王鹤^1,2,,
刘梦琳^1,2,
席振铢^1,2,,,
彭星亮^1,2,
何航^1,2
1. 中南大学地球科学与信息物理学院, 长沙 410083
2. 中南大学有色金属成矿预测与地质环境监测教育部重点实验室, 长沙 410083

基金项目: 国家自然科学基金（41304090）和国家重点研发计划课题（2016YFC0303104）联合资助


详细信息

作者简介: 王鹤, 女, 1978年生, 博士, 副教授, 主要从事电磁法数据处理与反演研究.E-mail:wanghe_46@163.com

通讯作者: 席振铢, 男, 1966年生, 教授, 博士生导师, 主要从事电磁法理论与应用研究.E-mail:xizhenzhu@163.com

中图分类号: P631

收稿日期:2017-05-31
修回日期:2017-06-27
上线日期:2018-04-05

Magnetotelluric inversion based on BP neural network optimized by genetic algorithm

WANG He^1,2,,
LIU MengLin^1,2,
XI ZhenZhu^1,2,,,
PENG XingLiang^1,2,
HE Hang^1,2
1. School of Geosciences and Info-Physics, Central South University, Changsha 410083, China
2. Key Laboratory of Metallogenic Prediction of Nonferrous Metals and Geological Environment Monitoring(Central South University), Ministry of Education, Changsha 410083, China


More Information

Corresponding author: XI ZhenZhu,E-mail:xizhenzhu@163.com

MSC: P631

--> Received Date: 31 May 2017
Revised Date: 27 June 2017
Available Online: 05 April 2018


摘要
摘要:为进一步提高大地电磁非线性反演的稳定性、运算效率及准确度，将遗传神经网络算法引入大地电磁反演.首先针对大地电磁二维地电模型建立BP（Back Propagation）神经网络基本框架进行学习训练，网络输入为已知地电模型的视电阻率参数，输出为该地电模型参数；再利用遗传算法对神经网络学习训练过程进行优化，计算出多种地电模型网络连接权值和阈值的最优解；最后将最优连接权值和阈值对未知模型进行反演测试，网络输入为未知地电模型的视电阻率参数，输出为该地电模型参数.模型实验表明：遗传神经网络算法充分结合了遗传算法的全局寻优性和神经网络的局部寻优性，相比单一神经网络算法，在网络学习训练中提高了解的收敛成功率和计算速度，在反演测试中能更准确地逼近真实模型.将遗传神经网络算法与最小二乘正则化反演进行对比，理论模型和实测数据都验证了遗传神经网络算法在大地电磁反演中的可行性和有效性.
关键词: 大地电磁/
反演/
遗传算法/
神经网络

Abstract:To improve nonlinear magnetotelluric (MT) inversion, this work introduces the genetic neural network algorithm. Firstly, a back propagation (BP) neural network frame is constructed for training in different models. The network inputs are the apparent resistivity values of known models, and the outputs are the model parameters. The reasonable network structure is designed by determining the number of network nodes. Secondly, the learning process of the neural network is optimized by using the genetic algorithm to obtain the optimal solution of network connection weights. Finally the trained genetic neural network is verified through inversion, in which the network inputs are the apparent resistivity values of unknown models, and the outputs are the corresponding model parameters. Experimental results show that the genetic neural network can make full use of the global searching capability of the genetic algorithm and the local optimization of the neural network. Compared with the single neural network inversion, the operation efficiency and calculation accuracy of the genetic neural network are improved significantly. By comparing the genetic neural network and least-squares regularization inversion, the tests on synthetic and real data show that this method can be applied to MT data inversion and achieve good results.
Key words:Magnetotelluric (MT)/
Inversion/
Genetic algorithm/
Neural network



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_14473