Power and gas use predictions based on historical data?

Question

My electricity and gas cumulative usage follows a general pattern that looks like a sigmoid (electricity) and flipped inversed sigmoid (gas) based on temperatures in the respectively summer and winter. See the attached file for examples (actual data).

What would be a good way to predict full-year cumulative utility usage based on the first 4-6 months of the year and all historical data? JMP can't fit to a sigmoid function (I think).

utilitydata.jmp

stan_koprowski · Answer

Hi,

Here is the link to the JMPer Cable article @Mark_Bailey mentioned.

Published monthly, JMP Newswire brings JMP news right to you. Subscribe to stay up to date on new software releases, events, training, tips and stories about how others are using JMP in their organizations.

cheers,

Stan

Mark_Bailey · Answer

Jim's modeling looks sufficient for your purpose. You could then integrated that fitted function for the cumulative result.

I can't find a copy of it at the moment but the Fall 2000 Issue 7 of the JMPer Cable newsletter published an article on page 10 describing how to fit models with trigonometric functions using a custom model with the Nonlinear platform. Pretty easy to do! These fitted models could also be integrated if necessary.

Here is a picture of the salient formula for the starting model.

You could easily modify this model to suit your purpose.

txnelson · Answer

Here is one way to model your data.  I just ran quartic(4) polynomial regression on your data.  I did have to create a numeric column for month, but the results are quite nice.  See below, and attached data table.

prediction.jmp