| | | Summary & Conclusion
 |
|
We have seen that current state-of-the-art optimisation techniques in the RAM problems arena can be improved upon. This was illustrated by comparing random samples with the Hybrid optimisation (as illustrated in figure 6). Sampling the Resources space randomly clearly shows that better designs can be found. Still, the random sample is not regarded as a competitive search technique, since we achieved that through a considerably high computational cost. Hence, better search algorithms could be explored, which would allow for better designs while maintaining a reasonable execution time. |
|
We proposed Genetic Algorithms as these benefit from a degree level of random sampling but at the same time channel this randomness towards a desired direction in an educated manner. As a result, we saw that fewer random points were shown in the results and that the search channelled towards the desired objective. However, since that brought up a competition between two objectives, we identified that the optimal results were not as good as those of the random search. This implies that the Genetic Algorithms technique can be improved upon, with an alternative search technique, such as Ranking.
|
|
Apart from the improvement of the results and of the effectiveness of the search process, the benefit of applying GA as the search engine to the RAM problems makes the optimisation process useful for any model scenario, because it does not require a preliminary knowledge of the physics of the problem. As a result, the optimisation is valid for any desired objective function and for any factors without regard to their effect on that objective function. |
|
last update: December 12, 2006 |
|
|
| | |
