CN-121995461-A - Resolution improving method and device for seismic imaging section

CN121995461ACN 121995461 ACN121995461 ACN 121995461ACN-121995461-A

Abstract

The embodiment of the invention relates to a resolution improving method and device for an earthquake imaging section, comprising the steps of collecting and analyzing various earthquake imaging sections, constructing a plurality of earthquake velocity models, obtaining a first target earthquake imaging section based on an earthquake convolution model and the plurality of earthquake velocity models, carrying out migration imaging on the plurality of earthquake velocity models by randomly setting different migration imaging parameters and utilizing a reverse time migration imaging method to obtain a second target earthquake imaging section, expanding the earthquake imaging section with diversity characteristics by utilizing self-adaptive synthetic sampling, adding earthquake noise to the earthquake imaging section, combining generation of an countermeasure network design, initially generating an countermeasure network, training to obtain an initial resolution improving model, and carrying out migration learning on the initial resolution improving model to obtain the target resolution improving model. Therefore, the signal-to-noise ratio and the resolution of the offset profile can be improved, the calculation efficiency is high, and no parameter dependence exists.

Inventors

SUN WEIGUO
LIU HAOJIE
KONG QINGFENG
GE DAMING
DIAO RUI
DU ZHENG
ZHAO MAOQIANG

Assignees

中国石油化工股份有限公司
中国石油化工股份有限公司胜利油田分公司

Dates

Publication Date: 20260508
Application Date: 20241108

Claims (10)

1. A method for improving resolution of a seismic imaging profile, comprising: collecting and analyzing various seismic imaging sections, and constructing a plurality of seismic velocity models; Obtaining a first target seismic imaging profile (high quality seismic imaging profile) based on the seismic convolution model and the plurality of seismic velocity models; Performing migration imaging on the plurality of seismic velocity models by using a reverse time migration imaging method by randomly setting different migration imaging parameters to obtain a second target seismic imaging section (low-quality seismic imaging section); Expanding the seismic imaging sections with diversity features by utilizing self-adaptive synthetic sampling aiming at the second target seismic imaging section, and adding seismic noise to each group of obtained seismic imaging sections; An initial generation countermeasure network combining a generator and a discriminator which generate the countermeasure network design with denoising and resolution improvement functions; taking the seismic imaging profile as the input of network training, and training the initial generation countermeasure network to obtain an initial resolution improvement model; Performing migration learning on the initial resolution enhancement model through actual seismic imaging profile data of a target work area to obtain a target resolution enhancement model applicable to the target work area; And processing the actual seismic imaging profile of the target work area based on the target resolution improvement model to obtain a target seismic imaging profile.
2. The method of claim 1, wherein the collecting and analyzing the plurality of seismic imaging profiles to construct a plurality of seismic velocity models comprises: Acquiring a theoretical seismic section from the existing seismic migration imaging data and acquiring different actual seismic sections from the existing seismic imaging sections of each type of oil field; constructing a seismic texture structure model corresponding to each actual seismic section based on the theoretical seismic section and the actual seismic section; And constructing a plurality of seismic velocity models based on the seismic texture structure model and the characteristics of the rock physical parameters of the work area corresponding to the seismic texture structure model.
3. The method of claim 2, wherein the obtaining a first target seismic imaging profile based on the seismic convolution model and the plurality of seismic velocity models comprises: Obtaining a plurality of reflection coefficient models through a first formula based on the plurality of seismic velocity models, wherein the first formula is: wherein R ( x ) is a reflectance model, v ( x ) is a seismic velocity model; And carrying out convolution operation on the high-frequency wavelet and the reflection coefficient models by adopting an earthquake convolution model theory to obtain a plurality of first target earthquake imaging sections.
4. A method according to claim 3, wherein said performing migration imaging on said plurality of seismic velocity models using a reverse time migration imaging method by randomly setting different migration imaging parameters to obtain a second target seismic imaging profile comprises: Randomly setting different migration imaging parameters for each seismic velocity model to obtain a group of imaging parameter combinations corresponding to each seismic velocity model; Obtaining a simulated seismic record of each seismic velocity model under the corresponding set of offset imaging parameters by using a finite difference numerical simulation method based on each set of offset imaging parameters; Performing migration imaging on the simulated seismic record by applying a reverse time migration imaging method to obtain a corresponding migration imaging section, wherein the migration imaging section is a second target seismic imaging section of the seismic velocity model under a corresponding group of migration imaging parameters.
5. The method of claim 4, wherein expanding the seismic imaging profile with diversity features for the second target seismic imaging profile using adaptive synthesis sampling and adding seismic noise to each resulting set of seismic imaging profiles comprises: performing self-adaptive sampling on each second target seismic imaging profile by using a self-adaptive synthetic sampling method, and expanding each second target seismic imaging profile into a plurality of seismic imaging profiles with diversity characteristics; Adding seismic noise to each seismic imaging section with diversity characteristics to obtain a noisy seismic imaging section containing Gaussian noise and simulated seismic noise.
6. The method of claim 5, wherein training the initially generated countermeasure network with the seismic imaging profile as an input to network training results in an initial resolution enhancement model, comprising: constructing a neural network training set required by network training based on semi-supervised learning based on the noisy seismic imaging section and the first target seismic imaging section; Inputting the first target seismic imaging section as tag-containing data into the initial generation countermeasure network to obtain an initial prediction model; and predicting the noisy seismic imaging section through the initial prediction model, performing confidence analysis on the prediction result, and adding the prediction result with the confidence higher than a preset confidence threshold into a tag set to obtain an initial resolution enhancement model.
7. The method of claim 6, wherein said performing a migration study on said initial resolution enhancement model from actual seismic imaging profile data of a target work area to obtain a target resolution enhancement model applicable to said target work area, comprises: training the initial generation countermeasure network through the actual seismic profile and the neural network training set, performing migration learning by taking the initial resolution enhancement model as a starting point, and optimizing parameters of the initial generation countermeasure network through round training; And dynamically adjusting the learning rate of each round in the training process by adopting the self-adaptive learning rate to obtain a target resolution improvement model applicable to the target work area.
8. A resolution enhancement device for a seismic imaging profile, comprising: The building module is used for collecting and analyzing various seismic imaging sections and building a plurality of seismic velocity models; The acquisition module is used for acquiring a first target seismic imaging section based on the seismic convolution model and the plurality of seismic velocity models; The acquisition module is used for carrying out offset imaging on the plurality of seismic velocity models by randomly setting different offset imaging parameters and utilizing a reverse time offset imaging method to acquire a second target seismic imaging section; The expansion module is used for expanding the seismic imaging sections with diversity characteristics by utilizing self-adaptive synthetic sampling aiming at the second target seismic imaging sections, and adding seismic noise to each group of obtained seismic imaging sections; the design module is used for combining the generation of the initial generation countermeasure network of the generator and the discriminator which are provided with the denoising and resolution improvement functions; the training module is used for training the initial generation countermeasure network by taking the seismic imaging section as the input of network training to obtain an initial resolution improvement model; The training module is used for performing migration learning on the initial resolution enhancement model through actual seismic imaging profile data of a target work area to obtain a target resolution enhancement model applicable to the target work area; And the processing module is used for processing the actual seismic imaging profile of the target work area based on the target resolution improvement model to obtain a target seismic imaging profile.
9. An electronic device comprising a processor and a memory, the processor configured to execute a resolution enhancement program for a seismic imaging profile stored in the memory, to implement the resolution enhancement method for a seismic imaging profile of any one of claims 1-7.
10. A storage medium storing one or more programs executable by one or more processors to implement the method of resolution enhancement of a seismic imaging profile of any of claims 1-7.

