用单纯形模拟退火混合算法反演地声参数

提出了一种基于下降的单纯形算法和模拟退火算法的混合优化算法用于反演地声参数。下降的单纯形方法对于目标函数的梯度信息相当敏感,因此在参数空间中能迅速地移向极小点,但由于它只能向下移动,因而往往易陷入局部极小且依赖初始条件。而模拟退火算法使用随机选取的目标函数值在参数空间中搜索最佳答案,其结构有利于从局部极小中跳出,但是由于它不能利用目标函数的梯度信息因而对参数空间的搜索效率很低。混合方法结合了下降的单纯形方法与快速模拟退火方法这二者的优点,能更有效、更迅速地获得全局最优点。

Geoacoustic Inversion Via Hybrid Simplex Simulated Annealing Algorithm

In this paper, hybrid inversion algorithm based on DHS(Downhill Simplex) algorithm and FSA(Fast Simulated Annealing) is developed and applied to the problem of determining geoacoustic properties. DHS is sensitive to gradients in objective function and moves effectively downhill, but as it has no mechanism to move uphill, it is prone to becoming trapped in local minima. FSA is designed to widely search the parameter space by using a random process to repeatedly perturb the model, and can move uphill in the objective function in order to escape from local minima. It requires only computation of the objective function. However, since the model perturbations are computed randomly(without gradient information), FSA is inefficient at move downhill and can be computationally expensive. Hybrid algorithm combined with DHS and FSA in an attempt to retain the advantages of each while overcoming their respective weaknesses, can search the parameter space efficeintly.