计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (12): 170-179.DOI: 10.3778/j.issn.1002-8331.2003-0321

• 模式识别与人工智能 • 上一篇    下一篇

融合学习策略和邻域搜索的飞蛾火焰算法

郭佳丽,王秋萍,王晓峰   

  1. 西安理工大学 理学院,西安 710054
  • 出版日期:2021-06-15 发布日期:2021-06-10

Hybrid Moth Flame Algorithm with Learning Strategy and Neighboring Search

GUO Jiali, WANG Qiuping, WANG Xiaofeng   

  1. Faculty of Sciences, Xi’an University of Technology, Xi’an 710054, China
  • Online:2021-06-15 Published:2021-06-10

摘要:

为进一步降低基本飞蛾火焰算法陷入局部最优的概率并提高种群多样性,提出一种融合学习策略和邻域搜索的飞蛾火焰算法。将拟反向学习策略嵌入到火焰更新过程,有助于火焰从局部最优中跳出,并且提供了更高的机会接近问题的未知最优解。对飞蛾种群基于适应度值分群,其中一个群采用排序配对学习策略以实现个体间的信息交流,另一个群采用邻域搜索策略以增加种群多样性,这种并行计算能更快地提升整个种群的质量。选取CEC2017测试函数进行数值实验,测试结果和统计分析表明了所提算法具有更高的求解精度和稳定性。将所提算法用于求解OR-Library中的标准实例,结果验证了所提算法对作业车间调度问题是有效的。

关键词: 飞蛾火焰算法, 作业车间调度问题, 拟反向学习策略, 排序配对学习策略, 邻域搜索策略

Abstract:

In order to decrease the probability of local optima stagnation and improve the population diversity, a hybrid moth flame algorithm with learning strategy and neighborhood search is proposed. The quasi-oppositional learning strategy is embedded into the flame updating process, which is helpful for the flame to jump out of the local optima and has a higher chance to be closer to the unknown optimal solution. Then, moths are divided into two subgroups according to their fitness values. One of the groups is updated by ranking paired learning strategy to realize the information exchange between individuals, and another group is updated by neighboring search strategy to increase the diversity of the population. This kind of parallel computing can improve the quality of the whole population more quickly. The CEC2017 test function set is selected for numerical experiment, and the test results and statistical analysis show that the proposed algorithm has higher accuracy and stronger robustness. Finally, the proposed algorithm is applied to solve the benchmark instances in OR-Library, the experimental results verify that the proposed algorithm is effective while solving the job shop scheduling problem.

Key words: moth flame algorithm, job shop scheduling problem, quasi-oppositional learning, ranking paired learning, neighboring search