Ventana software support forum

Hi everyone
I am looking documents or other materials about calibration of vensim model.
I read the manual of vensim. Are there other?
Thank you
Giuseppe

This is a very big topic - you might start by posing the particular problem you're working on, describing the data available and what you'd like to get out of the model.

Yes Tom, calbration is a very big problem. I have some questions.

The first question is how I can select the parameters to put into the vensim calibration form.

The second is how I select the model variables or costants that are calibrated with real data.

For now ... I have these questions.

Have you worked through the example in the Vensim help system?

Typically you have a calculation in your model that corresponds to data you have. The model calculation is what is compared to the real data. The constants that you select are those that you can modify to make the model match the data.

https://www.vensim.com/documentation/usr18.htm

It's generally best to do some hand calibration first, before you start automatically calibrating. I typically build myself a control panel, with graphs that compare model to data, or show other key outputs, and candidate parameters to include in the optimization. If you don't have any intuition about how the parameters affect the outcome or the fit, it's hard to make good decisions.

Nocegius,

Did you ever figured out the calibration? You can also try Kalman Filtering if you are working on prediction model:

https://www.vensim.com/documentation/23630.htm

Thank you for your answers. I did not know kalman filering it seems very interesting. I will try it.
So candidate parameters, could be parameters that are used to simulate growth ? In a disease model candidate parameter could be the infection rate, if I well understand.

It could be the infection rate, especially if you have real data for it to calibrate to.

Generally, a parameter is a candidate if it's (a) influential enough to affect the outcome, (b) not redundant with another parameter, and (c) substantially uncertain (not true by definition, like pi, or well known, like the speed of light).

Thank you for your answers. I have other questions .... sorry

About Kalman Filtering, which is the differences with "normal" calibration? And with behavior reproduction test ?

Thank you
Giuseppe