Optimization procedure

Use this forum to post Vensim related questions.
Post Reply
mikdale
Member
Posts: 28
Joined: Thu Jun 11, 2009 2:51 am

Optimization procedure

Post by mikdale »

Hi all

I have a (hopefully) quick question.

What optimisation method does VenSim use, i.e. what is the value that is being optimised, e.g. residual sum of squares, etc.

I couldn't seem to find it in the manuals

Cheers

Mik
bob@vensim.com
Senior Member
Posts: 1107
Joined: Wed Mar 12, 2003 2:46 pm

Post by bob@vensim.com »

Vensim uses a modifies Powell search. The book numerical recipes in C has a good description of this as do numerous other texts.
tomfid
Administrator
Posts: 3806
Joined: Wed May 24, 2006 4:54 am

Post by tomfid »

It sounds to me like you're actually asking about the objective function.

In a calibration optimization, it is the sum of squared residuals, by whatever weights you select. (If you're using Kalman filtering, it's the log likelihood, more complex but same general idea).

In a policy optimization, it's the simple sum of whatever variables you specify, again with your selected weights.

You can use a policy payoff to do things like robust estimation, where you define a variable
error = ABS(data - model)
and use that for the payoff.

Both kinds of payoffs integrate over the whole period of the simulation, so if you want to consider only the final value of something, you can create an added variable, like
payoffVar = modelVar*PULSE(FINAL TIME,TIME STEP)
i.e. something that returns 0 except at FINAL TIME.

Tom
mikdale
Member
Posts: 28
Joined: Thu Jun 11, 2009 2:51 am

Post by mikdale »

Thanks Tom and Bob

I was asking about the objective function - I couldn't think of the right term.

Cheers

Mik
Post Reply