CN-121995449-A - Denoising method and device for high-dimensional seismic data

CN121995449ACN 121995449 ACN121995449 ACN 121995449ACN-121995449-A

Abstract

The invention discloses a denoising method and device for high-dimensional seismic data. The method comprises the steps of dividing a high-dimensional seismic data body into a plurality of data blocks by using a sliding window, taking the data blocks with corresponding denoising data as one sample to obtain a sample set, training a transducer model to obtain a denoising model, inputting the plurality of data blocks obtained by dividing the high-dimensional seismic data body into the denoising model, and obtaining the denoised seismic data body of the high-dimensional seismic data body according to an output result of the model. The method can realize intelligent suppression of high-dimensional seismic data noise by utilizing the high-dimensional characteristics of the seismic data based on the deep neural network transducer model, and has high denoising precision.

Inventors

WANG WEI
YONG XUESHAN
GAO JIANHU
CHANG DEKUAN
CHEN DEWU
HE RUN
HE DONGYANG

Assignees

中国石油天然气集团有限公司

Dates

Publication Date: 20260508
Application Date: 20241105

Claims (13)

1. A method for denoising high-dimensional seismic data, comprising: dividing the high-dimensional seismic data volume into a plurality of data blocks using a sliding window; taking a data block with corresponding denoising data as a sample to obtain a sample set; Training a transducer model by using the sample set to obtain a denoising model; and inputting the plurality of data blocks into the denoising model, and obtaining the denoised seismic data body of the high-dimensional seismic data body according to the output result of the model.
2. The method of claim 1, further comprising, prior to training a transducer model using the set of samples: And carrying out waveform data forming 0 processing on the seismic channels contained in part of the obtained data blocks, inputting the data blocks into a transducer model, and carrying out self-supervision pre-training by interpolating the waveform data formed with the 0 seismic channels.
3. The method of claim 1, wherein the segmenting the high-dimensional seismic data volume into a plurality of data blocks using a sliding window comprises: setting the size and sliding step length of a window, wherein the side length of the window is larger than the step length; and sliding the window according to the step length in a shot point domain or a receiving point domain of the high-dimensional seismic data body, and cutting to obtain a data block aiming at each current position of the window to finally obtain a plurality of data blocks.
4. The method of claim 3, wherein the setting the window size and the sliding step size further comprises: the number of seismic traces contained in each data block is determined based on the size of the window and, correspondingly, After the cutting is performed to obtain a data block, the method further comprises the following steps: judging whether the number of the seismic channels contained in the data block is smaller than the determined number of the seismic channels, if so, complementing the seismic channels in the data block in a mode of giving 0 to the seismic waveform data.
5. The method of claim 1, wherein the high-dimensional seismic data volume is a five-dimensional seismic data volume comprising 4 spatial dimensions and 1 temporal dimension; the 4 spatial dimensions include two-dimensional coordinates, offset, and azimuth of the shot point and the receiving point.
6. The method of claim 1, wherein the transducer model comprises an input layer, an encoder, a decoder, and an output layer connected in sequence; The input layer is used for carrying out one-dimensional convolution operation on input data and converting the input data into vectors with fixed dimensions; The encoder comprises a plurality of self-attention modules, wherein each self-attention module calculates similarity by calculating query, key and value matrix of input data, and captures global dependency relationship of high-dimensional seismic data; The decoder is used for recovering signals in the seismic data and suppressing noise by applying a self-attention mechanism to time and space dimensions; The output layer is used for transmitting the output of the decoder to the one-dimensional convolution layer to generate denoised seismic data.
7. The method of claim 6, wherein the transducer model further comprises a position encoding layer disposed between the input layer and the encoder for determining a position encoding of the data block based on two-dimensional coordinates of the shot and the receiver points contained in the data block using a sin function and a cos function, the position encoding being used to determine relative position information of the data block, The decoder is used to suppress noise by applying a self-attention mechanism to the temporal and spatial dimensions, in conjunction with position coding to recover signals in the seismic data.
8. The method of claim 6, wherein the encoder employs a multi-head attention mechanism, each head learning a different characteristic pattern in the data.
9. The method of claim 6, wherein the decoder employs residual connection and layer normalization techniques; the decoder includes a plurality of self-attention modules.
10. The method of any of claims 1-9, wherein prior to dividing the high-dimensional seismic data volume into the plurality of data blocks using the sliding window, further comprising: And carrying out normalization processing on the waveform data of the high-dimensional seismic data volume.
11. A high-dimensional seismic data denoising apparatus, comprising: the data segmentation module is used for segmenting the high-dimensional seismic data volume into a plurality of data blocks by using the sliding window; The sample set establishing module is used for taking a data block with corresponding denoising data as a sample to obtain a sample set; the model training module is used for training the transducer model by utilizing the sample set to obtain a denoising model; And the seismic data denoising module is used for inputting the plurality of data blocks into the denoising model, and obtaining a denoised seismic data volume of the high-dimensional seismic data volume according to an output result of the model.
12. A computer storage medium having stored therein computer executable instructions which when executed by a processor implement the method of denoising high dimensional seismic data of any one of claims 1 to 10.
13. The server is characterized by comprising a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the high-dimensional seismic data denoising method according to any one of claims 1-10 when executing the program.

