摘要
从CSAMT信号中提取激电信息有利于提高频率域电磁法反演与解释的精度.目前的研究多以线性反演方法为主,存在依赖初始模型、易陷入局部极值的问题.针对CSAMT信号IP提取问题的非线性和非凸特征,本文提出了一种基于柯西分布和惯性权重的二阶段最小构造混合蛙跳反演方法来提取IP信息.该方法首先利用柯西算子取代随机算子来提高算法的全局搜索能力,并通过引入混沌震荡惯性权重来均衡进化过程中的个体经验和群体经验,保证算法后期的稳定收敛;然后通过引入第二阶段反演过程来强化极化率对观测数据的影响,同时将正则化参数引入混合蛙跳算法的适应度函数来改善反演的多解性问题;最后利用CPU并行计算加速了算法的模因组搜索过程.反演结果表明,上述方法能够较好地重构地电结构和提取激电信息,在加噪环境下具有较强的鲁棒性.相比其他非线性算法(标准混合蛙跳算法SFLA,差分进化算法DE和粒子群优化算法PSO)的反演结果,本文算法具有更强的全局搜索能力和更高的计算效率,适合对微弱的激电信息进行提取.
IP information extraction from CSAMT data contributes to better inversion results. The linear inversion method, which is commonly used in this aspect, has such problems as dependency on the initial model and easily falling into local minimum. Considering the nonlinearity and nonconvexity of extracting IP response, a shuffled frog leaping algorithm （SFLA） with inertia weight and Cauchy distribution is proposed, in the procedure of which an improved two-stage minimum structure inversion approach is adopted. Firstly, the Cauchy operator is used to substitute the random operator to enhance the global search ability of SFLA. And the inertia weight of chaotic oscillation is employed to balance the experience between individuals and groups in the evolutionary process for stable convergence. Secondly, a two-stage inversion strategy is applied to enhance the impact of polarizability in the inversion process. Meanwhile, to solve the multi-solution problem, the regularization factor is utilized to the fitness function of the SFLA. Thirdly, CPU parallel computing is employed to accelerate the local search process of memeplexes in the proposed method. The inversion results show that the proposed algorithm is capable of IP information extraction and geoelectric structure reconstruction, and is also robust in noisy environment. Compared with other nonlinear algorithms （such as SFLA, DE and PSO）, the proposed algorithm has better global searching performance and higher computational efficiency, which is suitable for extraction of weak IP information.
作者
董莉
李帝铨
江沸菠
DONG Li LI Di-Quan JIANG Fei-Bo(School of Geosciences and In fo-Physics, Central South University, Changsha 410083, China Department of Information Science and Engineering, Hunan International Economics University, Changsha 410205, China College of Physics and Information Science, Hunan Normal University, Changsha 410081, China)
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2017年第8期3264-3277,共14页
Chinese Journal of Geophysics
基金
国家重点研发计划深地专项项目(2016YFC0601100)
国家自然科学基金项目(41604117)
中国博士后科学基金项目(2015M580700)
湖南省教育厅科研优秀青年资助项目(16B147
15B138)
湖南省科技计划资助项目(2015JC3067)
中南大学中央高校基本科研业务费专项资金(2015zzts064)联合资助
关键词
混合蛙跳算法
可控源音频大地电磁法
激电信息提取
CPU并行计算
最小构造反演
Shuffled frog leaping algorithm
Controlled source audio frequency magnetotellurics
IP information extraction
CPU parallel eomputing
Minimum structure inversion
