【24h】

An Empirical Evaluation of Stochastic Search Methods in Real-World Telecommunication Domains

机译:真实世界电信领域中随机搜索方法的实证评估

获取原文
获取原文并翻译 | 示例

摘要

In the literature there exist several stochastic methods for solving NP-haid optimization problems approximatively. Examples of such algorithms include (in the order of increasing computational complexity) stochastic greedy search methods, simulated annealing, and genetic algorithms. In this paper we are interested in the problem, which one of these methods is likely to give best performance in practice, with respect to the computational effort required in applying the method. We study this problem empirically by selecting a set of stochastic algorithms with varying computational complexity, and by experimentally evaluating for each method how the goodness of the results achieved improves with increasing computational time. For the evaluation, we use two practical optimization problems closely related to real-world problems in telecommunications. To get a wider perspective of the quality of the results, the stochastic methods are also compared against special-case greedy heuristics. Our investigation indicates that although good results can be achieved by genetic algorithms, simpler stochastic algorithms can achieve similar performance with less computational resources. However, of the methods studied, genetic algorithms appear to be the most consistent in the sense that the variance in the quality of the results obtained is smallest with this approach.
机译:在文献中,存在几种用于近似解决NP-haid优化问题的随机方法。这种算法的示例包括(按计算复杂度递增的顺序)随机贪婪搜索方法,模拟退火和遗传算法。在本文中,我们对该问题感兴趣,就应用该方法所需的计算量而言,这些方法中的哪一种在实践中可能会提供最佳性能。我们通过选择一组具有不同计算复杂度的随机算法,并通过实验评估每种方法的经验性,通过增加计算时间来改善结果的优缺点,从而通过经验研究此问题。为了进行评估,我们使用了两个与电信中的实际问题密切相关的实际优化问题。为了更广泛地了解结果的质量,还将随机方法与特殊情况的贪婪启发法进行了比较。我们的研究表明,尽管遗传算法可以取得良好的结果,但是更简单的随机算法可以用更少的计算资源来实现类似的性能。但是,在所研究的方法中,从这种方法获得的结果质量的差异最小的意义上讲,遗传算法似乎是最一致的。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号