"预先"进化遗传算法研究

遗传算法在对单目标函数的静态寻优中表现出了良好的收敛性和鲁棒性。但是在优化目标动态变化,要求很快给出优化结果时,由于遗传算法运行时间较长,就难以实现。本文提出的“预先进化遗传算法”,就是为了使遗传算法具有动态性能,能够适应优化目标的动态变化,提高寻优的实时性。算法主要思路是借助于并行计算技术、在确定优化目标(决策)过程中就对可能的多个优化目标函数的优良个体进行培养。一旦优化目标确定,在预先培养的优良个体的基础上可以快速寻优。这样就使遗传算法能够适用于优化目标可变的动态环境。文中描述了“预先”进化遗传算法的实现算法,并证明了算法的有效性。最后对算法进行了验证。通过实例可以看出,采用传统遗传算法,单目标函数优化一般要迭代300次,才能够得到较理想的优化结果。而“预先”进化遗传算法经过150次左右的多个目标同时预先进化后,对具体目标的进化只需要50次左右就收敛到较理想的数值。由于预先进化是与决策具体的优化目标同时进行,因此可以实现优化目标确定后的快速寻优。

Study on Pre-evolution Genetic Algorithm

Genetic algorithm excels at astringency and robustness when used in single static function optimization. But due to it's time consuming operation, it can hardly be used in such applications which needs high real timing, and the optimization objects may change dynamically. In order to enable the genetic algorithm to be dynamic, to fit the dynamic decision optimal object and, consequently to optimize it, we propose the "pre-evolution genetic algorithm". The basic idea is, by the help of the technology of parallel computation, to foster the good individuals for all possible during the process of deciding the optimization object. Once the specific object is decided, it can be optimized quickly on the basis of fostered individuals. In this paper, the realization method of the algorithm and the algorithm's verification is introduced. In the end we testify the validity the algorithm through samples. It is found that it takes the traditional Genetic algorithm around iterating 300 generations to find a good solution for a single function, while it takes about 150 generations for the "pre-evolution" Genetic algorithm, based on the simultaneous evolution for every functions, and about iterating 50 generations for a single function. Because the "pre-evolution" is run together with decision-making process, so this algorithm can quickly find the good solution for a function after it is selected.