遗传算法的改进及其在各向异性介质参数反演中的应用

An improved genetic algorithm and its application in parameter inversion in anisotropic media

目前,常规的遗传算法对变异进程不能实行控制,当变异概率取得较大时,就会以极大的几率使种群中很多成员出现退化的现象,从而使遗传算法变得象蒙特卡洛方法那样,把大量的计算工作浪费在无意义的空间部分,其结果使常规遗传算法不得不采用很小的变异概率。这样做的结果显然不利于算法作全局搜索,存在易陷于局部极值的缺陷。为在扩大模型空间的搜索范围和保持种群多样性的同时,又能对变异过程进行控制,通过引入Metropolis抽样准则,提出了一种改进的遗传算法。同常规算法相比,改进算法更有利于作全局搜索,具有改善遗传算法全局收敛性的效果。针对正交各向异性介质模型,用改进的遗传算法进行了各向异性多参数反演工作,以说明新方法在处理这样高度非线性最优化问题时的处理效果。

The genetic algorithms currently used can not control the course of mutation. Many members of a population will degenerate with a great probability if the mutation probability is high, therefore the genetic algorithms will spend large amount of time on searching in the meaningless model space just like Monte Carlo method. Obviously this is adverse to search extrema globally, giving rise to search extrema locally. In order to extend the searching scope in the model space, at the same time keep the diversity of a population and control the course of mutation, this paper advances an improved genetic algorithm which controls the course of mutation by use of Metropolis sample rule. Comparing with conventional algorithms, the new algorithm is in favor of global searching and improves the global convergence of genetic algorithm The new method is demonstrated on an orthotropic medium model.