Research On Magnetotelluric Regularized Inversion

Magnetotelluric is an important method in geophysical exploration.It plays a great role in the exploration of the deep structure of the earth,the prospecting of oil and gas and geothermal resources,as well as the exploration of water resources and environment protection.The quality of the inversion directly affects the effect of exploration.After a long-term development,the magnetotelluric inversion technique has changed from the initial qualitative interpretation to quantitative analysis and continues to move toward refinement.The inversion interpretation itself is a fusion of multidisciplinary technologies,which integrates mathematics,computers,and geology.Therefore,any improvement in any of these technologies may improve the final inversion effect.In this thesis,we study the method of regularizing inversion of Tikhonov.The research focuses on the selection of regularization factors and regularization terms?i.e.model constraints?.In order to reduce the dependence of the inversion on the initial model and enhance the stability of the solution,an adaptive regularization factor is designed in this paper.Firstly,the initial value of the regularization factor is determined according to the relation between the size of the data fitting term and the model constraint term in the first iteration.Thenan adjustment coefficient q is set,and the trust region optimization algorithm is used for reference to dynamically adjust the qaccording to the ratio range of the reduction of the objective function of each iteration.Finally,according tocertain rules,the regularization factor at present is multipliedby qto get the regularization factor of the next iteration.In this thesis,six theoretical models?two kinds of one-dimensional and four kinds of two-dimensional?are designed.The synthesized data are used to invert and contrast the different regularization factors.The inversion results show that the inversion result of the adaptive regularization factor has a strong stability and increased inversion efficiency.The traditional smooth model constrained inversion results in the unsatisfactory position of the interface with large difference in resistivity.In order to make a clearer boundary of the geological body in the inversion result,the total variation method is used to restrain the model,that is???m-m_r??_L1,and thena very small betais introducedto ensure it has derivative everywhere,giving rise to the changed constraint model:(???m-m_r??_L1²+?²)^1/2.After the trials of the six models mentioned above,the inversion result of the synthesized data indicates that the total variation regularization offers the ability to recognize boundary,but it relies heavily on the initial model.Furthermore,this thesis combines the smooth constraints with the total variation constraints to form a mixed regularization method to build a new model constraint.Constructing a parameter gamma to balance the proportion of the two constraints,after the model trial calculation,the synthesized data inversion results show that the boundary of the abnormal body is clearer,and the dependence on the initial model is reduced.It made the invsersion more stable.