Timestral

Use case

Electricity load forecasting with a visible uncertainty band

Forecast hourly load from a CSV, read the daily cycle and the 80% band, and use it for scheduling, procurement and demand-response planning.

Run the samplePut this in production

Load forecasting is a scheduling input. The operator needs to know not only the expected load but how high it could reasonably go. A model that returns only a point estimate pushes the uncertainty into an informal reserve, where it is harder to audit.

The sample below is two days of hourly load in MW. For a real scheduling decision, use weeks of history so the model can learn the weekday/weekend difference and the seasonal shape. Keep the timestamp timezone consistent; daylight-saving transitions are a common source of phantom spikes.

The sample series

The studio below is preloaded with this hourly load (mw) series and a 24-step horizon. It is an illustrative example, not a real company’s data. Replace it with your own CSV to run the same workflow.

Show the CSV
time,load_mw
2026-06-01 00:00,61.2
2026-06-01 01:00,58.9
2026-06-01 02:00,57.1
2026-06-01 03:00,56.4
2026-06-01 04:00,57.8
2026-06-01 05:00,61.5
2026-06-01 06:00,68.2
2026-06-01 07:00,75.4
2026-06-01 08:00,80.1
2026-06-01 09:00,82.6
2026-06-01 10:00,83.4
2026-06-01 11:00,82.8
2026-06-01 12:00,81.2
2026-06-01 13:00,80.6
2026-06-01 14:00,81.8
2026-06-01 15:00,84.2
2026-06-01 16:00,86.1
2026-06-01 17:00,85.4
2026-06-01 18:00,82.0
2026-06-01 19:00,78.6
2026-06-01 20:00,75.2
2026-06-01 21:00,72.4
2026-06-01 22:00,68.9
2026-06-01 23:00,64.5
2026-06-02 00:00,60.8
2026-06-02 01:00,58.2
2026-06-02 02:00,56.9
2026-06-02 03:00,56.1
2026-06-02 04:00,57.2
2026-06-02 05:00,60.9
2026-06-02 06:00,67.8
2026-06-02 07:00,74.9
2026-06-02 08:00,79.8
2026-06-02 09:00,82.1
2026-06-02 10:00,83.0
2026-06-02 11:00,82.4
2026-06-02 12:00,80.9
2026-06-02 13:00,80.1
2026-06-02 14:00,81.4
2026-06-02 15:00,83.8
2026-06-02 16:00,85.7
2026-06-02 17:00,85.0
2026-06-02 18:00,81.6
2026-06-02 19:00,78.1
2026-06-02 20:00,74.8
2026-06-02 21:00,71.9
2026-06-02 22:00,68.4
2026-06-02 23:00,64.0

CSV forecast studio

Forecast your electricity load forecasting series.

This studio is preloaded with the hourly load (mw) example above. Replace it with a CSV of your own history, choose a horizon, and run the forecast. The public demo is free while the Hugging Face Space is available; the API is the paid product.

1 · Paste or upload a CSV2 · Pick a horizon3 · Download the 21 quantiles
1 credit per API call · the web demo is free while the public Space is available

What to watch in the output

Keep one timezone and one unit

A timestamp shift or a kW/MW mix looks like a step change. Normalize to a single timezone and a single unit before the file reaches the model.

Use the high band for reserve

The upper band is a natural starting point for operating reserve. It is not a substitute for your reliability standard; add the reserve margin your operator is required to hold.

Mark meter outages as missing

A missing interval is different from zero load. Leave the cell empty or write NaN; a zero will teach the model that the site can go dark.

Common mistakes

  • —Mixing measured and estimated intervals without a flag. The model cannot tell a bad estimate from a real reading.
  • —Ignoring holidays. A holiday can look like a 20% load drop; if the history has no holidays, the forecast will treat the next one as a normal day.
  • —Forecasting a year with two days of history. The model needs enough cycles to learn the weekly and seasonal shape.

How to put it to work

  1. 1. Start with the sample. Run the preloaded series once so you see the output shape before you debug your own CSV.
  2. 2. Replace it with your history. Keep one timestamp column and one or more numeric columns. Mark outages and stockouts as empty cells, not zeros.
  3. 3. Backtest before you plan. Hold out the last few periods, compare the forecast with what happened, and check whether the actuals land inside the 80% band.
  4. 4. Move it into production. Use the REST API for a nightly job or an internal tool. One credit per successful call; failed calls are refunded.

FAQ

Can I forecast multiple meters at once?

Yes, up to eight related numeric columns. If the meters are independent, separate files are easier to validate and debug.

Does the model use weather?

Not unless you include a weather series as a numeric column. Even then, it learns correlation, not a physical load model. For temperature-driven load, validate the covariate carefully.

What horizon should I use for day-ahead scheduling?

Twenty-four hourly steps is the common day-ahead horizon. The studio accepts up to 36 steps; the API accepts up to 1,024.

Need this as an API?

The same forecast is available as a metered REST endpoint with JSON and a downloadable quantile CSV.

See the API

Related reading