators.

Building an AI Model to Predict Transformer Hotspots

Traditional IEC-based models provide conservative estimates but miss the unique “thermal fingerprint” of each transformer.

ML6 took a different approach:

• Data inputs: 2 years of operational data from 17 transformers (load, hotspot temp, outside air temp).
• Feature engineering: Lagged and rolling averages to capture thermal inertia (“memory effect”).
• Models tested: Gradient-boosted trees (XGBoost, LightGBM).

The Results:

• 65% reduction in Mean Absolute Error (MAE) compared to Elia’s baseline model.
• A single global model with lagged features outperformed both transformer-specific and physical models.
• ML6 secured first place in Part 1 of the hackathon.

This new level of accuracy meant operators could safely push hardware closer to real limits — without compromising lifespan.

From Insight to Strategy — Enter Optimus

Winning the model was just the start. The bigger question: how do you act on it?

ML6 built Optimus, a strategic decision-support tool that turns predictions into action.

What Optimus does:

• Prioritization: Flags transformers by risk level (low/medium/high).
• Scenario simulation: Test strategies like raising temperature thresholds or adding battery support.
• Trade-off analysis: Balance disconnections avoided, costs, and asset lifespan.

Example: At one substation, raising the hotspot threshold from 98°C to 110°C cut disconnections by 64% — but shortened lifespan by 1.1 years. Adding 1.5 MW of battery support restored the lifespan while saving 80% of disconnections.

Optimus empowers operators with data-driven levers instead of one-size-fits-all rules.

Why This Matters: A Digital Layer for the Grid

Grid congestion can’t be solved with hardware alone. The future is a digital-first strategy:

• AI models that reveal the real limits of infrastructure.
• Decision tools that help operators act confidently.
• Decentralized flexibility, e.g., aggregating home batteries to support peak demand.

This approach transforms scarcity into resilience. Instead of waiting 8 years for new transformers, operators gain 8 years of extra capacity from existing assets.

Key Takeaways

• Grid congestion is a critical bottleneck for Europe’s energy transition.
• Transformer lifespan depends on hotspot temperatures — accurate prediction is key.
• ML6’s AI model cut prediction error by 65% vs. traditional methods.
• The strategy tool Optimus empowers TSOs to balance asset lifespan, cost, and load.
• A digital layer of intelligence is essential to bridge the infrastructure gap.

From Hackathon to Blueprint for the Future

Elia’s Hack the Grid 2025 showed that the energy transition isn’t just about concrete and steel — it’s about intelligence.

By combining an AI model with a strategic decision tool, ML6 demonstrated how we can safely unlock hidden capacity in today’s grid and buy precious time for the infrastructure of tomorrow.

For ML6, this was more than a hackathon win. It’s a blueprint for the future: using AI not just to analyze data, but to enable smarter, faster, and more sustainable decisions that power a cleaner energy system.