摘要
遗传算法是一种随机全局搜索算法，与常规的基于局部线性化的最优化方法相比对初始模型的依赖性大为减弱，但是存在着有效基因丢失和早熟收敛问题．采用多尺度逐次逼近反演思想而建立的多尺度逐次逼近遗传算法，能有效地解决上述问题．用该算法对大地电磁资料进行反演，理论曲线和实测资料的试算结果表明多尺度逐次逼近遗传算法能够自动反演地电参数．
Multiscale genetic algorithm (MGA) is proposed in mis paper bycombining multiscale inversion (MI) with genetic algorithm (GA) The new efficientalgorithm circumvents the problems of 'genehc drift' and premature convergenceexisting in classic genetic algorithm, which searches from a randomly chosenpopulation of models and work with binary code of the model parameter set. Byrepeating the GA ophndzation procedure for several times with different binary modelparameter code controlled by multiscale model space, we derive a very good subset ofmodels from the entire model space. The inversion results of synthetic and fieldmagnetotelluric sounding data indicate that MGA enhances the global convergence andimproves the convergence velocity.
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2000年第1期122-130,共9页
Chinese Journal of Geophysics
基金
国家自然科学基金!49674227
关键词
多尺度
遗传算法
大地电磁
非线性反演
最优化
Muhscale, Genetic algorithm, Magnetotelluric sounding, Nonlinearinversion, Optimization method.
