About AgilePredict

This model attempts to forecasts Octopus Agile electricity prices up to 14 days in advance using a Machine Learning model trained on data from the Balancing Mechanism Reporting System (BRMS), National Grid Electricity Supply Operator (NG ESO) and weather data from open-meteo.com. The table below lists the features used in the model.

How the Forecast Works

AgilePredict uses an XGBoost Gradient Boosting Regressor to forecast electricity prices based on a blend of weather and energy market data. Here's an overview of how it works:

• Data Collection: Forecasts and actual prices are sourced from NG ESO, BMRS, Open-Meteo, and historical Agile data.
• Feature Engineering: Time-based and derived features like peak hours, days since forecast, and weekend indicators are calculated.
• Training Focus: The model trains on the data closest to 16:15 daily (aligned with auction timing) and targets the PM auction period (22:00–22:00 next day).
• Modeling: The XGBoost model is trained using log-weighted errors to better capture high-volatility pricing. It uses cross-validation to validate accuracy.
• Prediction: New forecasts are generated, and confidence intervals are estimated using Kernel Density Estimation (KDE).
• Blending: Model outputs are blended with actual prices and extended into future time slots. Forecasts include a central prediction, and low/high bounds.
• Publishing: Results are stored and published for multiple UK regions with Agile-compatible pricing.

The forecasts are updated automatically four times per day: at 06:15, 10:15, 16:15, and 22:15. The accuracy tends to be stronger for general trends rather than exact half-hour slots, especially further into the future.

AgilePredict is a work in progress. Please use the forecasts with care and always check with official pricing when it matters.