CN-121978752-A - Method and device for identifying seismic data abnormal channels based on hybrid neural network

CN121978752ACN 121978752 ACN121978752 ACN 121978752ACN-121978752-A

Abstract

The invention belongs to the technical field of seismic exploration data processing, and relates to a method and a device for identifying seismic data abnormal channels based on a hybrid neural network, wherein the identification method comprises the steps of obtaining original single-shot seismic data; the method comprises the steps of obtaining standardized single shot seismic data according to original single shot seismic data, obtaining first characteristic data according to the standardized single shot seismic data, wherein the first characteristic data comprise low-frequency trend characteristic data and frequency spectrum distribution characteristic data or the first characteristic data comprise wavelet coefficient characteristic data and curved wave coefficient characteristic data, obtaining a first matrix according to the standardized single shot seismic data and the first characteristic data, dividing the first matrix into a plurality of first data sub-blocks, obtaining an identification model according to the plurality of first data sub-blocks and a hybrid neural network model, obtaining seismic data to be identified, and identifying abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model.

Inventors

XU YINPO
WANG NAIJIAN
HOU YUXIN
REN GUANG
HOU XICHANG
ZHOU XINYU

Assignees

中国石油集团东方地球物理勘探有限责任公司
中国石油天然气集团有限公司

Dates

Publication Date: 20260505
Application Date: 20251222

Claims (17)