Description

Resolution improving method and device for seismic imaging section Technical Field The embodiment of the invention relates to the field of exploration earthquakes, in particular to a method and a device for improving resolution of an earthquake imaging section. Background In the process of performing offset imaging, the problems of complex subsurface geological conditions, inaccurate speed models, noise and other factors are often faced, so that the resolution of an offset imaging section is limited. To solve this problem, the current mainstream method is to process the offset profile by deconvolution, time-varying spectrum prolongation, spectral whitening, and other means, so as to improve the resolution of the imaging profile. The method is characterized in that the time resolution is improved by compressing the seismic wavelet in the processing process, but noise amplification and other adverse effects are caused, the time-varying spectrum is extended to divide the seismic data into different time windows, continuous wavelet transformation is carried out on the seismic data of each time window, spectrum extension is carried out one by one, the seismic data are reconstructed by wavelet transformation inverse transformation, but the method can cause the problem of precision loss under the condition of multiple transformation, the frequency distortion caused by an underground structure is compensated by spectral whitening through smoothing the spectrum of the seismic data, particularly in a high-frequency part, the technology is helpful for balancing the energy distribution of different frequency components, improving the resolution of the seismic data, enabling details of the underground structure to be more clear and visible, but the method usually focuses on the amplitude information of the seismic data, is not sensitive enough for some geologic structures, particularly for fine features caused by phase information, and meanwhile, the low-frequency information can be lost when the spectrum is smoothed to improve the high-frequency resolution. With the advancement of artificial intelligence technology in recent years, a resolution improvement method based on deep learning has also been developed. Such methods improve the resolution of seismic offset imaging by introducing a Convolutional Neural Network (CNN), generating a deep learning network such as a countermeasure network (GAN), and a residual network (ResNet). The Convolutional Neural Network (CNN) realizes effective image reconstruction from low resolution to high resolution by learning the characteristics of the image, introducing up-sampling and the like, but the convolutional layer has insufficient global information capture for certain offset sections, the deep convolutional neural network has a large number of parameters and is easy to cause overfitting, and the residual network (ResNet) solves the gradient vanishing problem by introducing residual connection, so that the network can effectively learn and transfer advanced characteristics in the image when the network is deeper, thereby improving the resolution, but the residual structure pays more attention to larger-scale characteristics when information is transferred and is not sensitive to small-scale details. In the patent application of application number CN2020103655636, a multi-dictionary image super-resolution method based on Gaussian mixture model is disclosed, which features of low-resolution image are extracted by using stationary wavelet transformation, residual features of high-resolution image are extracted, training sample pairs are obtained by overlapping corresponding regions, training sample pairs are classified by Gaussian mixture model, corresponding dictionary pairs are learned for each class, super-resolution reconstruction is carried out on the image by using multiple dictionaries at the same time in reconstruction stage, and quality of reconstruction is further improved by using improved global optimization method. The method not only can train better dictionaries with generalization, but also can avoid the problem that a single global dictionary can not reconstruct image blocks with different structures well, and can reconstruct super-resolution of low-resolution images better. However, using a gaussian mixture model to classify training samples and learn the corresponding dictionary may be risky to fit, especially if the training samples are limited. Overfitting may result in poor performance of the model on unseen data. The patent application No. CN202110213988X relates to a seismic frequency band widening method, a device, a medium and electronic equipment, wherein the method comprises the steps of obtaining seismic data, determining seismic wavelet data of the seismic data according to logging data, inverting odd components and even components of reflection coefficients according to the seismic wavelet data and the seismic data, determining target reflection coeff