Choose a model, set horizon, and tune parameters. See metrics, confidence bands, and residual checks. Download results, validate assumptions, and iterate with new data.
Paste time series data, pick a model, and generate forecasts.
Use this structure for your own datasets.
| timestamp | load | temperature |
|---|---|---|
| 2026-02-22 06:00 | 1168 | 23.6 |
| 2026-02-22 07:00 | 1325 | 24.8 |
| 2026-02-22 08:00 | 1462 | 26.2 |
| 2026-02-22 09:00 | 1510 | 27.0 |
| 2026-02-22 10:00 | 1488 | 27.6 |
| 2026-02-22 11:00 | 1451 | 28.0 |
Electric load usually contains ramps, peaks, and recovery periods. In hourly datasets, daily seasonality is often 24, while weekly is 168. Many utilities see weekend demand drop 5–20% versus weekdays, and summer peaks can exceed winter peaks by 10–40% in hot regions. The calculator lets you test both season lengths and choose the better error profile.
Paste columns as timestamp, load, and optional drivers such as temperature, humidity, solar output, tariffs, or occupancy. As a rule of thumb, use at least 2,000 hourly rows, or 300 daily rows, to learn seasonal structure. Include one or two strong drivers before adding many weak features. Missing values can be forward filled or linearly interpolated; outliers can be clipped using an IQR factor (commonly 1.5 to 3.0) to reduce sensor spikes.
Moving averages provide fast smoothing when demand is stable. Simple exponential smoothing fits level-only behavior and reacts through α. Holt adds trend using α and β, which helps when demand grows or declines across months. Holt‑Winters adds seasonality and works well for repeating cycles, such as daily HVAC patterns. Regression combines time, season dummies, and feature variables, and is effective when temperature explains peaks or when policy changes shift baselines.
To reduce overfitting, hold out the last portion of history as a test slice. The calculator reports MAE for typical absolute error, RMSE to emphasize large misses, and MAPE for percentage error. With low-load periods, MAPE can inflate, so compare MAE and RMSE too. Use the chart to check bias: consistent under-forecasting near peaks suggests missing drivers or a longer season.
Results include cleaned history, test forecasts, future point forecasts, and a heuristic confidence band using RMSE with horizon scaling. Use CSV for dashboards, simulations, or capacity planning sheets. Use PDF for audits, procurement notes, and executive summaries. Re-run scenarios across horizons, seasons, and features to quantify risk, reserve margins, and demand-response headroom.
For planning, convert forecasts to energy by summing intervals, then add contingency factors. Track model drift weekly, and refresh parameters after major outages or significant tariff revisions.
Works with hourly, daily, or weekly series. Keep intervals consistent; if timestamps are irregular, aggregate first. For hourly, provide several weeks to capture weekday structure. For daily, include at least one seasonal cycle.
Use it when demand shows clear repeating seasonality and a gradual trend. Set season length to the cycle size, such as 24 or 168. If seasonality is weak or unstable, SES or Holt may generalize better.
Select drop, forward fill, or linear interpolation. Dropping removes incomplete rows; fill methods replace missing loads and features. Linear suits short gaps, while forward fill can be safer for step changes or meter resets.
It is a heuristic interval: forecast ± z·RMSE·√h. It reflects recent error scale and widens with horizon. It is not a guarantee and may be optimistic if errors change during extreme weather.
Yes. Add numeric columns after load for temperature, humidity, solar, price, or engineered calendar flags. For categories, convert them to 0/1 indicators before pasting so regression can use them correctly.
MAPE divides by actual load, so small loads inflate percentages in off-peak periods. If MAE and RMSE are stable, treat MAPE cautiously and inspect errors around peaks and operational thresholds.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.