Automatic Characterization Constrained Inversion Of Diverse Geological Structures

Seismic inversion technology plays an important role in seismic exploration.However,limited by the narrow frequency band of seismic data,the accuracy of the forward modeling physical model,and the unknown seismic wavelets,the traditional inversion method has certain problems such as ill-posedness and low resolution.The model-driven approach adds prior information constraints to reduce the impact of the above problems,but the traditional prior information constraints are often represented by one or two single constraints,resulting in limited application scope and difficult to meet complex oil and gas requirements.Exploration accuracy requirements for reservoirs.In recent years,with the development of computing power and machine learning algorithms,more and more machine learning methods have been applied in the field of geophysical exploration,and have shown excellent application potential.The thesis aims to combine the advantages of traditional inversion methods and machine learning,and try to reduce the influence of traditional seismic inversion methods such as difficulty in expressing complex geological structure features and strong dependence on the initial model on the inversion results.On the basis of sparse representation algorithm,based on physical model constraints,an inversion method based on sparse representation constraints is proposed.Specifically,the geological structure features are segmented and extracted from the logging data based on the sparse representation technology,so as to obtain a feature library with diverse structural features,because the feature library in the prior information constraint is derived from the logging data with rich frequency information.Therefore,the feature library also contains rich frequency information and structural features.The feature library is added to the regularization constraint of the inversion process to realize the two-wheel drive inversion of the model and data,which can greatly reduce the solution space.Finally,considering that the actual logging data will be disturbed by external factors and cause meaningless disturbances,this paper combines wavelet transform and sparse representation,and extends the logging data into the wavelet domain for multi-scale decomposition,and the multi-scale components are characterized.Extraction,get the sub-feature library,and finally use the wavelet inverse transform to obtain the pre-processed logging data.The logging data at this time filters out part of the interference information,thereby increasing the stability of the inversion process.