Micro-seismic seismic facies pickup method and device based on multi-channel characteristics

Abstract

The invention provides a microseismic seismic facies pickup method and device based on multichannel features, and relates to the field of seismic exploration, and the method comprises the steps: obtaining microseismic data, and carrying out the preprocessing of the microseismic data, and obtaining a chromaticity image and an artificial pickup data black and white mask image; constructing a seismic facies pickup model by introducing a coordinate channel and an attention mechanism; training a seismic phase pickup model through the chromaticity image and the manually picked data black and white mask image; obtaining micro-seismic data to be measured in a single micro-seismic event and performing preprocessing; and inputting the preprocessed micro-seismic data to be measured into the trained seismic phase pickup model to obtain seismic phase pickup data with spatial-temporal characteristics. Through joint analysis of multi-channel data, the method can overcome the limitation of single-channel feature learning of a traditional machine learning pickup method, and significantly improves the precision of microseism seismic phase pickup.

Description

Technical Field

[0001] The present application relates to the field of seismic exploration, and in particular to a method and device for picking up microseismic phases based on multi-channel features. Background Art

[0002] Microseismic monitoring is an important part of the field of seismic exploration, and phase picking is a key step in seismic signal processing. For microseismic monitoring caused by industrial activities such as hydraulic fracturing, the accuracy of phase picking directly affects the location of the earthquake source and the analysis of seismic activity. Traditional phase picking methods, such as the STA / LTA (short-time window average to long-time window average ratio) algorithm, often cannot accurately identify weak first-arrival wave characteristics under high noise and complex geological conditions. And they mainly focus on the processing of single-channel data, ignoring the correlation and characteristics between multi-channel signals.

[0003] Existing deep learning methods (such as PhaseNet) mainly deduce seismic phase picking based on the temporal characteristics of single-channel data, and fail to fully utilize the multi-channel characteristic information of regular detector arrays, resulting in poor picking effects in complex environments. Summary of the invention

[0004] The purpose of the present invention is to provide a microseismic phase picking method and device based on multi-channel features in order to solve the problem that the existing deep learning methods mainly derive seismic phase picking based on the time characteristics of single-channel data, fail to fully utilize the multi-channel characteristic information of regular detector arrays, resulting in poor picking effect in complex environments.

[0005] The above-mentioned purpose of the present application is achieved through the following technical solutions: S1: Acquire microseismic data and preprocess them to obtain chromaticity images and black and white mask images of artificially picked data; S2: Construct a seismic phase picking model by introducing coordinate channels and attention mechanism; S3: The phase picking model is trained using chromaticity images and black and white mask images of artificially picked data; S4: Acquire the microseismic data to be measured in a single microseismic event and preprocess them; input the preprocessed microseismic data to be measured into the trained phase picking model to obtain phase picking data with time and space characteristics.

[0006] Optionally, step S1 includes: The microseismic data is normalized to obtain a chromaticity image; Through the method of artificial picking and combining with the chromaticity image, a black and white mask image of artificially picked data is generated.

[0007] Optionally, step S2 includes: The seismic phase picking model includes: a CoordConv coordinate convolution layer, an encoder and a decoder; The CoordConv coordinate convolution layer is connected to the encoder; the encoder is connected to the decoder; The encoder consists of four layers, each of which includes: a maximum pooling layer, a first convolutional layer, a first LeakyRelu activation function, a second convolutional layer, a second LeakyRelu activation function and a residual block; The maximum pooling layer, the first convolutional layer, the first LeakyRelu activation function, the second convolutional layer, and the second LeakyRelu activation function are connected in sequence; The decoder includes: a transposed convolution layer, an upsampling layer, a maximum pooling layer, a residual block, a Sigmoid activation function, and an attention module.

