摘要
针对差分进化算法解决高维问题时求解精度低和过早收敛等问题,提出了一种多子群多策略差分进化算法。首先根据适应度值由小到大排列并按照一定比例将种群划分为三个子种群,针对不同子种群的特点使用不同的变异策略和控制参数。针对第二个适应度值一般的子种群引入学习参数,并提出了一种新的变异策略DE-rand-1 toDE-best-1,该变异策略通过学习参数在全局搜索和局部搜索之间建立一种平衡。针对第三个适应度值较差的子种群引入学习参数和均衡参数,通过学习参数向第一个子种群学习,并用均衡参数对提高收敛速度、精度和易陷入局部最优的状况进行改善。最后,用8个测试函数对所提出的算法性能进行测试,并与差分进化算法的四种基础变异策略进行了比较。实验结果表明,所提出的新算法全局寻优能力更强。
In order to solve the problems of low accuracy and premature convergence,when the differential evolution algo-rithm solves the high-dimensional problems,a differential evolution algorithm of multi-subgroup and multi-strategy is pro-posed.Firstly,arranging the population from small to large and according to the fitness value,t he population is divided into three sub-populations with a certain proportion.Different mutation strategies and control parameters are used for different sub-populations.For the second sub-population with a general fitness value,a learning parameter and a new mutation strat-egy called DE-rand-1 to DE-best-1 are introduced.The mutation strategy is established between global search and local search through the learning parameter to keep a kind of balance.A learning parameter and a equalization parameter are intro-duced for the third sub-population which with a poor fitness value.Learning from the first sub-population through the learning parameter,t he equalization parameter is also used to improve the convergence speed,accuracy and easy to fall into the local optimum.Finally,the performance of the propose algorithm is tested by 8 standard test functions and compared with four basic mutation strategies of differential evolution algorithm.The experimental results show that the proposed new algorithm has stronger global optimization ability.
作者
高媛
姜志侠
刘宇宁
GAO Yuan;JIANG Zhixia;LIU Yuning(School of Mathematics and Statistics,Changchun University of Science and Technology,Changchun 130022;CEC GienTech Technology Co.,Ltd.,Beijing 100192)
出处
《长春理工大学学报(自然科学版)》
2022年第5期123-129,共7页
Journal of Changchun University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金(11426045)
吉林省科技发展计划项目(20180101229JC)。
关键词
差分进化算法
多策略
学习参数
均衡参数
扰动
differential evolution algorithm
multi-strategy
learning parameter
equalization parameter
disturbance