摘要
震源参数和速度结构的联合反演是一个典型的非线性多参数最优化问题，常规的局部线性化反演方法往往易于陷入局部极值，且严重依赖于初始模型的选取。模拟生物界进化的遗传算法则是一种简单而高效的全局性搜索方法。它对初始模型的依赖性不大，不需要求导数，仅需对拟合差函数作出评价，能较好地解决速度结构和震源位置的联合反演问题。在此，简要介绍遗传算法的基本理论和特点，叙述用遗传算法联合反演京津唐张地区速度结构和震源位置的方法，分析遗传算法的迭代过程中拟合差的变化。首次提出在变异过程中引入可变变异概率的研究思路。由于可变变异概率的引入，拟合差明显减小，运用发生在京津唐张地区的１８个地震，２５３条Ｐ波到时数据，对具有７９个未知数的较大型问题进行遗传算法联合反演，得出的地壳模型与由人工爆破和天然地震资料得出的结果相当一致；对震源位置的修定使得台站的到时残差明显减小。研究结果表明遗传算法解决非线性多参数地球物理反演问题是很有效的。
The simultaneous inversion of hypocenter parameters and velocity structure can be considered as a typical nonlinear multiparameter optimization problem. The majority of inversion methods based on local linearization are usually trap in local maximum value,and they seriously depend on the selection of initial model. The genetic algorithm, which modelling the evolution of the biologicals,is a global search methods,which is hardly trap inlocal maximums and has less dependence on initial values.It does not use derivative information. The only requirement is the estimation of the misfit function. This paper gives a brief description of the basic theory and characteristics of the genetic algorithms.We apply genetic algorithms to the velocity structure and hypocenter simultaneous inversion in Jing Jin Tang Zhang, analysis the various of the misfit function in evolution iterations,presents a new research methods in which various mutation probability is used.We apply 18earthquakes took place in research zone, 253 P wave arrivals, simultaneously invert a complex problem which has 79 parameters, obtain a suitable crust model. The modifyed hypocenters reduce the residuals of the P wave arrival.It has been proved that the genetic algorithm is very effective on nonlinear multlparameter geophysical inversion.
出处
《地震地磁观测与研究》
1996年第3期57-66,共10页
Seismological and Geomagnetic Observation and Research
基金
地震科学联合基金会
关键词
联合反演
遗传算法
震源位置
速度结构
地震
combined inversion, crustal model, genetic algorithm, hypocentral location
