2

GenOPT optimisation

I use GenOPT coupled with Dymola to identify parameters for a thermal collector. As a result as per the GenOPT manual, the program approximates a function similar to the objective function and does parameter identification.

Question1 : Do I have control over the error of approximation in dymola.

Question2 : Since I was using GPS implementation of Hookes Jeeves algorithm, I was getting different minima points for different step lengths. I guess this is because of the nature of the algorithm and objective function.Solution?

ezeraj's avatar
151
ezeraj
asked 2017-06-23 08:30:15 -0500, updated 2017-06-27 02:58:25 -0500
edit flag offensive 0 remove flag close merge delete

Comments

Regarding 1: Yes, the Modelica experiment annotation allows specifying the tolerance, as does BuildingsPy in case you run Dymola with BuildingsPy. Also, the dsin.txt file exposes the tolerance, in case you directly execute dymosim.exe. But be careful as some parameters are evaluated during translation and hence cannot be changed when executing dymosim.exe, which is why we recommend not running dymosim.exe unless you really know what you are doing.

Regarding 2: If your cost function has multiple local minima, then there is no guarantee to find the global minimum. See also the GenOpt manual.

Michael Wetter's avatar Michael Wetter (2017-06-27 12:54:45 -0500) edit
add a comment see more comments

1 Answer

1

Unless your simulation model has a random number generator that is not initialized with the same seed in each run, GenOpt should give the same results if you do not make any change to the simulation model or the GenOpt files. There is nothing random in GenOpt, other than a random number generator in some algorithms, but this uses the same seed each time you start the optimization.

Michael Wetter's avatar
1.7k
Michael Wetter
answered 2017-06-23 13:59:04 -0500
edit flag offensive 0 remove flag delete link

Comments

Yes I agree to your point.

My fault, my question was not clear: I use GenOPT coupled with Dymola to identify parameters for a thermal collector. As a result as per the GenOPT manual, the program approximates a function similar to the objective function and does parameter identification.

Question1 : Do I have control over the error of approximation in dymola.

Question2 : Since I was using GPS implementation of Hookes Jeeves algorithm, I was getting different minima points for different step lengths. I guess this is because of the nature of the algorithm and objective function.Solution?

ezeraj's avatar ezeraj (2017-06-27 02:58:18 -0500) edit
add a comment see more comments