CN116068651A - Seismic data classification method based on time attention mechanism - Google Patents
Seismic data classification method based on time attention mechanism Download PDFInfo
- Publication number
- CN116068651A CN116068651A CN202310272897.2A CN202310272897A CN116068651A CN 116068651 A CN116068651 A CN 116068651A CN 202310272897 A CN202310272897 A CN 202310272897A CN 116068651 A CN116068651 A CN 116068651A
- Authority
- CN
- China
- Prior art keywords
- features
- spectrogram
- seismic data
- layer
- feature extraction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
- G01V1/40—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging
- G01V1/44—Seismology; Seismic or acoustic prospecting or detecting specially adapted for well-logging using generators and receivers in the same well
- G01V1/48—Processing data
- G01V1/50—Analysing data
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V2210/00—Details of seismic processing or analysis
- G01V2210/60—Analysis
- G01V2210/61—Analysis by combining or comparing a seismic data set with other data
- G01V2210/616—Data from specific type of measurement
- G01V2210/6169—Data from specific type of measurement using well-logging
Landscapes
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- Remote Sensing (AREA)
- General Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention relates to a seismic data classification method based on a time attention mechanism, which comprises the following steps: collecting seismic data, and processing the seismic data to generate a spectrogram, wherein the seismic data is time sequence data; extracting features of the spectrogram to obtain time period features and global features with weights; and obtaining a classification result of the seismic data according to the weighted time period characteristics and the weighted global characteristics. According to the method, the shallow layer features, the deep layer features, the time period features and the global features of the seismic data are extracted, fusion judgment is carried out, the attention of the deep learning model to the global attributes of the seismic data can be improved, the features of the original data are fully utilized, the processing capacity of the features of the seismic data can be optimized, and the classification precision of the seismic data is improved.
Description
Technical Field
The invention relates to the technical field of seismic detection, in particular to a seismic data classification method based on a time attention mechanism.
Background
The hydraulic fracturing technology is used as an evaluation means applied to low permeability oilfield crack exploration, plays an extremely important role in oil and gas exploration, and is widely applied to monitoring of oilfield yield increase, shale gas and other new energy exploitation. By deploying seismometers in wells and at the surface, a large amount of seismic data resulting from hydraulic fracturing can be recorded, so that in cases of limited computational resources, it is desirable to extract the useful information contained in the data from these data as efficiently as possible during data processing and analysis. The deep neural network has the advantages of automatic feature extraction and self-learning, and has succeeded in many classification problems, and the existing deep learning model is mostly used for solving related tasks of seismic data classification, but the same processing strategy is adopted for all features extracted from the seismic data, so that the correlation among the features is not well utilized.
The time-based attention mechanism is a resource allocation scheme for weakening the attention to useless information under the condition of limited computing resources, allocating the computing resources to more important tasks and solving the problem of information overload. When the deep learning model is utilized to process the seismic data, a time-based attention mechanism is introduced, so that the model focuses on information which is more critical to the current task, the attention degree to useless information is reduced, even the useless information is filtered, and the correlation among characteristics is fully utilized to solve the problem of seismic data classification. Therefore, the invention provides a seismic data classification method based on a time attention mechanism to solve the technical problems.
Disclosure of Invention
The invention aims at solving the technical problems and provides a seismic data classification method based on a time attention mechanism, which optimizes the capability of processing the characteristics of seismic data and improves the classification precision of the seismic data.
In order to achieve the above object, the present invention provides the following solutions:
a method of classifying seismic data based on a time-attention mechanism, comprising:
collecting seismic data, and processing the seismic data to generate a spectrogram, wherein the seismic data is time sequence data;
extracting features of the spectrogram to obtain time period features and global features with weights;
and obtaining a classification result of the seismic data according to the weighted time period characteristics and the weighted global characteristics.
Further, processing the seismic data includes:
and converting the time sequence data into frequency domain data by adopting frequency domain transformation, and storing the frequency domain data into a spectrogram, wherein the frequency domain transformation comprises fast Fourier transformation and short-time Fourier transformation.
Further, obtaining the weighted time period feature and the global feature includes:
and extracting the shallow features of the spectrogram, and respectively extracting the time period features and the global features of the spectrogram by adopting a parallel branch structure for the shallow features of the spectrogram.
