; GHOh+'0HP
$TU Delft Repository search results0TU Delft Repository search results (max. 1000)TU Delft LibraryTU Delft Library@
qW@
qW՜.+,0HPX`hp
x
WorksheetFeuilles de calcul
B=%r8X"1Calibri1Calibri1Calibri1
Calibri 83ffff̙̙3f3fff3f3f33333f33333.}0TU Delft Repositoryg n4uuidrepository linktitleauthorcontributorpublication yearabstract
subject topiclanguagepublication type publisherisbnissnpatent
patent statusbibliographic noteaccess restrictionembargo datefaculty
departmentresearch group programmeprojectcoordinates)uuid:c7baa01feb374bf1aceb3f58c575bdd1Dhttp://resolver.tudelft.nl/uuid:c7baa01feb374bf1aceb3f58c575bdd1[Parallel Approach to DerivativeFree Optimization: Implementing the DONE Algorithm on a GPUMunnix, J.H.T.Verhaegen, M. (mentor)Researchers at Delft University of Technology have recently developed an algorithm for optimizing noisy, expensive and possibly nonconvex objective functions for which no derivatives are available. The databased online nonlinear extremumseeker (DONE) was originally developed for sensorless wavefront aberration correction in optical coherence tomography (OCT) and optical beam forming network (OBFN) tuning. In order to make the DONE algorithm suitable for largescale problems, a parallel implementation using a graphics processing unit (GPU) is considered. This master thesis aims to develop such a parallel implementation which performs faster than the existing sequential implementation without much change in obtained accuracy. Since OBFN tuning is a problem that may involve a large amount of parameters, an OBFN simulation is to be used to compare the parallel implementation to the sequential implementation. The key of the DONE algorithm is solving a regularized linear leastsquares problem in order to construct a smooth and lowcost surrogate function which does provide derivatives and can be optimized fairly easily. This master thesis first discusses the basics of parallel computing, after which several linear leastsquares methods and several numerical optimization methods are investigated. These methods are compared and the most suitable methods for parallel computing are implemented and tested for increasing dimensions. The final parallel DONE implementation combines the recursive leastsquares (RLS) method with the BroydenFletcherGoldfarbShanno (BFGS) method and optimizes the largescale OBFN simulation almost twice as fast as the sequential DONE implementation, without much change in obtained accuracy.derivativefree; optimization; numerical; algorithm; linear; leastsquares; random; fourier; expansion; rfe; databased; online; nonlinear; extremumseeker; done; parallel; parallelization; graphics; processing; unit; gpu; compute; unified; device; architecture; cudaen
master thesis.Mechanical, Maritime and Materials Engineering+Delft Center for Systems and Control (DCSC))uuid:897b5abe07134c4aad5d6c76d33c8bd2Dhttp://resolver.tudelft.nl/uuid:897b5abe07134c4aad5d6c76d33c8bd2Well Trajectory OptimizationLoomba, A.K.+Jansen, J.D. (mentor); Ashoori, E. (mentor)Delineating the well placement and trajectory of production or injection wells is an important step of any field development program. Drilling a well is an expensive process in terms of money, time and effort invested. Being an expensive item, wells must be carefully studied before being drilled. In order to make this process more efficient, this dissertation presents a computerassisted approach of determining the optimal well trajectory using adjointbased optimization technique. The algorithm is based on surrounding the injection or production wells with dummy wells; a technique proposed in earlier studies. These dummy wells have a minimal rate of injection or production so as to minimalize their influence on the simulated output. The sum of the gradients of the objective function with respect to the flowrate in each dummy well over the lifetime of the reservoir is used to define an improved well trajectory. The whole process is repeated until the maximum net present value is achieved. In order to ensure that the newly optimized well trajectory is drillable, the algorithm restricts the curvature of the trajectory. The optimization algorithm was applied to synthetically generated homogeneous and heterogeneous reservoirs to improve a single we<ll trajectory. The algorithm was also successfully tested to improve multiple well trajectories in the Egg Model , a threedimensional heterogeneous channelized reservoir model. In addition to well trajectory optimization, the dissertation also presents an approach to optimize well numbers. This algorithm is based on drilling a dense quasiwell configuration to get an initial knowledge of the reservoir and utilizing this knowledge along with dummy well gradients to reduce the well count. Depending on the reservoir, fluid and economic parameters, both the algorithms display the adeptness to predict optimized type, trajectory and number of wells. Although the optimization results showed a considerable improvement in the net present value of the project, the algorithm can get stuck in local optimum.Eoptimization; algorithm; gradientbased; well trajectory; well number!Civil Engineering and GeosciencesGeoscience & Engineering!Section for Petroleum Engineering)uuid:38379080da964acda86df3b8f492dd1bDhttp://resolver.tudelft.nl/uuid:38379080da964acda86df3b8f492dd1b'Algorithmic Engineering in Public SpaceHulin, J.; Pavlicek, J.oThe paper reflects on a relationship between an algorithmic and a standard (intuitive) approach to design of public space. A realized project of a plaza renovation in Czech town Vsetin is described as a study case. The paper offers an overview of benefits and drawbacks of the algorithmic approach in the described study case and it outlines more general conclusions.?algorithm; public space; circle packing; optimization; pavementconference paper
*+&ffffff?'ffffff?(?)?"dXX333333?333333?U}}}}}}}}}} }
}}}
}}}}}}}}}}}}
@
!
"
#
$
%
&
'@
(
)
*
+
,

.
/
0t@
1
2
3
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~>@ddyKyKhttp://resolver.tudelft.nl/uuid:c7baa01feb374bf1aceb3f58c575bdd1yKyKhttp://resolver.tudelft.nl/uuid:897b5abe07134c4aad5d6c76d33c8bd2yKyKhttp://resolver.tudelft.nl/uuid:38379080da964acda86df3b8f492dd1bgg
Root Entry F
qW
qW@SummaryInformation( F<Workbook F͆DocumentSummaryInformation8 F
!"#$%&'()*+,./0123456789:;<=>?@ABCDEF