论文部分内容阅读
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。
Traditionally, the time-domain aeronautical electromagnetic data was calculated by fitting the iterative inversion to obtain the geodetic model. However, due to the strong correlation between the aeromagnetic data tracks, the ill-posed inversion resulted in overdetermined problems. At the same time, the correlation of electromagnetic data The complexity of the mapping relationship between the model parameters and the model parameters increases the complexity of the inversion. Using principal component analysis to transform aerial electromagnetic data into a smaller number of orthogonal principal components not only reduces the correlation between data channels, reduces the amount of data, but also restrain the uncorrelated noise of data. In this paper, the artificial neural network (ANN) is used to approximate the mapping relationship between the principal component and the parameters of the geodetic model, which avoids the complicated calculation of Jacobi matrix in the traditional inversion algorithm. The comparison between the principal component neural network and the data neural network shows that the principal component neural network inversion method has the advantages of simple network structure, small training steps and good inversion results, especially for noisy data. The inversion results of principal component ANN, data ANN and Zhody method of quasi-two-dimensional model show that the principal component neural network has the inversion effect closer to the real model and further proves that the principal component neural network inversion method is suitable for mass airborne electromagnetic detection data Inversion.