Further, extracting shallow features of the spectrogram includes:
the method comprises the steps of constructing a shallow feature extraction module based on a convolutional neural network, inputting a spectrogram into the shallow feature extraction module to extract shallow features of the spectrogram, wherein the shallow feature extraction module comprises two-dimensional convolutional layers, a batch normalization layer and an ELU activation layer, the two-dimensional convolutional layers are connected in a residual form, and the batch normalization layer and the ELU activation layer are connected behind the two-dimensional convolutional layers.
Further, the parallel branching structure includes an upper branching portion and a lower branching portion;
the upper branch part is used for extracting deep features of the spectrogram through shallow features of the spectrogram and extracting time period features of the spectrogram based on the extracted deep features;
the lower branch part is used for extracting global features of the spectrogram through shallow features of the spectrogram.
Further, extracting deep features of the spectrogram includes:
the deep feature extraction method comprises the steps of constructing a deep feature extraction module based on a residual network, inputting shallow features of a spectrogram into the deep feature extraction module to extract deep features of the spectrogram, wherein the deep feature extraction module comprises a residual block, the residual block comprises a direct mapping unit and a residual unit, summation is carried out after output results of the direct mapping unit and the residual unit are obtained, the direct mapping unit comprises a two-dimensional convolution layer, a batch normalization layer, an ELU activation layer and a Dropout layer, and the residual unit comprises the ELU activation layer and a maximum pooling layer.
Further, extracting the time period feature of the spectrogram includes:
a time-based attention mechanism is used for constructing a time period feature extraction module, deep features of the spectrogram are input into the time period feature extraction module to extract time period features of the spectrogram, wherein the time period feature extraction module comprises a weight calculation unit and a feature extraction unit, and feature stitching is performed after a result of parallel calculation output of the weight calculation unit and the feature extraction unit is obtained;
the weight calculation unit comprises a global average pooling layer, a full connection layer, a Mish activation layer and a Sigmoid function, performs feature channel compression on the deep features based on the global average pooling layer, generates weights of the feature channels through the full connection layer, the Mish activation layer and the Sigmoid function, and weights the weights to the deep features; the feature extraction unit comprises a two-dimensional convolution layer and an ELU activation layer, and further feature extraction is carried out on the deep features.
Further, extracting global features of the spectrogram includes:
the method comprises the steps of constructing a global feature extraction module based on a convolutional neural network, inputting shallow features of a spectrogram into the global feature extraction module to extract global features of the spectrogram, wherein the global feature extraction module comprises a two-dimensional convolutional layer, a global average pooling layer, a full-connection layer and an ELU activation layer, performing feature degradation and further deep feature extraction on the shallow features through the two-dimensional convolutional layer and the global average pooling layer, mapping the extracted further deep features by utilizing the full-connection layer to obtain the global features, and connecting the ELU activation layer after the full-connection layer.
Further, obtaining the classification result of the seismic data includes:
and carrying out feature fusion on the extracted time period features and the global features of the spectrogram, and inputting a Softmax function to obtain a classification result of the seismic data.
The beneficial effects of the invention are as follows:
the method extracts shallow features of seismic data through a convolutional neural network, then adopts a parallel branch structure, applies a residual error network to extract deep features in an upper branch, and uses a time-based attention mechanism to obtain time period features with weights for the deep features; in the lower branch, global features are extracted from shallow features using convolutional neural networks. The fusion judgment is carried out on the extracted depth features, so that the attention of the deep learning model to the global attribute of the seismic data can be improved, the features of the original data are fully utilized, the processing capacity of the seismic data features is optimized, and the classification precision of the two types of seismic data is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for classifying seismic data based on a time attention mechanism according to an embodiment of the present invention;
FIG. 2 is a block diagram of a deep feature extraction module according to an embodiment of the invention;
FIG. 3 is a block diagram of a time period feature extraction module according to an embodiment of the invention;
fig. 4 is a block diagram of a global feature extraction module according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
The embodiment provides a seismic data classification method based on a time attention mechanism, as shown in fig. 1, and the specific flow includes:
s1, acquiring seismic data, and performing frequency domain transformation on the seismic data
And detecting signals through a seismometer, acquiring time series data, then converting the time series data into frequency domain data by applying frequency domain transformation methods such as fast Fourier transformation, short-time Fourier transformation and the like, and storing the frequency domain data in a spectrogram form.
