|
Title:
|
|
Design optimization for wind turbines: optimization algorithms: a stateof the art study
|
|
|
|
Author(s):
|
|
|
|
|
|
Published by:
|
Publication date:
|
|
ECN
|
1996
|
|
|
|
ECN report number:
|
Document type:
|
|
ECN-C--96-030
|
ECN publication
|
|
|
|
Number of pages:
|
Full text:
|
|
31
|
Download PDF
|
Abstract:
The aim of the study on the title subject was to search for a suitablealgorithm for numerical optimizing of the geometry of wind turbine rotor
blades. The objectives were to look for a minimization procedure which also
minimizes the number of function evaluations. This is mainly due to the fact
that function evaluation costs huge amounts of computer time. On a 468 100
MHz computer the evaluation of a specific rotor design still costs a few
minutes, and this is only for determining one set of the stationary
performance coefficients, e.g. the power coefficient Cp or the rotor torque
coefficient Cq or the rotor axial coefficient Cdax. For future extensions
of the BLADOPT program, e.g. to HATOPT, with dynamic time simulations with
PHATAS or an other aeroelastic code, it is necessary to use these function
evaluations as efficiently as possible, because these kind of function
evaluations still costs more than 1 hr on a modern work station. The
optimization scheme chosen is based on a so called global approximate
optimization. With approximate optimization is meant to make an approximate
model of the real problem and perform the optimization on that simplified
model. Such an approximate model can be made by using e.g. (orthonormal)
polynomial curve fitting algorithms. The parameters in the approximate model
are the design parameters. Using classical optimization algorithms on the
approximate problem, the combination of design parameter values is found to
come up with a minimum value of the object function. The design parameters
belonging to this solutions are then used in a 'real' function evaluation.
With the new function value the approximate model will be updated. This
procedure will be performed in a loop until the new function evaluations do
not lead to other minima. The foremost virtue of the global approximate model
is that it successively includes all real function evaluations, thus not
discarding or forgetting any parameter combinations that have been
calculated. This prevents repeating the real evaluations of a parameter
combination that has already been explored. Such repetitions frequently
afflict (very) simple search algorithms. 24 refs.
Back to List