计算机科学与探索 ›› 2020, Vol. 14 ›› Issue (8): 1427-1440.DOI: 10.3778/j.issn.1673-9418.1909043

• 人工智能 • 上一篇    

大规模优化问题的改进花朵授粉算法

李煜,郑娟,刘景森   

  1. 1. 河南大学 管理科学与工程研究所,河南 开封 475004
    2. 河南大学 商学院,河南 开封 475004
    3. 河南大学 智能网络系统研究所,河南 开封 475004
  • 出版日期:2020-08-01 发布日期:2020-08-07

Improved Flower Pollination Algorithm for Large Scale Optimization Problems

LI Yu, ZHENG Juan, LIU Jingsen   

  1. 1. Research Institute of Management Science and Engineering, Henan University, Kaifeng, Henan 475004, China
    2. School of Business, Henan University, Kaifeng, Henan 475004, China
    3. Institute of Intelligent Networks System, Henan University, Kaifeng, Henan 475004, China
  • Online:2020-08-01 Published:2020-08-07

摘要:

花朵授粉算法(FPA)寻优结构新颖,寻优能力良好,但求解高维优化问题易陷入“维数灾难”。为提高FPA求解大规模优化问题的性能,提出一种改进花朵授粉算法(IFPA)。采用反向学习策略增加种群多样性,充分搜索解空间,提高初始种群质量;在自花授粉阶段,发挥当代最优位置的牵引作用,减少算法迭代代价,提高搜索效率,提出避免维间干扰的方法,采用逐维随机扰动策略对花粉个体进行更新,整体评价后接受更优解,提高了算法局部迭代质量。IFPA仅需3~5个种群个体即可达到满意的优化效果,15个测试函数在100、1 000和5 000维下的仿真结果表明:IFPA的求解精度大幅提高,收敛速度明显加快,鲁棒性强,与FPA、PSO和BA的对比表明,改进算法在处理不同类型大规模优化问题上是具有竞争力的。

关键词: 花朵授粉算法, 反向学习, 逐维随机扰动, 维间干扰, 大规模优化

Abstract:

Flower pollination algorithm (FPA) has a novel optimization structure and good optimization ability, but it is easy to fall into “dimension disaster” when solving high-dimensional optimization problems. In order to improve the performance of FPA in solving large-scale optimization problems, an improved flower pollination algorithm (IFPA) is proposed. The opposition-based learning strategy is adopted to increase the diversity of population and search the solution space sufficiently to improve the quality of the initial population. In the stage of self-pollination, the traction effect of contemporary optimal position is exerted to reduce the iterative cost of the algorithm and the search efficiency is improved. And this paper proposes a new method to avoid interference phenomena among dimensions to improve the local iteration quality. The pollen individual is updated with the strategy of dimension-  by-dimension random disturbance, and the better solution is accepted after the overall evaluation. IFPA only needs 3 to 5 population individuals to achieve satisfactory optimization effect. The simulation results of 15 test functions at the dimensions of 100, 1000 and 5000 show that the solution accuracy of IFPA is greatly improved, the convergence speed is obviously accelerated, and the robustness is strong. Meanwhile, the results also reveal the proposed algorithm is competitive for different types of large scale optimization problems compared with FPA, PSO and BA.

Key words: flower pollination algorithm, opposition-based learning, dimension-by-dimension random disturbance, interference among dimensions, large scale optimization