16.01.2015 Views

A Tutorial on Variable Neighborhood Search

A Tutorial on Variable Neighborhood Search

A Tutorial on Variable Neighborhood Search

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.

Les Cahiers du GERAD G–2003–46 20<br />

10 C<strong>on</strong>clusi<strong>on</strong>s<br />

The general schemes of <strong>Variable</strong> <strong>Neighborhood</strong> <strong>Search</strong> have been presented, discussed<br />

and illustrated by examples. References to many further successful applicati<strong>on</strong>s are given<br />

in the Appendix. In order to evaluate the VNS research program, <strong>on</strong>e needs a list of<br />

desirable properties of metaheuristics. The following eight <strong>on</strong>es are presented in Hansen<br />

and Mladenović (2003):<br />

(i) Simplicity: the metaheuristic should be based <strong>on</strong> a simple and clear principle, which<br />

should be largely applicable;<br />

(ii) Precisi<strong>on</strong>: steps of the metaheuristic should be formulated in precise mathematical<br />

terms, independent from the possible physical or biological analogy which was an<br />

initial source of inspirati<strong>on</strong>;<br />

(iii) Coherence: all steps of heuristics for particular problems should follow naturally from<br />

the metaheuristic’s principle;<br />

(iv) Efficiency: heuristics for particular problems should provide optimal or near-optimal<br />

soluti<strong>on</strong>s for all or at least most realistic instances. Preferably, they should find<br />

optimal soluti<strong>on</strong>s for most problems of benchmarks for which such soluti<strong>on</strong>s are<br />

known, when available;<br />

(v) Effectiveness: heuristics for particular problems should take moderate computing<br />

time to provide optimal or near-optimal soluti<strong>on</strong>s;<br />

(vi) Robustness: performance of heuristics should be c<strong>on</strong>sistent over a variety of instances,<br />

i.e., not just fine-tuned to some training set and less good elsewhere;<br />

(vii) User-friendliness: heuristics should be clearly expressed, easy to understand and,<br />

most important, easy to use. This implies they should have as few parameters as<br />

possible and ideally n<strong>on</strong>e;<br />

(viii) Innovati<strong>on</strong>: preferably, the metaheuristic’s principle and / or the efficiency and effectiveness<br />

of the heuristics derived from it should lead to new types of applicati<strong>on</strong>s.<br />

As argued there, as well as in the more recent am<strong>on</strong>g the surveys listed in the Appendix,<br />

VNS possesses, to a large extend, all of those properties. This has led to heuristics am<strong>on</strong>g<br />

the very best <strong>on</strong>es for several problems, but more importantly to insight into the soluti<strong>on</strong><br />

process and some innovative applicati<strong>on</strong>s.<br />

Appendix. Sources of Additi<strong>on</strong>al Informati<strong>on</strong> about VNS<br />

Some of Web addresses with sources of informati<strong>on</strong> about <strong>Variable</strong> <strong>Neighborhood</strong> <strong>Search</strong><br />

include

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

Saved successfully!

Ooh no, something went wrong!