Description

Denoising method and device for high-dimensional seismic data Technical Field The invention relates to the technical field of seismic data processing in oil and gas geophysical exploration, in particular to a high-dimensional seismic data denoising method and device. Background Under the influence of complex underground structural characteristics and earth surface conditions, the oil and gas exploration in China faces a plurality of problems of large construction organization difficulty, low data signal-to-noise ratio, difficult seismic imaging and interpretation and the like. How to accurately judge the underground medium structure by utilizing the seismic exploration is a continuously pursuing target of scientific researchers, especially in the times of high petroleum exploitation cost and trend fluctuation of petroleum price at present. How to effectively break through the bottleneck of the traditional geophysical data processing method, distinguish noise, see data clearly, fully utilize the data, further effectively reduce the difficulty and risk of exploration and development, improve the exploration precision and oil and gas recovery ratio, control the development cost, and are the troublesome problems which are urgently needed to be solved by the petroleum industry. The seismic data has a large noise type, is easy to couple with effective signals, and is a problem of continuous concern in the exploration geophysical field for identifying and separating noise and keeping the effective signals. Seismic data denoising is one of key steps of preprocessing an oil and gas resource exploration early-stage data set, and is important to improving the resolution of seismic data imaging. The traditional seismic exploration data denoising technical framework is based on signal processing, and a plurality of bottlenecks are encountered at present. 1. The exploration data volume is increased explosively, and the conventional technology is difficult to mine the internal relation of the seismic big data. 2. Depending on the signal and noise model assumptions too much, existing algorithms have difficulty achieving efficient denoising in the face of complex noise types. 3. Traditional algorithms rely on professionals for parameter adjustment, limiting automation of data processing. The exploration industry is urgent to research 'intelligent denoising', and breakthroughs are realized in three aspects of exploration data denoising efficiency, effect and automation by utilizing an artificial intelligent deep learning technology. In seismic survey data processing, seismic data is typically processed by extracting common shots, geophones, or offset gathers from shots, geophones, or offset gathers. The mathematically extracted co-shot gathers (or co-detector, co-offset gathers) are a subset of the original seismic data and physically represent the dimensionality reduction of the original data (five-dimensional seismic data becomes three-dimensional in three-dimensional exploration), which makes the relevant features of the seismic data in high-dimensional space underutilized. Processing based on co-shot gathers, for example, fails to take into account the correlation of adjacent shot data, which to some extent destroys the structure of the original data. The method fully utilizes the correlation characteristics of different azimuth angles, offset distances and the like in the high-dimensional seismic data, and is a physical basis of the high-dimensional seismic data processing (such as denoising and interpolation reconstruction) superior to the low-dimensional condition. Five-dimensional seismic data interpolation based on the traditional method has been studied for many years, and five-dimensional data denoising research based on the traditional method is relatively few. The problems and bottlenecks encountered by the conventional method are generally clear, but five-dimensional intelligent denoising based on deep learning basically belongs to an unmanned area. Disclosure of Invention In order to enrich the types of products, enrich the process routes and increase the selection space, the embodiment of the invention provides a high-dimensional seismic data denoising method and device, and the intelligent suppression of high-dimensional seismic data noise is realized. In a first aspect, an embodiment of the present invention provides a method for denoising high-dimensional seismic data, including: dividing the high-dimensional seismic data volume into a plurality of data blocks using a sliding window; taking a data block with corresponding denoising data as a sample to obtain a sample set; Training a transducer model by using the sample set to obtain a denoising model; and inputting the plurality of data blocks into the denoising model, and obtaining the denoised seismic data body of the high-dimensional seismic data body according to the output result of the model. Optionally, before training the transducer model by using