计算机科学与探索 ›› 2021, Vol. 15 ›› Issue (12): 2421-2437.DOI: 10.3778/j.issn.1673-9418.2008044

• 理论与算法 • 上一篇    下一篇

最优权动态控制学习机制的多种群遗传算法

潘家文,钱谦,伏云发,冯勇   

  1. 1. 昆明理工大学 信息工程与自动化学院,昆明 650500
    2. 昆明理工大学 云南省计算机技术应用重点实验室,昆明 650500
  • 出版日期:2021-12-01 发布日期:2021-12-10

Multi-population Genetic Algorithm Based on Optimal Weight Dynamic Control Learning Mechanism

PAN Jiawen, QIAN Qian, FU Yunfa, FENG Yong   

  1. 1. Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China
    2. Yunnan Key Laboratory of Computer Technology Applications, Kunming University of Science and Technology, Kunming 650500, China
  • Online:2021-12-01 Published:2021-12-10

摘要:

遗传算法(GA)的全局搜索能力强,易于操作,但收敛速度慢,易陷入局部极值。为克服上述缺陷,首先对算法初始化方法进行改进,采用海明距离作为聚类划分的相似性度量提出了一种均匀分区多种群初始化方法。该方法以相似性度量为准则划分出不同集合的聚类中心点,然后以偏好随机的方式产生多个不同的种群,避免算法因种群初始个体在解空间分布不够均匀而陷入局部收敛。其次在遗传算法中引入多种群并行机制和学习机制来提高算法的性能,通过对已有研究中两种机制在遗传算法中的作用进行分析,指出各自的优势和不足,分别对两种机制进行改进,提出改进的多种群并行机制与最优权动态控制的学习机制,并从理论角度探讨了改进的两种机制的合理性。最后,将两种机制有机结合起来,充分发挥两种机制的优点,抑制各自的不足之处。仿真实验结果表明,算法中经过改进的两种机制具有良好的沟通能力,结合新的初始化方法,使得算法在收敛速度和精度上都要优于其他几种已有的改进算法。

关键词: 遗传算法(GA), 最优权, 多种群并行机制, 学习机制

Abstract:

Genetic algorithm (GA) has strong global search ability and is easy to operate, but it has some disadvantages, such as slow convergence speed, and easy to fall into local extreme. In order to overcome these disadvantages, an improved genetic algorithm is proposed in this paper. Firstly, instead of the random initialization method, a uniform partition multi-population initialization method is used to generate the initial populations. This method calculates clustering centers by the criterion of Hamming distance, so as to generate different populations. The algorithm can make the initial solutions disperse in the solution space as much as possible, thus avoiding the problem of local extremes. Secondly, the ideas of multi-population parallel mechanism and learning mechanism are introduced to further improve the performance of algorithm. Based on the analysis of advantages and disadvantages of the two mechanisms, new improvements are made to these two mechanisms. Modified multi-population parallel mechanism and optimal weight dynamic control learning mechanism are proposed. In addition, the rationality of the two improved mechanisms is discussed. At last, the above mentioned two mechanisms and the new initialization method are combined. Simulation results show that the proposed algorithm has better performance in convergence speed and accuracy than other genetic algorithms.

Key words: genetic algorithm (GA), optimal weight, multi-population parallel mechanism, learning mechanism