In particle sizing by light extinction method, the regularization parameter plays an important role in applying regularization to find the solution to ill-posed inverse problems. We combine the generalized cross-valid...In particle sizing by light extinction method, the regularization parameter plays an important role in applying regularization to find the solution to ill-posed inverse problems. We combine the generalized cross-validation （GCV） and L-curve criteria with the Twomey-NNLS algorithm in parameter optimization. Numerical simulation and experimental validation show that the resistance of the newly developed algorithms to measurement errors can be improved leading to stable inversion results for unimodal particle size distribution.展开更多
In order to minimize uncertainty of the inversed parameters to the largest extent by making full use of the limited information in remote sensing data, it is necessary to understand what the information flow in quanti...In order to minimize uncertainty of the inversed parameters to the largest extent by making full use of the limited information in remote sensing data, it is necessary to understand what the information flow in quantitative remote sensing model inversion is, thus control the information flow. Aiming at this, the paper takes the linear kernel-driven model inversion as an example. At first, the information flow in different inversion methods is calculated and analyzed, then the effect of information flow controlled by multi-stage inversion strategy is studied, finally, an information matrix based on USM is defined to control information flow in inversion. It shows that using Shannon entropy decrease of the inversed parameters can express information flow more properly. Changing the weight of a priori knowledge in inversion or fixing parameters and partitioning datasets in multi-stage inversion strategy can control information flow. In regularization inversion of remote sensing, information matrix based on USM may be a better tool for quantitatively controlling information flow.展开更多
Conventional artificial neural networks used to solve electrical resistivity imaging （ERI） inversion problem suffer from overfitting and local minima. To solve these problems, we propose to use a pruning Bayesian ne...Conventional artificial neural networks used to solve electrical resistivity imaging （ERI） inversion problem suffer from overfitting and local minima. To solve these problems, we propose to use a pruning Bayesian neural network （PBNN） nonlinear inversion method and a sample design method based on the K-medoids clustering algorithm. In the sample design method, the training samples of the neural network are designed according to the prior information provided by the K-medoids clustering results; thus, the training process of the neural network is well guided. The proposed PBNN, based on Bayesian regularization, is used to select the hidden layer structure by assessing the effect of each hidden neuron to the inversion results. Then, the hyperparameter αk, which is based on the generalized mean, is chosen to guide the pruning process according to the prior distribution of the training samples under the small-sample condition. The proposed algorithm is more efficient than other common adaptive regularization methods in geophysics. The inversion of synthetic data and field data suggests that the proposed method suppresses the noise in the neural network training stage and enhances the generalization. The inversion results with the proposed method are better than those of the BPNN, RBFNN, and RRBFNN inversion methods as well as the conventional least squares inversion.展开更多
For a typical marine shale reservoir in the Jiaoshiba area, Sichuan Basin of China, P-impedance is sensitive for identifying lithology but not suitable for indicating good shale reservoirs. In comparison, density is a...For a typical marine shale reservoir in the Jiaoshiba area, Sichuan Basin of China, P-impedance is sensitive for identifying lithology but not suitable for indicating good shale reservoirs. In comparison, density is an important quantity, which is sensitive for identifying the organic-rich mud shale from non-organic-rich mud shale. Due to the poor data quality and incidence angle range, density cannot be easily inverted by directly solving the ill-posed pre-stack seismic inversion in this area. Meanwhile, the traditional density regularizations implemented by directly using the more robust P-impedance inversion tend to be inaccurate for recovering density for this shale reservoir. In this paper, we combine the P-impedance and the minus uranium to construct the pseudo-P-impedance（PIp） at well locations. The PIp is observed to be sensitive for identifying organic-rich mud shale and has a good correlation with density in this area. We employ the PIp–density relation into the pre-stack inversion framework to estimate density. Three types of regularization are tested on both numerical and field data： These are no regularization, traditional regularization and the proposed approach. It is observed that the proposed method is better for recovering the density of organic-rich mud shale in the Jiaoshiba area.展开更多
基金The present work is supported by National Science Foundation of China （NSFC 50376041）the National High Technology Development 863 Program （2006AA03Z349）the ShuGuang project of Shanghai Educational Development Foundation （04SG49）, which are gratefully acknowledged.
