多维连续参数的随机试验蚁群算法优化反演分析

针对一般蚁群算法难以求解优化反演中的多维连续参数优化问题,借鉴进化思想,提出随机试验蚁群算法求解多维连续参数全局优化问题的方法。该方法主要运用随机试验求出每个参数的任一水平对目标函数的影响度,以此影响度为蚂蚁选择下一个参数水平的能见度,从而确定参数水平被选中的概率。同时考虑到当参数被划分的水平很多时目标函数收敛较慢的缺点,提出用方差分析法来确定参数的水平对目标函数的影响是否显著,将对目标函数影响不显著的参数水平其信息素设置为很小的值,从而加速目标函数的收敛。通过对一个岩土工程优化实例,表明了该算法的有效性。

Stochastic experiment ant colony algorithm used in back analysis of multidimensional continuous parameters

For being difficult to solve the problem of multidimensional continuous parameters in the back analysis by the ant colony algorithm,stochastic experiment ant colony algorithm is brought out by the idea of evolution of ant colony algorithm.The method does stochastic experiment to determine the influence of each level of parameter.So the visibility and the probability of each level of parameter is determined by the influence of each level of parameter.At the same time,the analysis of variance is used to make sure whether the level of parameter is important or not.If the level of parameter is not important,then the pheromone is very low,so the speed of the convergence of objective function is accelerated.The validity of the method is proved by an example used in the back analysis of geotechnical engineering.