Dutch electricity load forecasting app with backtested, calibrated predictions
gridcast-nl predicts the Netherlands' hourly electricity demand up to a week ahead using LightGBM models trained on public data. It provides a Streamlit interface that visualizes forecasts, baseline comparisons, segment‑wise scores and calibrated 80 % prediction intervals. The rolling‑origin backtest and quarterly refits ensure realistic performance evaluation, making it useful for grid operators and analysts. Compared to generic demos, it delivers a production‑ready pipeline with thorough validation and actionable uncertainty metrics.
View on GitHub →iArslan95/gridcast-nl