CN-122017963-A - Method, device and equipment for generating amplitude-preserving offset angle gather by utilizing machine learning acceleration

CN122017963ACN 122017963 ACN122017963 ACN 122017963ACN-122017963-A

Abstract

The application provides a method, a device and equipment for generating a vibration-preserving offset angle gather by utilizing machine learning, wherein the method comprises the steps of obtaining preprocessed three-dimensional pre-stack seismic data, generating all the offset angle gathers of a quasi-imaging area through pre-stack time offset imaging, selecting a partial classical line of the quasi-imaging area, adopting an imaging domain hessian array diagonalization approximate true amplitude gather generation method to obtain the vibration-preserving offset angle gather, taking the offset angle gathers of all CDP point gathers of the classical line as input data, taking the vibration-preserving offset angle gathers corresponding to the input data as labels to establish a training data set, training a deep learning neural network by the training data set to obtain a trained neural network, and processing the quasi-imaging full-area offset angle gather by utilizing the trained neural network to obtain the full-area vibration-preserving offset angle gather. The method can greatly improve the efficiency of generating the three-dimensional seismic migration imaging amplitude-preserving offset angle gather, and can also obtain the amplitude-preserving offset superposition profile.

Inventors

WEI ZHEFENG
ZHU CHENGHONG

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石油勘探开发研究院

Dates

Publication Date: 20260512
Application Date: 20241112

Claims (10)

1. The method for generating the amplitude-preserving offset angle gather by utilizing machine learning acceleration is characterized by mainly comprising the following steps of: Acquiring preprocessed three-dimensional pre-stack seismic data; Directly generating all offset angle gathers of a quasi-imaging area from the three-dimensional pre-stack seismic data through pre-stack time offset imaging; Selecting a partial typical line of a to-be-imaged region, and obtaining a amplitude-preserving offset angle gather by adopting an imaging domain hessian array diagonalization approximate true amplitude gather generation method; Taking the offset angle gathers of all CDP point gathers of the typical line as input data, and taking the amplitude-preserving offset angle gathers corresponding to the offset angle gathers in the input data as tag data to establish a deep learning training data set; training the constructed deep learning neural network according to the deep learning training data set to obtain a trained neural network; and processing the to-be-imaged full-area offset angle gather by using the trained neural network, and predicting to obtain the full-area amplitude-preserving offset angle gather.
2. The method for generating the offset angle gather by machine learning acceleration amplitude-preserving method according to claim 1, wherein the three-dimensional pre-stack seismic data are generated into all the offset angle gather of the quasi-imaging area by pre-stack time offset imaging, and the calculation formula of the offset angle gather is as follows: Wherein x, y represents the horizontal coordinate of an imaging point, t represents imaging time depth, theta represents a prestack migration imaging domain incident angle, M represents a three-dimensional prestack seismic data trace number, n represents a three-dimensional prestack seismic data total trace number accumulated in a total aperture, τ s and τ g respectively represent travel times from a shot point and a detector point to the imaging point, a vector M (x, y: theta) represents a migration angle trace set, and f' (τ s +τ g ; theta) represents derivatives of a seismic data time sequence.
3. The method for generating the amplitude-preserving offset angle gather by machine learning according to claim 1, wherein the calculation formula of the diagonalized approximate true amplitude gather generation method by using the imaging domain hessian array is as follows: R(x,y,t:θ)=H -1 (x,y,t:θ)M(x,y,t:θ) Wherein, the vector R (x, y, t: theta) represents a real reflection coefficient matrix, H -1 (x, y, t: theta) is the inverse of the Heisen matrix corresponding to the incident angle theta, and M (x, y, t: theta) represents the offset angle gather.
4. The method for generating a bias angle gather using machine learning acceleration as claimed in claim 1, wherein the bias angle gather of all CDP point gathers of the classical line is used as input data, and the bias angle gather corresponding to the bias angle gather in the input data is used as tag data to establish a deep learning training data set, specifically: And splitting the data of the offset angle gathers of all CDP point gathers of the typical line by adopting a sliding window, taking each piece of split small data as input data, taking the data of the amplitude-preserving offset angle gathers corresponding to the input data as tag data, and establishing a deep learning training data set.
5. The method for generating the amplitude-preserving offset angle gather by machine learning acceleration according to claim 1, wherein the deep learning neural network adopts an improved U-net structure and comprises 32 convolution layers, a ReLU function adopted by an activation function and 2 fully connected layers added in the middle.
6. The method for generating the amplitude-preserving offset angle gather by machine learning according to claim 1, wherein a calculation formula of a Loss function Loss for training the constructed deep learning neural network is as follows: Where M represents the M-th group of data of the deep learning training data set, M represents the total number of data groups of the deep learning training data set, r m (x, y, t: θ) represents the predicted data of the input data in the M-th group of data, and q m (x, y, t: θ) represents the tag data in the M-th group of data.
7. The method for generating the amplitude-preserving offset angle gather by machine learning according to claim 6, wherein training the constructed deep learning neural network according to the deep learning training data set to obtain a trained neural network comprises the following specific steps: And inputting the deep learning training data set into the constructed deep learning neural network, training the deep learning neural network, and stopping the network training when the Loss function Loss does not continuously descend, wherein the deep learning training data set is the trained neural network.
8. The device for generating the acceleration amplitude-preserving offset angle gather by utilizing the machine learning is characterized by mainly comprising the following modules: The data acquisition module is used for acquiring the preprocessed three-dimensional pre-stack seismic data; The all offset angle gather generation module is used for generating all offset angle gathers of the imaging-simulated area from the three-dimensional pre-stack seismic data through pre-stack time offset imaging; The classical line amplitude-preserving offset angle gather generation module is used for selecting part of classical lines of a to-be-imaged area and obtaining an amplitude-preserving offset angle gather by adopting an imaging domain hessian array diagonalization approximate true amplitude gather generation method; The deep learning training data set building module is used for building a deep learning training data set by taking offset angle gathers of all CDP point gathers of the typical line as input data and taking a amplitude-preserving offset angle gather corresponding to the offset angle gathers in the input data as tag data; The neural network training module is used for training the constructed deep learning neural network according to the deep learning training data set to obtain a trained neural network; And the prediction module is used for processing the to-be-imaged full-area offset angle gather by using the trained neural network and predicting to obtain the full-area amplitude-preserving offset angle gather.
9. An electronic device, comprising: A processor; a memory; and a computer program, wherein the computer program is stored in the memory, the computer program comprising instructions that, when executed by the processor, cause the electronic device to perform the method of any one of claims 1 to 7.
10. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored program, wherein the program, when run, controls a device in which the computer readable storage medium is located to perform the method of any one of claims 1 to 7.

