Hourly load (kW): teal = lags, blue band = rolling window, amber = forecast | hour as unit circle
Test RMSE (kW) by feature set, lower is better

Lag Features, Rolling Windows and Cyclical Hour Encoding: Forecast Error and Unit-Circle Map

Turn a noisy hourly electricity-load series into the columns a model actually learns from: lag features, rolling-window mean and spread, and a sine/cosine clock. A ridge regression forecasts the next hour and is scored on nine unseen days against persistence and the noise floor, with built-in numerical checks. Desk, window posts, pegs and dial modelled in Blender.

Made with 🟠 Blender