文摘In particle sizing by light extinction method, the regularization parameter plays an important role in applying regularization to find the solution to ill-posed inverse problems. We combine the generalized cross-validation （GCV） and L-curve criteria with the Twomey-NNLS algorithm in parameter optimization. Numerical simulation and experimental validation show that the resistance of the newly developed algorithms to measurement errors can be improved leading to stable inversion results for unimodal particle size distribution.
基金This work was supported by the Special Funds for the Major State Basic Research Project(Grant No.G2000077903)the National Natural Science Foundation of China(Grant No.40171068).
文摘In order to minimize uncertainty of the inversed parameters to the largest extent by making full use of the limited information in remote sensing data, it is necessary to understand what the information flow in quantitative remote sensing model inversion is, thus control the information flow. Aiming at this, the paper takes the linear kernel-driven model inversion as an example. At first, the information flow in different inversion methods is calculated and analyzed, then the effect of information flow controlled by multi-stage inversion strategy is studied, finally, an information matrix based on USM is defined to control information flow in inversion. It shows that using Shannon entropy decrease of the inversed parameters can express information flow more properly. Changing the weight of a priori knowledge in inversion or fixing parameters and partitioning datasets in multi-stage inversion strategy can control information flow. In regularization inversion of remote sensing, information matrix based on USM may be a better tool for quantitatively controlling information flow.
基金supported by the National Natural Science Foundation of China(Grant No.41374118)the Research Fund for the Higher Education Doctoral Program of China(Grant No.20120162110015)+3 种基金the China Postdoctoral Science Foundation(Grant No.2015M580700)the Hunan Provincial Natural Science Foundation,the China(Grant No.2016JJ3086)the Hunan Provincial Science and Technology Program,China(Grant No.2015JC3067)the Scientific Research Fund of Hunan Provincial Education Department,China(Grant No.15B138)
文摘Conventional artificial neural networks used to solve electrical resistivity imaging （ERI） inversion problem suffer from overfitting and local minima. To solve these problems, we propose to use a pruning Bayesian neural network （PBNN） nonlinear inversion method and a sample design method based on the K-medoids clustering algorithm. In the sample design method, the training samples of the neural network are designed according to the prior information provided by the K-medoids clustering results; thus, the training process of the neural network is well guided. The proposed PBNN, based on Bayesian regularization, is used to select the hidden layer structure by assessing the effect of each hidden neuron to the inversion results. Then, the hyperparameter αk, which is based on the generalized mean, is chosen to guide the pruning process according to the prior distribution of the training samples under the small-sample condition. The proposed algorithm is more efficient than other common adaptive regularization methods in geophysics. The inversion of synthetic data and field data suggests that the proposed method suppresses the noise in the neural network training stage and enhances the generalization. The inversion results with the proposed method are better than those of the BPNN, RBFNN, and RRBFNN inversion methods as well as the conventional least squares inversion.
基金NSFC and Sinopec Joint Key Project (U1663207)the China Geology Survey Project (DD20160195)+2 种基金973 Program (2014CB239104)National Key S&T Projects (2017ZX05049002)China Postdoctoral Science Foundation for the financial support
文摘For a typical marine shale reservoir in the Jiaoshiba area, Sichuan Basin of China, P-impedance is sensitive for identifying lithology but not suitable for indicating good shale reservoirs. In comparison, density is an important quantity, which is sensitive for identifying the organic-rich mud shale from non-organic-rich mud shale. Due to the poor data quality and incidence angle range, density cannot be easily inverted by directly solving the ill-posed pre-stack seismic inversion in this area. Meanwhile, the traditional density regularizations implemented by directly using the more robust P-impedance inversion tend to be inaccurate for recovering density for this shale reservoir. In this paper, we combine the P-impedance and the minus uranium to construct the pseudo-P-impedance（PIp） at well locations. The PIp is observed to be sensitive for identifying organic-rich mud shale and has a good correlation with density in this area. We employ the PIp–density relation into the pre-stack inversion framework to estimate density. Three types of regularization are tested on both numerical and field data： These are no regularization, traditional regularization and the proposed approach. It is observed that the proposed method is better for recovering the density of organic-rich mud shale in the Jiaoshiba area.
