| 基于主成分的时间域航空电磁数据神经网络反演仿真研究(英文) |
| |
| 引用本文： | 朱凯光,马铭遥,车宏伟,杨二伟,嵇艳鞠,于生宝,林君.基于主成分的时间域航空电磁数据神经网络反演仿真研究(英文)[J].应用地球物理,2012,9(1):1-8. |
| |
| 作者姓名： | 朱凯光 马铭遥 车宏伟 杨二伟 嵇艳鞠 于生宝 林君 |
| |
| 作者单位： | 吉林大学仪器科学与电气工程学院 |
| |
| 基金项目： | supported by the National Natural Science Foundation of China (Grant No. 40974039);High-Tech Research and Development Program of China (Grant No.2006AA06205);Leading Strategic Project of Science and Technology, Chinese Academy of Sciences (XDA08020500) |
| |
| 摘 要： | 传统上,时间域航空电磁数据通过拟合迭代反演计算得到大地模型,然而,由于航空电磁数据道间的较强相关性,导致病态反演,并引起超定问题;同时电磁数据的相关性使其与模型参数的映射关系复杂,增加了反演的复杂度。采用主成分分析法将航空电磁数据变换为正交的较少数量的主成分,不仅降低了数据道间的相关性,减小了数据量,同时压制了数据的不相关噪声。本文利用人工神经网络(ANN)逼近主成分与大地模型参数间的映射关系,避免了传统反演算法中雅克比矩阵的复杂计算。层状模型的主成分神经网络与数据神经网络的反演结果对比显示,主成分神经网络反演方法网络结构简单,训练步数少,反演结果好,特别是对于含噪数据。准二维模型的主成分ANN、数据ANN以及Zhody方法的反演结果显示了主成分神经网络具有更接近真实模型的反演效果,进一步证明了主成分神经网络反演方法适合海量航空电磁探测数据反演。
|
| 关 键 词： | 主成分分析 人工神经网络 时间域航空电磁法 反演 电导率 |
PC-based artificial neural network inversion for airborne time-domain electromagnetic data |
| |
| Kai-Guang Zhu,Ming-Yao Ma,Hong-Wei Che,Er-Wei Yang,Yan-Ju Ji,Sheng-Bao Yu,Jun Lin.PC-based artificial neural network inversion for airborne time-domain electromagnetic data[J].Applied Geophysics,2012,9(1):1-8. |
| |
| Authors: | Kai-Guang Zhu Ming-Yao Ma Hong-Wei Che Er-Wei Yang Yan-Ju Ji Sheng-Bao Yu Jun Lin |
| |
| Institution: | 1. Key Laboratory of Geo-exploration Instrumentation, College of Instrumentation and Electrical Engineering Changchun, Jilin University, Changchun, Jilin, 130026, China |
| |
| Abstract: | Traditionally, airborne time-domain electromagnetic (ATEM) data are inverted to derive the earth model by iteration. However,
the data are often highly correlated among channels and consequently cause ill-posed and over-determined problems in the inversion.
The correlation complicates the mapping relation between the ATEM data and the earth parameters and thus increases the inversion
complexity. To obviate this, we adopt principal component analysis to transform ATEM data into orthogonal principal components
(PCs) to reduce the correlations and the data dimensionality and simultaneously suppress the unrelated noise. In this paper,
we use an artificial neural network (ANN) to approach the PCs mapping relation with the earth model parameters, avoiding the
calculation of Jacobian derivatives. The PC-based ANN algorithm is applied to synthetic data for layered models compared with
data-based ANN for airborne time-domain electromagnetic inversion. The results demonstrate the PC-based ANN advantages of
simpler network structure, less training steps, and better inversion results over data-based ANN, especially for contaminated
data. Furthermore, the PC-based ANN algorithm effectiveness is examined by the inversion of the pseudo 2D model and comparison
with data-based ANN and Zhody’s methods. The results indicate that PC-based ANN inversion can achieve a better agreement with
the true model and also proved that PC-based ANN is feasible to invert large ATEM datasets. |
| |
| Keywords: | Principal component analysis artificial neural network airborne time-domain electromagnetics inversion conductivity |
| 本文献已被 CNKI SpringerLink 等数据库收录！ |
|
