Font Size: a A A

Research On CSAMT 3D Inversion Based On Quasi Newton Method And Data Space Conjugate Gradient Method

Posted on:2021-03-28Degree:DoctorType:Dissertation
Country:ChinaCandidate:C S DuanFull Text:PDF
GTID:1360330647463051Subject:Earth Exploration and Information Technology
Abstract/Summary:PDF Full Text Request
CSAMT(controlled source audio frequency magnetotelluric method)plays an important role in geophysical exploration due to its advantages of high signal-to-noise ratio and high resolution.However,the CSAMT problem becomes very complicated due to the influence of the near-field effect,shadow effect and copy effect.In many cases,interpretation method of MT(the magnetotelluric sounding method)can not be used to process the CSAMT data,otherwise the interpretation results will be far different from the actual situation,resulting the wrong conclusion of the explanation in exploration solutions.Therefore,it is urgent for scientific research and production to launch practical 3D high resolution CSAMT forward and inverse algorithm and software,and the research in this field has become a hot issue in the field,which has been widely valued by the industry.The core work and innovative achievements of this paper mainly include: first,the 3D CSAMT finite element forward modeling with tilter data is completed,and the multi parameter inversion of tensor impedance and tilter data is realized,which improves the lateral resolution of CSAMT inversion interpretation;second,DCG(data space conjugate gradient)inversion algorithm based on Gauss Newton method is proposed,which is effective for complex model inversion;third,the structural coupling and cross gradient inversion algorithm is studied to provide the basis for multi parameter constrained inversion and additional prior structural information constrained inversion.Finally,using the 3D CSAMT forward and inverse software developed in this paper,the inversion interpretation of DCG and L-BFGS inversion methods is carried out for the measured 3D data of coalbed methane electromagnetic exploration in southern Yanchuan,which verifies the effectiveness of the method and achieves the reverse in depth understanding of the algorithm.The 3D CSAMT forward simulation algorithm of long wire field source is completed.Based on the need of multi parameter and orthogonal source inversion,the tipper parameter is introduced into forward modeling.In order to improve the calculation accuracy of complex medium model,the 3D CSAMT finite element numerical simulation method based on tetrahedral element mesh generation is adopted to realize the forward simulation of tipper data with different frequencies and offsets of long wire source,which verifies the effectiveness and correctness of the algorithm.The practical inversion method of 3D CSAMT with tilter data is introduced and realized.In this paper,the Egbert parameter transformation scheme is applied to the limited memory L-BFGS(Broyden-fletcher-Goldfarb-Shanno)method,and the CSAMT 3D inversion algorithm based on tensor source two direction tilter data is realized.The simulation results show that the high-frequency and far offset distribution characteristics of CSAMT are similar to the magnetotelluric model of plane wave hypothesis,while the near offset or low-frequency data may be strongly affected by the transmitter.The composite data inversion further verifies that CSAMT tilter has higher signal-to-noise ratio and stronger boundary recognition ability than the tilter data of magnetotelluric field.Therefore,in the actual data processing and interpretation,the tipper should be combined with apparent resistivity and phase inversion,which helps to improve inversion resolution and obtain more reasonable results and interpretation.The CSAMT inversion of data space conjugate gradient method(DCG)algorithm is realized.The algorithm is essentially an improved algorithm of iterative Gauss Newton method.Compared with the commonly used gradient algorithm,the algorithm retains more partial derivative information and can improve the resolution of inversion.In the inner loop of the algorithm,CG iteration method is used to avoid the Jacobian matrix calculation problem,which greatly saves the computing resources.From the perspective of the algorithm itself,it has the potential to significantly improve the computational efficiency.However,compared with L-BFGS,the time consumption of DCG is much higher than that of L-BFGS.The reason is that the inner cycle of DCG retains more partial derivative information,which greatly increases the calculation times of "quasi forward modeling" and reduces the practicability of DCG.However,DCG retains more information of sensitivity matrix,and its inversion resolution is better than L-BFGS.In this sense,DCG inversion algorithm has certain practical significance.In order to further improve the inversion resolution by using a variety of prior structural information,a 3D CSAMT inversion method based on structural coupling and cross gradient is proposed,and the joint inversion algorithm of multiple geophysical data or attributes is studied.In this paper,the cross gradient term and prior structure information are selected,and three schemes of fixed,step-by-step and real-time structure are proposed to implement strong constraints on CSAMT.Through the reasonable use of cross gradient inversion method,the constraints on resistivity parameters are not limited to any attribute.It can be synchronously inversed with other geophysical methods or completely separated,but the cross gradient should be matched with a priori input structure.The simulation tests of different field sources and different data volumes show that the cross gradient theory and structural coupling scheme can effectively reproduce the deep medium morphology of complex model,and help to improve the inversion resolution of CSAMT,especially scalar CSAMT.The effectiveness of the three-dimensional DCG and L-BFGS inversion methods is verified by the inversion interpretation of CSAMT orthogonal dual source measured data of coalbed methane in southern Yanchuan.The data processing results show that DCG and L-BFGS are consistent with the test results of the above methods in terms of execution efficiency and inversion resolution,and both obtain good inversion results,which are consistent with the actual geological conditions,showing that both methods have high reliability.According to the actual data of multiple inversion experiments,the results show that: when introducing structural constraints,the input velocity structure can significantly affect the electrical distribution of inversion results.Therefore,the input constraints should meet the actual geological conditions,otherwise the inversion effect will be affected.
Keywords/Search Tags:CSAMT, Tipper, Data Space Conjugate Gradient Method(DCG), Cross Gradient, Limited Memory Broyden-Fletcher-Goldfarb-Shanno(L-BFGS)
PDF Full Text Request
Related items