简单多智能体进化算法求解高维数值优化问题  

Simple Multi-agent Evolutionary Algorithm for High Dimensional Numerical Optimization Problems

作　　者：刘逸[1,2] 慕彩红[2] 寇卫东[1] 

机构地区：[1]西安电子科技大学通信工程学院,陕西西安710071 [2]西安电子科技大学电子工程学院,陕西西安710071

出　　处：《科技视界》2012年第23期20-24,152,共6页Science & Technology Vision

基　　金：国家自然科学基金资助项目(61003199);中央高校基本科研业务费专项资金资助项目(K50510020015;K5051202019)

摘　　要：为解决高维无约束数值优化问题,提出了一种新的利用智能体寻优的进化算法:简单多智能体进化算法(SimpleMulti-Agent Evolutionary Algorithm,SMAEA)。算法在各世代中均从单个智能体出发进行进化,该智能体代表了待优化函数的一个候选解,它通过自翻转算子加速寻优,并通过自学习过程进化为更好的智能体。在自学习过程中,对原有智能体执行局部搜索算子以产生一个环状多智能体系统,并通过交叉翻转、正交交叉、变异等操作使智能体不断改进。对标准测试函数的仿真实验表明,当问题维数从20增至1,000时,该算法能以较少的评价次数收敛到全局最优值。An evolutionary Mgorithm （Simple Multi-agent Evolutionary Algorithm, SMAEA） is proposed for high dimensional unconstrained numerical optimization problems. During the every generation of SMAEA, the evolution always starts from one agent, which represents a candidate solution to the optimization problem in hand. The agent accelerates the optimization process by using self-flipping operator, and subsequently evolves into a better one through a self-learning process, where a multi-agent system with ring topology is produced after the local search operator is executed on the original agent and the agent is then improved through op- erators including cross-flipping, orthogonal crossover and mutation. Tests on benchmark problems show that when the dimensions are increased from 20 to 1,000, the algorithm can find the good solutions with a small number of function evaluations.

关 键 词：优化 数值优化 进化算法 智能体 多智能体 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]