用于函数优化的正交Multi-Agent遗传算法

将Multi Agent系统、遗传算法和正交试验设计方法相结合,提出了一种混合进化算法———正交Multi Agent遗传算法。它以Multi Agent系统为基础,通过Agent间的相互作用与每个Agent所具有的知识和自学习功能来提高算法的全局优化能力和收敛速度;同时利用正交试验设计方法产生较好的初始种群和设计正交交叉算子以获得更好的后代;针对正交试验设计产生初始化种群在函数维数很高时需很大存贮空间的缺点,提出了子空间分割法来产生所需的初始化种群,它只需要原来存贮空间的十分之一。首先,对维数为30或100的12个标准测试函数进行仿真试验,结果表明正交Multi Agent遗传算法具有很强的全局优化能力和较快的收敛速度;其次,算法对这些标准测试函数进行高维优化(高达200维),实验结果表明正交Multi Agent遗传算法具有较好的高维搜索能力。

Orthogonal Multi-Agent genetic algorithm and its application in the function optimization problem

Based on Multi-Agent systems, genetic algorithm and orthogonal experimental design method, a hybrid evolutionary algorithm, Orthogonal Multi-Agent genetic algorithm (OMAGA), is proposed. It realizes the global optimal computation via the local interacting agents with abilities of local perceptivity, competition and cooperation, self-learning etc. Then orthogonal design is introduced to generate an initial population of points that are scattered uniformly over the feasible solution space and to generate a crossover operator. As a result, the resulting algorithm is more robust and statistically sound. For high dimensional functions, a large memory is needed in generating an initial population with orthogonal design. To overcome this problem, a method called subspace partition is proposed, so that the memory is one tenth that of original one. In simulation experiments, first the proposed algorithm is applied to 12 benchmark functions with 30 or 100 dimensions, the results show that the OMAGA has a strong ability of global optimization and a high convergence speed. Second, the proposed algorithm is applied to these 12 benchmark functions with 200 dimensions, the result demonstrates that the OMAGA can find optimal or close-to-optimal solutions to high dimension functions.