15-Minute Market Transition
Background
The European electricity market is transitioning from hourly to 15-minute market time units (MTU15) as part of the EU Clean Energy Package. OMIE, the Iberian electricity market operator, began publishing 15-minute clearing prices in October 2025, moving toward full MTU15 market clearing.
Why 15 Minutes?
Hourly pricing was designed for an era of large, dispatchable power plants. The growth of variable renewable energy (wind and solar) creates significant intra-hour price variation:
- Solar ramps: A cloud passing over a solar farm can shift generation by hundreds of MW within minutes
- Wind gusts: Wind generation can fluctuate substantially within a single hour
- Demand peaks: Industrial loads and EV charging create sub-hourly demand spikes
- Battery dispatch: Storage systems optimize on sub-hourly price signals
15-minute pricing better reflects real-time supply-demand balances and sends more accurate economic signals.
Impact on Forecasting
The MTU15 transition changes the forecasting problem in several ways:
| Aspect | Hourly Era | MTU15 Era |
|---|---|---|
| Predictions per day | 24 | 96 |
| 7-day forecast size | 168 values | 672 values |
| Price volatility | Lower (hour-averaged) | Higher (quarter-hourly) |
| Historical data | 5+ years | Months (since Oct 2025) |
| Feature availability | All indicators hourly | Mixed (some hourly, some 15-min) |
EPF’s Three-Strategy Response
The system implements three approaches to handle the transition period:
1. Expanded (Immediate)
Replicates hourly forecasts to 15-minute slots. Available immediately with no retraining:
Hourly: [52.30 EUR/MWh for hour 14] → 14:00 = 52.30, 14:15 = 52.30, 14:30 = 52.30, 14:45 = 52.30Trade-off: No intra-hour variation, but leverages the full, proven hourly feature set.
2. Pure 15-Minute (Native)
Trains directly on 15-minute data with quarter-hourly lags and encoding:
Features: price_lag_4 (1h), price_lag_96 (24h), quarter_of_day_sin/cosTrade-off: Captures real sub-hourly dynamics but requires 6+ months of 15-minute history and has a reduced feature set.
3. Hybrid 15-Minute (Combined)
Merges hourly features (expanded) with native 15-minute data:
Hourly features (×4 expanded) + Native 15-min features → Combined modelTrade-off: Maximum feature coverage with intra-hour variation, but higher complexity.
Data Collection at 15-Minute Resolution
The same REE ESIOS indicators collected at hourly resolution are also fetched at 15-minute resolution using time_trunc=fifteen_minutes. These are stored in the ree_15min database table.
Key parameters:
time_trunc:fifteen_minutes(instead ofhour)time_agg:average(critical — same as hourly, prevents summation artifacts)- Interconnection data is excluded (only available at hourly resolution)
Transition Timeline
| Phase | Status | Description |
|---|---|---|
| Hourly only | Complete | Traditional 24-prediction forecasts |
| MTU15 publication | Active (Oct 2025+) | OMIE publishes 15-min prices alongside hourly |
| Expanded approach | Active | Hourly models expanded to 15-min output |
| Pure/Hybrid training | Active | Building 15-min training history |
| Full MTU15 | Future | Native 15-min market clearing replaces hourly |
Practical Considerations
Data history gap: The biggest challenge during transition is limited 15-minute history. Hourly models have years of training data; 15-minute models are starting fresh. The expanded approach bridges this gap while native history accumulates.
Feature availability: Not all data sources publish at 15-minute resolution. Interconnection flows, commodity prices, and some weather variables remain hourly. The hybrid approach handles this by expanding hourly features to 15-minute slots.
Storage and computation: 4× more data points per day increases storage requirements and training time. Origin step optimization (training on every 4th 15-minute step) mitigates the computational cost.