|May 1, 2014, 09:51||
what are the scaling factors of Optimization objective and constraint functions ?
Join Date: Jun 2011
Posts: 151Rep Power: 7
please let me know why the objective and constraint functions are multiplied by the 0.001 ?
% Optimization objective function with scaling factor
% ex= Objective * Scale
OPT_OBJECTIVE= DRAG * 0.001
% Optimization constraint functions with scaling factors, separated by semicolons
% ex= (Objective = Value ) * Scale, use '>','<','='
OPT_CONSTRAINT= ( MAX_THICKNESS > 0.04 ) * 0.001
|May 15, 2014, 16:16||
Sean R. Copeland
Join Date: Jan 2013
Posts: 40Rep Power: 6
For the optimal design problems, SU2 uses the scipy optimization libraries. There is an initial step that the optimizer takes to perturb the design in its search for the local optimum. We generally scale the objective function to control the size of this initial step to ensure it is not too large. For different objective functions, the value of this scaling factor needs to be adjusted -- in general there's no set of guidelines to select the value, it just takes a little experimentation.
|January 27, 2015, 09:37||
Join Date: Sep 2013
Location: Delft, The Netherlands
Posts: 44Rep Power: 5
it happened to me that if I set the scaling factor to a small number ( e.g. OPT_OBJECTIVE=DRAG*0.001) and I start an optimization with already a solution_flow.dat and a solution_adj_cd.dat, the optimization stops at the end of DSN_001 with the following message:
Optimization terminated successfully. (Exit mode 0)
Current function value: 0.0003926472324
Function evaluations: 1
Gradient evaluations: 1
Where the current function value corresponds exactly to the value of the objective function at the end of the direct simulation multiplied by the scaling factor 0.001. Whereas if the scale is increase to 0.1, or 0.01, the optimization works but it produces negative volume cells at later stages (around DSN_006) and the FGMRES solver diverges.
Any hint here?