PROJECT 02 / MACHINE LEARNING & TIME SERIES

GridCast.

Forecast hourly electricity demand using real U.S. grid history. Inspect the error before you trust the forecast.

Data archive ↗

●Loading historical ERCOT electricity demand…

Observed series

—

Continuous hourly records

Model test MAE

—

Megawatts, held-out hours

Seasonal baseline MAE

—

Same hour yesterday

24-hour forecast peak

—

Megawatts

Observed demand & next 24 hours

Forecast is out-of-sample beyond the final historical observation.

—

Model configuration

Ridge regression with hourly lag features

Regularization strength1.0

Evaluation window

Demand scenario0%

The scenario adjustment is applied after modeling and is not part of the learned forecast.

Model versus baseline

Chronological holdout, no random train/test shuffle.

Load your own time series

Use historical U.S. load CSV files. All calculations run locally in the browser.

CSV import
date_time + cleaned demand (MW), or timestamp + load

At least 120 contiguous hourly values recommended. UTC timestamps avoid daylight-saving ambiguities. The included sample has 336 real historical observations.

↓ Download example input CSV

Forecast output

Predictions are model estimates. Error bands are approximate residual-based ranges, not calibrated probability intervals.

Hour (UTC)	Predicted MW	Scenario MW	Approx. low	Approx. high

Data attribution: Ruggles et al., Scientific Data (2020), cleaned EIA Form-930 hourly demand, CC BY 4.0. Historical example only. Methods publication ↗