1. The method for identifying the seismic data abnormal channel based on the hybrid neural network is characterized by comprising the following steps of: Acquiring original single shot seismic data; obtaining standardized single shot seismic data according to the original single shot seismic data; Acquiring first characteristic data according to the standardized single shot seismic data, wherein the first characteristic data comprises low-frequency trend characteristic data and spectrum distribution characteristic data, or the first characteristic data comprises wavelet coefficient characteristic data and curvelet coefficient characteristic data; acquiring a first matrix according to the standardized single shot seismic data and the first characteristic data; dividing the first matrix into a plurality of first data sub-blocks; Acquiring an identification model according to the plurality of first data sub-blocks and the hybrid neural network model; Acquiring seismic data to be identified; And identifying abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model.
2. The method for identifying abnormal seismic data channels based on a hybrid neural network according to claim 1, wherein the obtaining standardized single shot seismic data according to the original single shot seismic data specifically comprises: And carrying out normalization processing on the original single shot seismic data to obtain the standardized single shot seismic data.
3. The method for identifying abnormal seismic data traces based on a hybrid neural network according to claim 1, wherein, in the case that the first feature data includes low-frequency trend feature data and spectrum distribution feature data, the acquiring the first feature data according to the normalized single shot seismic data specifically includes: performing median filtering calculation on the standardized single shot seismic data to obtain the low-frequency trend characteristic data; and carrying out Fourier transform on the standardized single shot seismic data to obtain the spectrum distribution characteristic data.
4. The method for identifying seismic-data anomalies based on a hybrid neural network as claimed in claim 3, wherein, In the case of median filtering calculation of the normalized single shot seismic data, the window size of median filtering is 3 to 9.
5. The method for identifying abnormal seismic data traces based on a hybrid neural network according to claim 1, wherein, in the case that the first characteristic data includes wavelet coefficient characteristic data and curvelet coefficient characteristic data, the acquiring the first characteristic data according to the standardized single shot seismic data specifically includes: Performing wavelet transformation on the standardized single shot seismic data to obtain wavelet coefficient characteristic data; And performing curvelet transformation on the standardized single shot seismic data to obtain the curvelet coefficient characteristic data.
6. The method for identifying abnormal seismic data traces based on a hybrid neural network according to claim 1, wherein the acquiring a first matrix according to the normalized single shot seismic data and the first feature data specifically comprises: Acquiring first matrix data according to the standardized single shot seismic data; Acquiring second matrix data and third matrix data according to the first characteristic data; and acquiring the first matrix according to the first matrix data, the second matrix data and the third matrix data.
7. The method for identifying seismic data anomalies based on a hybrid neural network as recited in claim 1, wherein, after dividing the first matrix into a plurality of first data sub-blocks, the identification method further comprises: And storing the corresponding relation between each first data sub-block and the seismic channel in the original single shot seismic data.
8. The method for identifying abnormal channels of seismic data based on a hybrid neural network according to claim 1, wherein the identifying abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model specifically comprises: acquiring standardized seismic data to be identified according to the seismic data to be identified; acquiring second characteristic data according to the standardized seismic data to be identified; acquiring a second matrix according to the standardized seismic data to be identified and the second characteristic data; dividing the second matrix into a plurality of second data sub-blocks; acquiring an abnormal probability map of each second data sub-block according to the plurality of second data sub-blocks and the identification model; acquiring the comprehensive abnormal probability of each seismic channel according to the abnormal probability map; And identifying the abnormal channel in the seismic data to be identified according to the comprehensive abnormal probability and the judging threshold value.
9. The method for identifying seismic-data anomalies based on a hybrid neural network of claim 8, wherein after identifying anomalies in the seismic data to be identified according to the composite anomaly probability and a decision threshold, the method further comprises: And outputting an abnormal track list comprising a track index and an abnormal mark according to the abnormal track.
10. The method for identifying abnormal channels of seismic data based on a hybrid neural network according to claim 9, wherein outputting an abnormal channel list including channel indexes and abnormal markers according to the abnormal channels specifically comprises: performing spatial distribution smoothing processing on the abnormal tracks, and outputting an abnormal track list containing track indexes and abnormal marks; the spatial distribution smoothing is realized by adopting median filtering or Gaussian filtering, and the window size of the median filtering is 3 to 9 under the condition of adopting the median filtering.
11. The method for identifying seismic-data anomalies based on a hybrid neural network as recited in claim 8, wherein, After dividing the second matrix into a plurality of second data sub-blocks, the identification method further comprises: Storing the corresponding relation between each second data sub-block and the seismic channel in the seismic data to be identified; The step of obtaining the comprehensive abnormal probability of each seismic channel according to the abnormal probability map specifically comprises the following steps: Acquiring the corresponding relation between each second data sub-block and the seismic channel in the seismic data to be identified; Mapping the probability value of each second data sub-block in the abnormal probability map back to the corresponding seismic channel according to the corresponding relation; And acquiring the comprehensive abnormal probability of each seismic channel according to all probability values mapped to the same seismic channel.
12. The method of any one of claims 1 to 11, wherein prior to deriving the identification model from the plurality of first data sub-blocks and the hybrid neural network model, the method further comprises: constructing the hybrid neural network model; the hybrid neural network model comprises a feature extraction module formed by alternately connecting a convolution layer and a window multi-head self-attention module.
13. The method for identifying seismic-data anomalies based on a hybrid neural network as recited in claim 12, wherein, The convolution layer uses a 3 x 3 or 5 x 5 convolution kernel, and the window multi-head self-attention module has a window size of 4 x 4 to 16 x 16.
14. The method for identifying seismic-data anomalies based on a hybrid neural network as recited in claim 12, wherein, The hybrid neural network model is a structure with connected encoders and decoders, wherein the encoders downsample through max pooling or step convolution and the decoders upsample through transpose convolution or interpolation.
15. A device for identifying seismic data anomalies based on a hybrid neural network, the device comprising: the first acquisition unit is used for acquiring original single-shot seismic data; The first processing unit is used for acquiring standardized single-shot seismic data according to the original single-shot seismic data; The second processing unit is used for acquiring first characteristic data according to the standardized single shot seismic data, wherein the first characteristic data comprises low-frequency trend characteristic data and spectrum distribution characteristic data, or the first characteristic data comprises wavelet coefficient characteristic data and curvelet coefficient characteristic data; the third processing unit is used for acquiring a first matrix according to the standardized single-shot seismic data and the first characteristic data; a fourth processing unit, configured to divide the first matrix into a plurality of first data sub-blocks; A fifth processing unit, configured to obtain an identification model according to the plurality of first data sub-blocks and the hybrid neural network model; the second acquisition unit is used for acquiring the seismic data to be identified; and the sixth processing unit is used for identifying abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model.
16. A storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the hybrid neural network-based seismic data anomaly trace identification method of any one of claims 1 to 14.
17. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method for identifying seismic data anomalies based on a hybrid neural network of any one of claims 1 to 14 when the computer program is executed.

