Guidelines for developing effective Estimation of Distribution Algorithms in solving single machine scheduling problems
Guidelines for developing effective Estimation of Distribution Algorithms in solving single machine scheduling problems
复制标题
DOI：
10.1016/j.eswa.2010.02.073
复制
发表时间：
2010-09
期刊：
影响因子：
--
通讯作者：
Shih-Hsin Chen;Min-Chih Chen;P. Chang;Qingfu Zhang;Yuh-Min Chen
中科院分区：
文献类型：
--
作者：
Shih-Hsin Chen;Min-Chih Chen;P. Chang;Qingfu Zhang;Yuh-Min Chen
本研究的目标是推断设计有效的分布估计算法 (EDA) 的重要指南。这些指南将增强设计的算法，以平衡 EDA 的集约化和多样化效应。大多数 EDA 的优点是结合了概率模型，可以在不破坏显着基因的情况下生成染色体。然而，这种优势可能会导致 EDA 过早收敛的问题，导致概率模型不再生成多样化的解决方案。此外，由于搜索空间的过度拟合，概率模型无法真正代表总体的一般信息。因此，本研究将通过不同计算时间下 EDA 的收敛速度分析，为设计有效的 EDA 算法得出重要的指导方针。主要思想是通过将 EDA 与其他元启发式方法混合并取代采样新解决方案的程序来逐渐增加群体多样性。据此，本研究进一步提出了一种自适应EA/G来提高EA/G的性能。该算法解决了即时调度环境中具有提前/延迟成本的单机调度问题。实验结果表明，自适应 EA/G 优于 ACGA，并且 EA/G 在不同停止标准下具有统计显着性。因此，本文对于 EDA 领域以及研究调度问题的研究人员来说具有重要意义。
The goal of this research is to deduce important guidelines for designing effective Estimation of Distribution Algorithms (EDAs). These guidelines will enhance the designed algorithms in balancing the intensification and diversification effects of EDAs. Most EDAs have the advantage of incorporating probabilistic models which can generate chromosomes with the non-disruption of salient genes. This advantage, however, may cause the problem of the premature convergence of EDAs resulted in the probabilistic models no longer generating diversified solutions. In addition, due to overfitting of the search space, probabilistic models cannot really represent the general information of the population. Therefore, this research will deduce important guidelines through the convergency speed analysis of EDAs under different computational times for designing effective EDA algorithms. The major idea is to increase the population diversity gradually by hybridizing EDAs with other meta-heuristics and replacing the procedures of sampling new solutions. According to that, this research further proposes an Adaptive EA/G to improve the performance of EA/G. The proposed algorithm solves the single machine scheduling problems with earliness/tardiness cost in a just-in-time scheduling environment. The experimental results indicated that the Adaptive EA/G outperforms ACGA and EA/G statistically significant in different stopping criteria. This paper, hence, is of importance in the field of EDAs as well as for the researchers in studying the scheduling problems.