S2, extracting shallow layer characteristics
Firstly, extracting shallow features of an input spectrogram by using a convolutional neural network, wherein a shallow feature extraction module consists of 15 two-dimensional convolutional layers, and comprises 1 two-dimensional convolution with a convolution kernel 7*7, 3 two-dimensional convolutions with a convolution kernel 5*5 and 11 two-dimensional convolutions with a convolution kernel 3*3, different convolution kernels are connected in a residual form, and each convolution layer is connected with a batch of normalization layers and an ELU activation layer. The application of the batch normalization layer can play a certain regularization effect, improve the convergence rate of the model and prevent the model from being over fitted.
S3, extracting time period features and global features by adopting a parallel branch structure
S3.1 extracting time period features in the upper branch
The upper branch is composed of a deep feature extraction module and a time period feature extraction module, and the two modules are connected in series. In the upper branch, deep feature extraction is carried out on shallow features through a residual error network, and then a time-based attention mechanism is used on the extracted deep features so as to obtain time period features with importance weights.
As shown in fig. 2, in the deep feature extraction module, deep feature extraction is performed through a residual network, where the residual network is composed of two residual blocks, each of which is divided into a direct mapping unit and a residual unit, and summation operation is performed after output results of the direct mapping unit and the residual unit are obtained.
The method comprises the steps that a residual unit of two residual blocks consists of an ELU activating layer and a maximum pooling layer, a direct mapping unit of a first residual block consists of two convolution kernels of 3*3 two-dimensional convolution layers, a batch normalization layer, an ELU activating layer and a Dropout layer, first deep feature extraction is carried out, input shallow features are subjected to feature splicing with the first extracted deep features after passing through the ELU activating layer and the maximum pooling layer, and the first residual block is repeatedly stacked for two times; the direct mapping unit of the second residual block consists of two convolution layers of 3*3 two-dimensional convolution layers, two batch normalization layers, two ELU activation layers and one Dropout layer, the second deep feature extraction is carried out, the input shallow features are subjected to feature splicing with the deep features extracted for the first time after passing through the two ELU activation layers and one maximum pooling layer, and the second residual block is repeatedly stacked for seven times.
As shown in fig. 3, in the time period feature extraction module, a time period feature with importance weight is acquired from deep features by using a time-based attention mechanism, the time period feature extraction module is stacked six times in series and is divided into a weight calculation branch and a feature extraction branch, and the output result of the weight calculation branch and the output result of the feature extraction branch are spliced to obtain the time period feature with the weight.
The weight calculation branch performs further time period feature extraction on deep features extracted from the residual structure, and the deep features are composed of a global average pooling layer, two full-connection layers, a Mish activation layer and a Sigmoid function, wherein the stacking sequence is a global average pooling layer, a full-connection layer, a Mish activation layer, a full-connection layer and a Sigmoid function. The unit firstly performs feature compression on the deep features along the space dimension through a global average pooling layer, compresses each two-dimensional feature channel into a real number, and the output dimension is matched with the input feature channel number; generating respective weights for each characteristic channel, namely correlation among the characteristic channels, by using two full connection layers, a Mish activation layer and a sigmoid function; and finally, weighting all the calculated weights to the previous features channel by channel through multiplication to finish recalibration of the original features in the channel dimension.
The feature extraction branch further extracts the deep features extracted by the residual network, and the feature extraction unit consists of three two-dimensional convolution layers and two ELU activation layers, wherein the three-dimensional convolution layers with one convolution kernel of 3*3, the ELU activation layer, the two-dimensional convolution layers with two convolution kernels of 3*3 and the ELU activation layer are stacked.
S3.2 extracting global features in the lower branches
In the lower branch, the shallow feature is input to perform global feature extraction, as shown in fig. 4, the global feature extraction module performs feature dimension reduction on the shallow feature by using a two-dimensional convolution layer with a convolution kernel of 3*3 and two-dimensional global average pooling layers, and stacks the shallow feature according to the sequence of the global average pooling layer, the two-dimensional convolution layer with the convolution kernel of 3*3 and the global average pooling layer; extracting deeper features by using four two-dimensional convolution layers with convolution kernels of 3*3 and two global average pooling layers, and stacking three two-dimensional convolution layers with convolution kernels of 3*3, one global average pooling layer, one two-dimensional convolution layer with convolution kernels of 3*3 and one global average pooling layer in sequence for 3 times in series; finally, mapping the extracted deeper features to a new feature space by using five full-connection layers to obtain global features, wherein an ELU activation layer is accessed after each full-connection layer.