Description

Method and device for identifying seismic data abnormal channels based on hybrid neural network Technical Field The invention belongs to the technical field of seismic exploration data processing, and particularly relates to a method and a device for identifying seismic data abnormal channels based on a hybrid neural network. Background In the related technology, the seismic exploration is used as a key link of petroleum and natural gas resource development, the quality of seismic data directly determines the reliability of underground structure interpretation and reservoir prediction, and abnormal channel identification is a core premise for guaranteeing data quality. As exploration targets extend toward deep complex formations, abnormal lane identification presents a significant challenge. The traditional identification method has the obvious defects in the characteristic extraction and analysis modes that the statistical method based on the sliding window can only capture local waveform mutation characteristics but can not identify cross-track global abnormality, and the global analysis method based on inter-track correlation can sense inter-track correlation but is extremely insensitive to local fine distortion. The problem of local-global characteristic fracture directly leads to the great reduction of the identification precision in a complex scene, and severely restricts the improvement of the exploration precision. Under the background, research and development of an abnormal channel identification method capable of considering local features and global association becomes an urgent need for improving seismic data processing quality. Disclosure of Invention In a first aspect, an embodiment of the invention provides a method for identifying seismic data abnormal channels based on a hybrid neural network, which comprises the steps of obtaining original single shot seismic data, obtaining standardized single shot seismic data according to the original single shot seismic data, obtaining first characteristic data according to the standardized single shot seismic data, wherein the first characteristic data comprises low-frequency trend characteristic data and spectrum distribution characteristic data or the first characteristic data comprises wavelet coefficient characteristic data and curvelet coefficient characteristic data, obtaining a first matrix according to the standardized single shot seismic data and the first characteristic data, dividing the first matrix into a plurality of first data sub-blocks, obtaining an identification model according to the plurality of first data sub-blocks and the hybrid neural network model, obtaining seismic data to be identified, and identifying abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model. In a second aspect, the embodiment of the invention provides a device for identifying seismic data abnormal channels based on a hybrid neural network, which comprises a first acquisition unit, a first processing unit, a second processing unit and a sixth processing unit, wherein the first acquisition unit is used for acquiring original single shot seismic data, the first processing unit is used for acquiring standardized single shot seismic data according to the original single shot seismic data, the second processing unit is used for acquiring first characteristic data according to the standardized single shot seismic data, the first characteristic data comprises low-frequency trend characteristic data and spectrum distribution characteristic data, or the first characteristic data comprises wavelet coefficient characteristic data and curved wave coefficient characteristic data, the third processing unit is used for acquiring a first matrix according to the standardized single shot seismic data and the first characteristic data, the fourth processing unit is used for dividing the first matrix into a plurality of first data sub-blocks, the fifth processing unit is used for acquiring an identification model according to the plurality of first data sub-blocks and the hybrid neural network model, the sixth processing unit is used for acquiring seismic data to be identified, and the abnormal channels in the seismic data to be identified according to the seismic data to be identified and the identification model. In a third aspect, embodiments of the present invention provide a storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method of the first aspect. In a fourth aspect, embodiments of the present invention provide an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method of the first aspect when the computer program is executed. The beneficial effects brought by the invention are as follows: According to the