20.04.2014 Views

design space pruning heuristics and global optimization method for ...

design space pruning heuristics and global optimization method for ...

design space pruning heuristics and global optimization method for ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

CHAPTER V<br />

APPLICATION OF METHODOLOGY TO LARGER PROBLEMS<br />

In this chapter, the <strong>method</strong>ology is applied in full to two larger problems where<br />

the <strong>global</strong> optimum solution is unknown. The first problem is derived as a modified<br />

version of the 3 rd Global Trajectory Optimization Competition, while the second problem<br />

is modified version of the 2 nd Global Trajectory Optimization Competition. For each<br />

problem, a number of known good solutions exists from the competition results, which<br />

will serve as benchmarks to evaluate the effectiveness of the <strong>method</strong>ology. For both of<br />

these problems, the goal is to find a suite of good solutions <strong>for</strong> subsequent analysis with<br />

higher fidelity <strong>method</strong>s. Additionally, the <strong>method</strong>ology is applied to the full version of<br />

the GTOC2 problem, subject to the time limitations of the competition, in order to<br />

determine where the best solution found in that timeframe would have placed in the<br />

competition.<br />

For the larger problems, which will require greater computational resources, a<br />

computer cluster is utilized to carry out the low-thrust trajectory <strong>optimization</strong>s. The<br />

genetic algorithm is run on a computer cluster comprised of fifteen nodes. A Matlab<br />

code runs on the master node of the cluster, which executes the genetic algorithm <strong>and</strong><br />

distributes the MALTO runs to each of the nodes. The master node contains two AMD<br />

Opteron processors at 2.2 GHz each. MALTO then runs on the cluster nodes. A Fortran<br />

script creates the input files required <strong>for</strong> each of the MALTO runs, based on a batch<br />

script sent to each node by Matlab. Seven of these nodes contain two AMD Opteron<br />

processors with 2.2 GHz each <strong>and</strong> 5 GB of RAM. The remaining eight nodes contain<br />

two dual core AMD Opteron processors with 2.4 GHz each <strong>and</strong> 12 GB of RAM. The<br />

operating system on the computer cluster is Ubuntu 8.04 LTS.<br />

Because of limitations in the Matlab Distributed Computing Server, each<br />

generation of the genetic algorithm was manually distributed to the nodes – e.g., if sixty<br />

105

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!