S4, performing feature fusion on the extracted time period features and the global features, and then inputting a Softmax function to obtain a classification result of the seismic data.
The invention extracts shallow features of seismic data through a convolutional neural network, and then adopts a parallel branch structure: in the upper branch, extracting deep features by using a residual error network, and acquiring time period features with weights for the deep features by using a time-based attention mechanism; in the lower branch, global features are extracted from shallow features using convolutional neural networks. The fusion judgment is carried out on the extracted depth features, so that the attention of the deep learning model to the global attribute of the seismic data can be improved, the features of the original data are fully utilized, the processing capacity of the seismic data features is optimized, and the classification precision of the two types of seismic data is improved.
The above embodiments are merely illustrative of the preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, but various modifications and improvements made by those skilled in the art to which the present invention pertains are made without departing from the spirit of the present invention, and all modifications and improvements fall within the scope of the present invention as defined in the appended claims.
Claims (9)
1. A method for classifying seismic data based on a time-based attention mechanism, comprising:
collecting seismic data, and processing the seismic data to generate a spectrogram, wherein the seismic data is time sequence data;
extracting features of the spectrogram to obtain time period features and global features with weights;
and obtaining a classification result of the seismic data according to the weighted time period characteristics and the weighted global characteristics.
2. The time-attention-mechanism-based seismic data classification method of claim 1, wherein processing the seismic data comprises:
and converting the time sequence data into frequency domain data by adopting frequency domain transformation, and storing the frequency domain data into a spectrogram, wherein the frequency domain transformation comprises fast Fourier transformation and short-time Fourier transformation.
3. The time-attention-mechanism-based seismic data classification method of claim 1, wherein acquiring the weighted time period features and global features comprises:
and extracting the shallow features of the spectrogram, and respectively extracting the time period features and the global features of the spectrogram by adopting a parallel branch structure for the shallow features of the spectrogram.
4. A method of classifying seismic data based on a time-awareness mechanism according to claim 3, wherein extracting shallow features of the spectrogram comprises:
the method comprises the steps of constructing a shallow feature extraction module based on a convolutional neural network, inputting a spectrogram into the shallow feature extraction module to extract shallow features of the spectrogram, wherein the shallow feature extraction module comprises two-dimensional convolutional layers, a batch normalization layer and an ELU activation layer, the two-dimensional convolutional layers are connected in a residual form, and the batch normalization layer and the ELU activation layer are connected behind the two-dimensional convolutional layers.
5. A time-attention-mechanism-based seismic data classification method according to claim 3, wherein the parallel branching structure includes an upper branching portion and a lower branching portion;
the upper branch part is used for extracting deep features of the spectrogram through shallow features of the spectrogram and extracting time period features of the spectrogram based on the extracted deep features;
the lower branch part is used for extracting global features of the spectrogram through shallow features of the spectrogram.
6. The method of time-attention-based seismic data classification of claim 5, wherein extracting deep features of said spectrogram comprises:
the deep feature extraction method comprises the steps of constructing a deep feature extraction module based on a residual network, inputting shallow features of a spectrogram into the deep feature extraction module to extract deep features of the spectrogram, wherein the deep feature extraction module comprises a residual block, the residual block comprises a direct mapping unit and a residual unit, summation is carried out after output results of the direct mapping unit and the residual unit are obtained, the direct mapping unit comprises a two-dimensional convolution layer, a batch normalization layer, an ELU activation layer and a Dropout layer, and the residual unit comprises the ELU activation layer and a maximum pooling layer.
7. The time-attention-mechanism-based seismic data classification method of claim 5, wherein extracting a temporal feature of the spectrogram comprises:
a time-based attention mechanism is used for constructing a time period feature extraction module, deep features of the spectrogram are input into the time period feature extraction module to extract time period features of the spectrogram, wherein the time period feature extraction module comprises a weight calculation unit and a feature extraction unit, and feature stitching is performed after a result of parallel calculation output of the weight calculation unit and the feature extraction unit is obtained;
the weight calculation unit comprises a global average pooling layer, a full connection layer, a Mish activation layer and a Sigmoid function, performs feature channel compression on the deep features based on the global average pooling layer, generates weights of the feature channels through the full connection layer, the Mish activation layer and the Sigmoid function, and weights the weights to the deep features; the feature extraction unit comprises a two-dimensional convolution layer and an ELU activation layer, and further feature extraction is carried out on the deep features.
