Research On Seismic Data Interpolation Methods For Different Data Missing Cases

As an important geophysical exploration method,seismic exploration has the characteristics of high precision and high resolution,and is an important methods of detecting resources such as oil,natural gas,and solid minerals.Seismic exploration receives the seismic wave response of the underground medium by deploying detectors on the surface,processes and images the collected data,and obtains the structural and physical parameters of the underground medium.Ideally,the collected seismic data should be regular and cover the entire space(in time and space directions).However,due to the design of the observation system(e.g.,undersampling of spatial directions due to economic factors),the field acquisition environment(e.g.,ground obstacles,complex terrain conditions,etc.),and the hardware problems of the signal acquisition instruments(e.g.,receivers failure,etc.),the acquired seismic data may be irregularly sampled or sparsely.Many subsequent data processing and imaging methods require regular and fully sampled data as a prerequisite,such as 3D true amplitude wave equation migration(WEM),common azimuth migration,and 3D surface related multiple elimination(SRME)in the early stage of data processing,and the fourdimensional time-shifted seismic exploration in the monitoring phase of oil field development.Seismic data interpolation is an effective method to solve missing data and irregular distribution.However,seismic missing data has various cases depending on the actual environment and production requirements,such as random missing data caused by bad traces during the actual layout of the field observation system;regular data loss caused by increasing the distance between measurement lines and detectors to reduce acquisition costs;large continuous data loss caused by the inability to arrange shot points or bury detectors due to surface conditions such as lakes,industrial and mining sites,and urban areas;and the entire single-shot record is missing along the shot line direction due to problems such as stimulation costs or inability to arrange shot points.Different seismic data interpolation methods need to be selected to deal with the above typical data loss scenarios.More accurate interpolation effects can be achieved by establishing a matching relationship between the interpolation problem and the interpolation method.This paper first describes the spatial distribution characteristics of seismic data from the perspective of the observation system,and analyzes the causes of different data missing cases.At the same time,it reviews the different interpolation basic theories involved in this paper.To solve the problem of limited filter coefficient update capability and inability to handle large areas of continuous missing data in low-order streaming prediction filtering interpolation,a high-order streaming prediction filter interpolation method is proposed,which also reduces the difficulty of parameter selection in similar matrix streaming prediction filter interpolation.To solve the interpolation path problem in the past streaming prediction filter,a "snake" interpolation processing path is developed to solve the problem of multiple initializations of streaming prediction filtering coefficients in the calculation process.In order to solve the problem that the filter coefficients cannot be updated during the interpolation process of streaming prediction filter,and the problem that the similar matrix streaming prediction filter interpolation method requires too many artificial input parameters,a higher-order streaming prediction filter interpolation method is proposed.The "snake" processing path is developed to solve the problem of multiple initializations during calculating filter parameters.In response to the problem of missing data in single-shot recordings,this paper combines the basic principles of the seislet transform with the shot continuation properties to propose an SC-seislet transform method that can effectively compress shot-gather data.Under the framework of compressive sensing theory,a projection onto convex sets iterative threshold interpolation method based on the sparsity of the SCseislet transform is constructed.To solve the intelligent interpolation problem of seismic data with missing values,this paper analyzes the selection strategy of training sets and network structures in deep learning interpolation methods.Based on the forward and backward propagation processes in deep learning theory,and considering the limitations of traditional U-Net networks in achieving ideal training effects and increasing network storage space,this paper uses deep supervision theory and U-Net++ network as the core of the interpolation method to reduce storage space while ensuring calculation accuracy.Given the difficulty of obtaining training shot-gather data with different features and the lack of a universal seismic data training set,this paper proposes a new method of constructing a natural image to seismic data,which is applied to solve seismic data interpolation problems using the obtained standard natural-seismic training set.While ensuring effective seismic signal interpolation reconstruction,the robustness of interpolation for noisy data is also achieved.Finally,different interpolation methods proposed in this paper are compared,and the advantages and disadvantages of each interpolation method are verified.This paper also explores and tests the interpolation methods for natural seismic and fivedimensional data interpolation problems,which expands the application areas of the proposed interpolation methods.This paper proposes a diversified seismic data interpolation scheme for different situations of seismic missing data,providing theoretical basis and technical support to address the problem of seismic missing data with non-stationary features.