Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
The scale in optimization
Posted Jan 24, 2017, 3:45 a.m. EST Optimization Version 5.2a 2 Replies
Please login with a confirmed email address before reporting spam
I am interested in solving problem of optimization, I'm puzzled about 'Scale' in control variables and parameters set.
I wander how 'Scale' influnets the progress of solving, and how to set this value?
I would greatly appreciate any help.
Thanks
I wander how 'Scale' influnets the progress of solving, and how to set this value?
I would greatly appreciate any help.
Thanks
2 Replies Last Post Jan 24, 2017, 9:53 p.m. EST
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Hello,
For numerical reasons, in an optimization problem we want all variables to have roughly the same magnitude, so it is good practice to scale all design variables so that they have similar magnitudes.
For example, if you have an optimization problem with two design variables, Length & Thickness, where:
90 < Length < 110
0.05 < Thickness <0.15
we would apply a scale of 100 to Length, and 0.1 to Thickness. That way, the optimization solver will (internally) solve for the design variables:
0.9 < ScaledLength < 1.1
0.5 < ScaledThickness < 1.5
The latter problem has design variables of similar magnitude, which is better for numerical reasons.
Note also that the objective function and constraint functions should also be scaled such that they are close to unity. For a more detailed example see:
www.comsol.com/blogs/designing-new-structures-with-shape-optimization/
Best Regards,
For numerical reasons, in an optimization problem we want all variables to have roughly the same magnitude, so it is good practice to scale all design variables so that they have similar magnitudes.
For example, if you have an optimization problem with two design variables, Length & Thickness, where:
90 < Length < 110
0.05 < Thickness <0.15
we would apply a scale of 100 to Length, and 0.1 to Thickness. That way, the optimization solver will (internally) solve for the design variables:
0.9 < ScaledLength < 1.1
0.5 < ScaledThickness < 1.5
The latter problem has design variables of similar magnitude, which is better for numerical reasons.
Note also that the objective function and constraint functions should also be scaled such that they are close to unity. For a more detailed example see:
www.comsol.com/blogs/designing-new-structures-with-shape-optimization/
Best Regards,
Please login with a confirmed email address before reporting spam
Posted:
7 years ago
Thanks for your help Walter！Now the function of scale is very clear by your explanation.
Could you explan how to set the step length of sweep in optimization?
Could you explan how to set the step length of sweep in optimization?