Dear NREL community,
I am working on vibration and fatigue mitigation of a barge-type FOWT. The aim is to obtain optimal TMD parameters (stiffness and damping ratio).
I succeeded in coupling OpenFAST and MATLAB for the optimization and i succeeded to run the optimization on HPC (High Performance Computer).
For the optimization, i am using PSO optimizer (Particle Swarm Optimization).
I did two optimization procedures: the first has the aim of finding the optimal damping ratio of the TMD (i.e. 1D optimization) and the second has the aim of finding the optimal damping ratio and stiffness of the TMD (i.e. 2D optimization). The optimization are ran using parallel computing with 30 cores.
The problem is that i obtain the same simulation time for 1D and 2D optimization. Is this possible ?
I expect that the time required for 2D optimization should be twice the one needed for 1D optimization.
What do you think ?
Thank you in advance,
Best regards,
Riad
Dear @Riad.Elhamoud,
How many OpenFAST simulations are being run for each optimization problem? I would guess the computational time is tied to that, as well as how you set up to run the OpenFAST/MATLAB optimization on your HPC, which I am not familiar with.
Best regards,
1 Like
Dear @Jason.Jonkman,
Thank you for your response.
In fact, i think that it is not obvious that the time used for 2D optimization problem should be twice that used for 1D optimization problem. It depends mainly on the type of optimizer used. If it was a gradient-based algorithm, it shoud be twice since those types of algorithm search for the first paramter along a direction then, search for the second parameter in another direction. However, when using population-based algorithm like genetic algorithm or particle swarm algorithm, the global optimum is searched for two optimization parameters at the same time.
Moreover, it also depends on the stopping criteria of the optimization procedure.
Best regards,
Riad
1 Like