Natural Gas Load Forecasting

A day-ahead forecast that scores itself, every day

Demo, illustrative data
All demos
Have a look around
  • The forecast. Switch between winter, shoulder and summer. The two forecasts separate most in summer, which is the opposite of what most people expect.
  • What it is worth. Accuracy converted into the imbalance charges it avoids, on one year and one system.

Illustrative product demo. No client is shown, named or described here, and no figure on this page reflects any client’s measured results. Generic illustrative series, no client or utility data. Illustrative system and tariff. No utility's data, tariff or forecast is reproduced here.

Weather-only baseline
7.71%
1,025 Dth/day average miss
With the model
4.53%
719 Dth/day average miss
Error removed
41%
of the baseline's error, across 365 days
Imbalance avoided
$43,777/yr
on 6,046,026 Dth delivered
Day-ahead demand against what actually happened

The shaded band is the 5% balancing tolerance around actual demand. A forecast inside the band costs nothing; outside it, the excess is charged.

System peak day
38,141 Dth
against 7,800 Dth with no heating load at all
Weather sensitivity
690 Dth
per heating degree day
Days outside the band
191 to 139
baseline, then with the model, across all 365 days

Switch to Summer and the two forecasts separate the most, which is the opposite of what people expect. In winter the weather term dominates and any regression tracks it; in the shoulder and summer the load is mostly behavior, and a weather-only model has nothing to work with.