基于改进差分进化算法的大地电磁反演  

Magnetotelluric inversion based on the improved differential evolution algorithm

作　　者：王天意 侯征 何元勋 杨进[3] 晁曼妮 刘国辉 WANG TianYi;HOU Zheng;HE YuanXun;YANG Jin;CHAO ManNi;LIU GuoHui(Key Laboratory of Intelligent Detection and Equipment for Underground Space of BeijingTianjinHebei Urban Agglomeration,Ministry of Natural Resources,Hebei GEO University,Shijiazhuang 050031,China;Hebei Key Laboratory of Strategic Critical Mineral Resources,Hebei GEO University,Shijiazhuang 050031,China;School of Geophysics and Information Technology,China University of Geosciences(Beijing),Beijing 100010,China;The 6th Geological Brigade of Hebei Geology and Mineral Bureau,Shijiazhuang 050051,China)

机构地区：[1]京津冀城市群地下空间智能探测与装备重点实验室,石家庄050031 [2]河北省战略性关键矿产资源重点实验室,石家庄050031 [3]中国地质大学(北京)地球物理与信息技术学院,北京100010 [4]河北省地矿局第六地质大队,石家庄050051

出　　处：《地球物理学进展》2022年第4期1605-1612,共8页Progress in Geophysics

基　　金：国家重点研发计划项目“综合航空物探探测系统集成与方法技术示范研究”(2017YFC0602201);河北省自然科学基金项目“大地电磁三维非线性反演新技术研究”(D2018403090)联合资助。

摘　　要：为弥补传统线性化大地电磁(MT)反演方法需计算高阶偏导矩阵,对细节信息反映能力弱的不足,达到提高大地电磁反演精度的目的,本文尝试引入优化领域的智能优化算法差分进化算法(DE)实现大地电磁的全局优化反演.针对经典DE算法在进行全局寻优时存在收敛慢、易陷入局部最优等问题,从算子自适应及寻优策略集成方面对算法进行研究,提出改进差分进化(TDE)算法,并利用标准测试函数对其进行测试,结果证实了改进后的TDE算法具有寻优速度快、鲁棒性强、不易陷入局部最优的特点.并构建了基于TDE算法的大地电磁优化反演模型,通过理论模拟及实测数据对所构建模型进行验证,结果表明TDE算法可有效提高大地电磁的反演的精度,具较好的可推广性.In order to make up for the deficiencies of traditional linearized Magnetotelluric(MT)inversion methods(such as a calculation of high-order partial derivative matrix and a weak ability to reflect the detailed information),and improve the inversion accuracy of MT,this paper attempts to introduce the intelligent optimization-Differential Evolution algorithm(DE)to realize the global optimization inversion of MT.Differential evolution which was a global optimization algorithm in intelligent optimization has been introduced to the magnetotelluric inversion.This algorithm had high efficiency and strong robustness,and could avoid to calculate the partial derivative matrix in the inversion and to overlook the details information easily by making full use of the whole space parallel search in dealing with the optimization problem.Through the analysis of the standard differential evolution algorithm,we found that the single evolutionary strategy and the fixed operator in the classical DE algorithm will make the optimization process fall into the risk of local optimization easily.To solve the problem of slow convergence and local optimum of classical DE algorithm in the global optimization,the operator adaptation and optimization strategy of classical DE algorithm are integrated and improved,and the Improved Differential Evolution(TDE)algorithm has been tested by standard function.In addition,comparisons in performance of optimization have been made between TDE,standard differential evolution,simulated annealing and genetic algorithm.Combined with the test results of the standard function and other algorithms,the improved TDE algorithm has the following merits:fast optimization speed,strong robustness and not easy getting in local optimization.Finally,an optimal inversion model based on TDE algorithm is constructed,which is verified by combining theoretical simulation and measured data.The results show that the TDE algorithm can effectively improve the accuracy of magnetotelluric inversion,and can be widely used.

关 键 词：改进差分进化算法 大地电磁 非线性反演 

分 类 号：P631[天文地球—地质矿产勘探]