Expanded Approach

Overview

The expanded approach is the simplest path to 15-minute resolution forecasts. It trains models on hourly data using the standard hourly pipeline, then replicates each hourly prediction to four quarter-hourly slots. This produces 96 forecasts per day (4 × 24) without requiring any native 15-minute input data.

How It Works

Hourly model → 24 predictions/day → Expand × 4 → 96 predictions/day

Hour 14 prediction: 52.30 EUR/MWh
  → 14:00 = 52.30
  → 14:15 = 52.30
  → 14:30 = 52.30
  → 14:45 = 52.30

Train: Standard hourly models using the full hourly feature set
Predict: Generate 24 hourly point forecasts per day
Expand: Replicate each hourly value to its four constituent 15-minute slots
Intervals: Confidence intervals are similarly replicated (all four quarters share the parent hour’s bounds)

Advantages

Zero retraining cost — uses existing hourly models without modification
Full feature coverage — leverages all hourly indicators (interconnections, etc.)
Immediate deployment — works today, even before 15-minute data is widely available
Proven accuracy — hourly models have the longest track record and most stable performance

Limitations

Flat intra-hour profile — all four quarters within an hour get the same price, missing genuine intra-hour variation
No sub-hourly learning — cannot capture 15-minute-specific patterns (ramp events, solar intermittency)
Oversmoothed — real 15-minute prices have higher variance than hour-averaged prices

When to Use

The expanded approach is the recommended starting point for 15-minute forecasting. It’s appropriate when:

Native 15-minute historical data is limited (less than 6 months)
The priority is reliability over sub-hourly granularity
Downstream consumers can tolerate flat intra-hour profiles
Comparing 15-minute results against hourly baselines

Output Format

The expanded predictions use target_minute values of 0, 15, 30, 45:

target_hour	target_minute	predicted_price
14	0	52.30
14	15	52.30
14	30	52.30
14	45	52.30
15	0	48.10
…	…	…

Comparison with Other Approaches

Aspect	Expanded	Pure15	Hybrid15
Training data	Hourly	15-min only	Both
Intra-hour variation	None	Full	Full
Feature coverage	Full	Limited	Full
Data requirement	Minimal	6+ months 15-min	6+ months 15-min
Complexity	Low	Medium	High