8. The time-attention-mechanism-based seismic data classification method of claim 5, wherein extracting global features of the spectrogram comprises:
the method comprises the steps of constructing a global feature extraction module based on a convolutional neural network, inputting shallow features of a spectrogram into the global feature extraction module to extract global features of the spectrogram, wherein the global feature extraction module comprises a two-dimensional convolutional layer, a global average pooling layer, a full-connection layer and an ELU activation layer, performing feature degradation and further deep feature extraction on the shallow features through the two-dimensional convolutional layer and the global average pooling layer, mapping the extracted further deep features by utilizing the full-connection layer to obtain the global features, and connecting the ELU activation layer after the full-connection layer.
9. A method of classifying seismic data based on a time-awareness mechanism according to claim 3, wherein obtaining the classification result of the seismic data comprises:
and carrying out feature fusion on the extracted time period features and the global features of the spectrogram, and inputting a Softmax function to obtain a classification result of the seismic data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310272897.2A CN116068651A (en) | 2023-03-17 | 2023-03-17 | Seismic data classification method based on time attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310272897.2A CN116068651A (en) | 2023-03-17 | 2023-03-17 | Seismic data classification method based on time attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116068651A true CN116068651A (en) | 2023-05-05 |
Family
ID=86171700
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310272897.2A Pending CN116068651A (en) | 2023-03-17 | 2023-03-17 | Seismic data classification method based on time attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116068651A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116660992A (en) * | 2023-06-05 | 2023-08-29 | 北京石油化工学院 | Seismic signal processing method based on multi-feature fusion |
-
2023
- 2023-03-17 CN CN202310272897.2A patent/CN116068651A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116660992A (en) * | 2023-06-05 | 2023-08-29 | 北京石油化工学院 | Seismic signal processing method based on multi-feature fusion |
| CN116660992B (en) * | 2023-06-05 | 2024-03-05 | 北京石油化工学院 | Seismic signal processing method based on multi-feature fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103487832B (en) | Supervision waveform classification is had in a kind of 3-D seismics signal | |
| CN111562612B (en) | Deep learning microseismic event identification method and system based on attention mechanism | |
| CN107678059B (en) | A kind of method, apparatus and system of reservoir gas-bearing identification | |
| Zhang et al. | Extracting dispersion curves from ambient noise correlations using deep learning | |
| Ma et al. | Automated arrival-time picking using a pixel-level network | |
| Dou et al. | Attention-based 3-D seismic fault segmentation training by a few 2-D slice labels | |
| CN116068651A (en) | Seismic data classification method based on time attention mechanism | |
| Yang et al. | Denoising of distributed acoustic sensing data using supervised deep learning | |
| Yang et al. | Automatic reconstruction method of 3D geological models based on deep convolutional generative adversarial networks | |
| CN112231974B (en) | Deep learning-based method and system for recovering seismic wave field characteristics of rock breaking seismic source of TBM (Tunnel boring machine) | |
| Chang et al. | Generative adversarial networks for seismic data interpolation | |
| CN114492521A (en) | Intelligent lithology while drilling identification method and system based on acoustic vibration signals | |
| CN112464721A (en) | Automatic microseism event identification method and device | |
| CN105277979B (en) | The optimization method and device of a kind of seismic properties | |
| Vinard et al. | Localizing microseismic events on field data using a U-Net-based convolutional neural network trained on synthetic data | |
| Li et al. | Distributed acoustic sensing vertical seismic profile data denoising based on multistage denoising network | |
| CN115436993A (en) | Seismic signal P wave detection method based on PPNet network | |
| CN111894563A (en) | Method and system for determining classification of fractured reservoir in buried hill section | |
| Zhong et al. | Multiscale residual pyramid network for seismic background noise attenuation | |
| CN115220094B (en) | Microseism event detection method based on BilSTM and attention mechanism | |
| CN117251674A (en) | Seismic phase classification method, device, electronic equipment and medium | |
| CN112329804A (en) | Naive Bayes lithofacies classification integrated learning method and device based on feature randomness | |
| CN116577821A (en) | Time-lapse earthquake prediction method and system based on neural network | |
| Guo et al. | Seismic random noise removal based on a multiscale convolution and densely connected network for noise level evaluation | |
| Zhang et al. | Application of u-net for the recognition of regional features in geophysical inversion results |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |