Distributed Electricity Model
Model Summary
- Model version
- demand_production_drift
- Target
- Annual distributed electricity (GWh)
- Geography
- 34 harmonized modeling regions
- Selection basis
- Rolling and recursive out-of-sample backtesting
- Input
- Historical distributed-electricity series by harmonized region
- Regional intervals
- Not published
Why This Model Is Used
Drift extrapolates the average annual change implied by the first and latest observations in each regional historical series. It was retained because it provided stable out-of-sample performance under the current short annual dataset and validation protocol, while more complex challenger models did not demonstrate a sufficiently consistent improvement.
Local Generation Model
Model Summary
- Model version
- supply_production_drift_v1
- Target
- Annual local electricity generation (GWh)
- Geography
- 34 harmonized modeling regions
- Selection basis
- Nested robustness validation
- Input
- Historical local-generation series by harmonized region
- Regional intervals
- Not published
Why This Model Is Used
Drift V1 extrapolates the average annual change implied by the first and latest observations in each regional local-generation series. It was retained after challenger models did not demonstrate sufficiently robust improvement under nested validation. The V1 designation identifies the validated production implementation used by the dashboard.
Validation Design
Out-of-sample evaluation used for production-model assessment
Rolling One-Step
- Target years
- 2019–2024
- Forecast horizon
- 1 year
- Purpose
- Near-term stability
Recursive Multi-Step
- Origins
- 2018–2021
- Maximum horizon
- 3 years
- Purpose
- Forecast-path robustness
Supported Performance Measures
Measures supported by the current forecasting methodology
Metric values will be displayed after the production-metric outputs are integrated into the frontend data contract. Challenger win rate is used during model comparison against the Drift benchmark and is not interpreted as a standalone accuracy measure for the active Drift model.
Validation and Reporting Note
Production models are evaluated using rolling one-step and recursive multi-step backtesting. Supported evaluation measures include sMAPE, MASE, RMSE, aggregate bias, and challenger comparisons against the Drift benchmark. Generic R², unsupported feature-importance displays, and unvalidated regional uncertainty intervals are intentionally excluded from the production dashboard.