An end-to-end AutoML pipeline that trains an AutoML model on historical ONS open data and produces monthly forecasts of electric energy consumption for Brazil's Southeast (SE) subsystem: the country's largest, covering São Paulo, Rio de Janeiro, Minas Gerais and Espírito Santo. Updated automatically on the 1st of each month via GitHub Actions.
Tip: Click a legend item to show/hide it. Double-click to isolate a single series. Use the range buttons or the slider below the chart to zoom into a period. The orange region is the forecast; to the left of the line, actual data overlaps the forecast for comparison.
Daily energy load, reservoir storage (EAR) and generation mix from the ONS open data portal.
H2O AutoML — best model: GBM with 50 engineered features (temporal, lag, rolling, ratios). Test MAPE: 1.83%. Current month MAPE (26 days): 3.7%. Tracked and versioned in MLflow, with a rolling-window accuracy check that automatically retrains and redeploys the model on sustained drift.
GitHub Actions runs on the 1st of every month: checks for new ONS data, incrementally updates the consolidated dataset, and re-runs inference. GitHub: Energy_Consumption_AutoML.