【摘要】 本文在分析标准遗传算法的优点和不足的基础上,基于AER模型提出了一种新的遗传算法——Multi-A-gent遗传算法.它利用Agent的局部感知、竞争协同和自学习等特性来实现生物对环境的自适应,从而实现全局优化计算.理论分析证明这种算法是以概率1收敛的.在实验中,我们首先用10个维数为30的标准测试函数来全面测试算法的性能,然后用50～200维的Rastrigin函数来测试算法处理高维函数的能力.结果表明本文算法具有较强的全局优化能力,鲁棒性强,且具有良好的处理高维函数的能力.更多还原

【Abstract】 Based on AER model, a novel genetic algorithm (Multi-Agent (ienetic Algorithm, MAGA) is proposed. The agents in MAGA have local perceptivity, the abilities of competition, cooperation, self-learning etc. MAGA realizes the global optimal computation via these local interacting agents. It is proved theoretically that the MAGA is convergent with probability 1. In experiments, we first apply the proposed algorithm to 10 benchmark functions with 30 dimensions, and the results show that MAGA has strong abilities in finding global optimum and avoiding premature convergence. Afterwards, we test the proposed algorithm on the Rastrigin function from 50 to 200 dimensions, the results indicate that MAGA can optimize the problems with high dimensions and is quite robust.更多还原