Grid Stress & Heat wave Load Forecasting with Temporal Fusion Transformers
Simultaneous severe heatwaves triggered by super El Niño events cause record peak electricity demand for air conditioning, straining transmission grids and increasing transformer failure rates. Utilities and grid risk insurers need sub-seasonal forecasting models that capture complex interactions between ambient temperature, humidity, industrial calendar schedules, and clean energy dispatch. In this challenge, you will implement a Temporal Fusion Transformer (TFT) using PyTorch Forecasting. Your model will predict multi-horizon electrical load demand and transformer overheating risk across regional transmission sub-stations. Your solution will process multi-variate continuous weather signals and static infrastructure attributes, emitting multi-step quantile forecasts with calibrated uncertainty bounds.
What you are building
The core problem, expected build, and operating context for this challenge.
Implement a Temporal Fusion Transformer in PyTorch Forecasting to predict regional electrical grid load stress during extreme El Niño heat waves.
How work is evaluated
Evaluates multi-horizon load forecast quantile loss against ground truth grid telemetry.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
quantile_monotonicity
Ensure p90 prediction is greater than or equal to p50
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
quantile_loss_p50
Normalized quantile loss at 50th percentile • target: 0.035 • range: 0-0.2
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Build a Temporal Fusion Transformer (TFT) deep learning model for time-series forecasting.
Process multi-variate weather and calendar inputs using PyTorch Forecasting DataLoaders.
Generate multi-horizon quantile predictions (p10, p50, p90) for power demand.
Analyze variable importance metrics to identify key drivers of peak load stress.
Reference links and supporting material
Synthetic multi-variate time series dataset containing hourly electrical grid substation demand, temperature readings, humidity, and calendar flags.
Challenge setup
Evaluates multi-horizon load forecast quantile loss against ground truth grid telemetry.
Challenge input
JSON with substation_id, historical_load_mw, forecasted_temp_c, forecasted_humidity
PyTorch Forecasting
Provides official Temporal Fusion Transformer implementation.
Evaluated output
JSON with peak_load_p50_mw, peak_load_p90_mw, grid_stress_warning (bool)
- Ensure p90 prediction is greater than or equal to p50
- Normalized quantile loss at 50th percentile • target: 0.035 • range: 0-0.2
- Benchmark: GridLoadBench
- Quantile Loss P50 target: 0.035
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Transformers · Required
- Claude 3.5 Haiku · Optional
- Fireworks AI · Optional
- Fireworks AI · Optional
Evaluation contract
- quantile_monotonicity · Weight 1
- quantile_loss_p50 · Weight 1
Metrics
- quantile_loss_p50 · Target 0.035
Benchmarks
- GridLoadBench
- energy
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with substation_id, historical_load_mw, forecasted_temp_c, forecasted_humidity
Output: JSON with peak_load_p50_mw, peak_load_p90_mw, grid_stress_warning (bool)
Execution harness: Python
Evaluation tools
- PyTorch Forecasting · Deep Learning Time Series Framework
Recipe freeze marker
No recipe freeze time is recorded.