Description

Method, device and equipment for generating amplitude-preserving offset angle gather by utilizing machine learning acceleration Technical Field The application relates to the technical field of seismic exploration, in particular to a method, a device and equipment for generating a amplitude-preserving offset angle gather by utilizing machine learning acceleration. Background In the prior art, the conventional migration imaging method technology generally approximates an inverse operator of a seismic wave propagation operator by using an accompanying operator, so that the method greatly reduces the calculated amount of seismic migration and improves the stability, but also has the problems of reduced spatial resolution, incomplete imaging amplitude corresponding to the reflection coefficient (amplitude preservation) of a reflection interface which changes along with an incident angle, and the like. To obtain a magnitude-preserving offset gather, factors such as wave field prolongation, imaging conditions, data coverage, local superposition and the like must be comprehensively considered. At present, in the prior art, the amplitude-preserving imaging is realized mainly by utilizing the least square offset of the inverse operator of the hessian array. However, the least square offset needs to be iterated for a plurality of times, and is large in calculation amount, so that the least square offset is difficult to apply to actual three-dimensional large-scale data, and only a coverage offset gather of a local area (a target imaging line) can be obtained generally. Currently, machine learning neural network algorithms based on data driving have a certain potential in seismic data processing, especially in accelerating high-computation-power prestack migration imaging, but the prior art mainly processes based on superimposed profile images, and has not been applied to migration trace sets. Disclosure of Invention In view of the above, the application provides a method, a device and equipment for generating a amplitude-preserving offset angle gather by utilizing machine learning acceleration, which aims to solve the defect that the amplitude-preserving offset angle gather of large-scale three-dimensional seismic data is difficult to generate in the prior art. In a first aspect, an embodiment of the present application provides a method for generating a gather of amplitude-preserving offset angles by using machine learning acceleration, which mainly includes the following steps: Acquiring preprocessed three-dimensional pre-stack seismic data; Directly generating all offset angle gathers of a quasi-imaging area from the three-dimensional pre-stack seismic data through pre-stack time offset imaging; Selecting a partial typical line of a to-be-imaged region, and obtaining a amplitude-preserving offset angle gather by adopting an imaging domain hessian array diagonalization approximate true amplitude gather generation method; Taking the offset angle gathers of all CDP point gathers of the typical line as input data, and taking the amplitude-preserving offset angle gathers corresponding to the offset angle gathers in the input data as tag data to establish a deep learning training data set; training the constructed deep learning neural network according to the deep learning training data set to obtain a trained neural network; and processing the to-be-imaged full-area offset angle gather by using the trained neural network, and predicting to obtain the full-area amplitude-preserving offset angle gather. In one possible implementation manner, the three-dimensional pre-stack seismic data is generated into a set of offset angle traces of all the quasi-imaging areas through pre-stack time migration imaging, and a calculation formula of the set of offset angle traces is as follows: Wherein x, y represents the horizontal coordinate of an imaging point, t represents imaging time depth, theta represents a prestack migration imaging domain incident angle, M represents a three-dimensional prestack seismic data trace number, n represents a three-dimensional prestack seismic data total trace number accumulated in a total aperture, τ s and τ g respectively represent travel times from a shot point and a detector point to the imaging point, a vector M (x, y: theta) represents a migration angle trace set, and f' (τ s+τg; theta) represents derivatives of a seismic data time sequence. In one possible implementation, the calculation formula of the diagonalized approximate true amplitude gather generation method using the imaging domain hessian array is as follows: R(x,y,t:θ)=H-1(x,y,t:θ)M(x,y,t:θ) Wherein, the vector R (x, y, t: theta) represents a real reflection coefficient matrix, H -1 (x, y, t: theta) is the inverse of the Heisen matrix corresponding to the incident angle theta, and M (x, y, t: theta) represents the offset angle gather. In one possible implementation manner, the deep learning training data set is established by using t