Very Large Scale Optimization

G. Vanderplaats
8th AIAA/USAF/NASA/ISSMO Symposium at Multidisciplinary Analysis and Optimization, Long Beach, CA
Publication Date: 
Sep. 6-8, 2000

As optimization problems grow in size, modern algorithms such as Sequential Quadratic Programming require large amounts of computer memory. Also, the optimizer itself begins to use considerable CPU time, relative to the function and gradient computations. A method is presented here, based on Sequential Unconstrained Minimization Techniques using an Exterior Penalty Function, which requires very little memory to store information and does not require time consuming optimization calculations. The penalty for these advantages is that the optimization requires 3-5 times as many function evaluations to converge to a solution. Examples are presented to demonstrate the method.

Login or register to download the full paper.

Follow Us: Facebook Youtube