LCOE trend in a RAFT Optimization

Wind & Water Computer-Aided Engineering Software Tools
1

Dear all,

I was exploring a constrained optimization problem in WEIS, using the level1 (RAFT) option for a single floating turbine (15MW).
I have defined the following modeling and analysis options:

modelling options:

WISDEM:
RotorSE:
flag: True
spar_cap_ss: Spar_Cap_SS
spar_cap_ps: Spar_Cap_PS
te_ss: TE_reinforcement_SS
te_ps: TE_reinforcement_PS
TowerSE:
flag: True
DriveSE:
flag: True
FloatingSE:
flag: True
BOS:
flag: true

Level1:
flag: True
potential_model_override: 0
trim_ballast: 0
heave_tol: 1
save_designs: True

analysis options:

constraints:
control:
rotor_overspeed:
flag: False
min: 0.0
max: 0.25
Max_PtfmPitch:
flag: True
max: 5.5
Std_PtfmPitch:
flag: True
max: 2.
Max_Offset:
flag: True
max: 30.
floating:
stress:
flag: True
global_buckling:
flag: True
shell_buckling:
flag: True

merit_figure_user:
name: financese.lcoe

driver:
optimization:
flag: True # Flag to enable optimization
solver: COBYLA # Optimization solver. Other options are ‘SLSQP’ - ‘CONMIN’
tol: 1.e-6 # Optimality tolerance
max_iter: 100 # Maximum number of iterations (SLSQP)

The optimization seems to be successfully completed, converging to a final geometry of the platform. Nonetheless, I have noticed that during the process a single iteration shows a lower LCOE value compared to the final converged result, seemingly with no violation of the imposed constraints.

Optimization_path
Optimization_path570×420 31.3 KB

May you please help me to understand the trend followed by the optimization process and the possible issues related to the singular lower LCOE value observed along the convergence path.

Thanks in advance
Best regards,

2

Hi Vincenzo,

Thanks for the question.

I don’t know of any WEIS studies using the merit figure you’ve selected, so my response will contain some level of conjecture.

I am guessing that some constraint has stopped the LCOE from decreasing further. Otherwise, your optimization would not have converged. I recommend finding which constraints are violated or nearly violated. It’s possible that you have found another local minima in those points. The COBYLA solver constructs linear approximations of the merit figure and constraints within a trust region.

We typically use the LN_COBYLA solver (the NLOpt implementation) because the standard scipy solver is known to violate constraints. Also, it’s known that the stress and buckling constraints may not be adequately modeled in WISDEM for floating substructures, so I would verify those constraints against another method, as well.

I hope this helps.

Best, Dan