[0008] Optionally, the CoordConv coordinate convolution layer specifically includes: By introducing the coordinate channel, the CoordConv coordinate convolution layer is constructed; For a given input feature map, generate two coordinate maps of the same size, one for the x-coordinate and the other for the y-coordinate; Normalize the coordinate values ​​to the range of [-1, 1], and the coordinate value calculation formula is:

[0009]

[0010] in, and The pixels are respectively and height The coordinate index on .

[0011] Optionally, step S4 includes: S41: inputting the chromaticity image of the microseismic data to be measured into the trained seismic phase picking model to generate a corresponding black and white mask image; S42: Analyze the pixel positions of the boundary edge lines between the black area and the white area in the black-and-white mask image to obtain the first arrival picking data in a single microseismic event: S43: Repeat steps S41 to S42 to obtain the first arrival picking data in three directions in a single microseismic event and perform an average operation to obtain seismic phase picking data with time and space characteristics.

[0012] Optionally, step S42 includes: Design a post-processing algorithm; the post-processing algorithm is used to ignore black noise in the black and white mask image and identify the longest white chain of pixels in each column of the chromaticity image column by column; Through the post-processing algorithm, the longest white chain of pixels in each column of the black-and-white mask image is identified column by column; The starting position of the longest white chain in each column is picked as the first arrival time to obtain the first arrival picking data.

[0013] An electronic device includes a processor, a memory, a communication interface and graphic computing hardware, wherein the memory is used to store instructions, the communication interface is used to communicate with other devices, the graphic computing hardware is used to calculate images, and the processor is used to execute the instructions stored in the memory so that the electronic device performs a microseismic phase picking method based on multi-channel features.

[0014] A computer-readable storage medium stores instructions, and when the instructions are executed, a microseismic phase picking method based on multi-channel features is executed.

[0015] The beneficial effects of the technical solution provided by this application are: 1. The spatial coordinate information is introduced through the coordinate convolution (CoordConv) layer, so that the model can simultaneously utilize the spatial and temporal characteristics of the signal to improve the accuracy of phase picking. By introducing the attention mechanism, it is possible to effectively focus on the key features of the microseismic signal and effectively capture the multi-channel related information and regular morphological information of the microseismic first-arrival waveform, which has higher accuracy and robustness than traditional machine learning methods. The present invention uses multi-channel signal fusion and the CoordConv convolution layer to learn the spatial and temporal characteristics of microseismic signals, which greatly improves the accuracy of phase picking and can overcome the limitations of traditional machine learning picking methods that learn single-channel features.

[0016] 2. Through the designed post-processing algorithm, the black noise in the black and white mask image is ignored, and the longest white chain of pixels in each column of the chromaticity image is identified column by column to improve the picking accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 is a step diagram in an embodiment of the present application; Figure 2 is a schematic diagram of the original seismic channel of data preprocessing in the embodiment of the present application; Figure 3 is the chromaticity diagram in the embodiment of the present application; Figure 4 is a black and white mask image in the embodiment of the present application; Figure 5 It is a schematic diagram of the model structure in the embodiment of the present application; Figure 6 Schematic diagram of the coordinate convolution layer in the embodiment of the present application; Figure 7 is a schematic diagram of an attention module in an embodiment of the present application; Figure 8 It is a schematic diagram of the optimization algorithm in the embodiment of the present application; Fig. 9 is a schematic diagram of the structure of an electronic device in an embodiment of the present application; Fig.10 It is a data result diagram in the examples of this application. DETAILED DESCRIPTION

[0018] In order to have a clearer understanding of the technical features, purposes and effects of the present application, the specific implementation methods of the present application are now described in detail with reference to the accompanying drawings.

[0019] An embodiment of the present application provides a microseismic phase picking method based on multi-channel features.

[0020] Please refer to Figure 1 , Figure 1 : is a step diagram of a microseismic phase picking method based on multi-channel features in an embodiment of the present application, comprising: S1: Acquire microseismic data and preprocess them to obtain chromaticity images and black and white mask images of artificially picked data; Step S1 includes: The microseismic data is normalized to obtain a chromaticity image; Through the method of artificial picking and combining with the chromaticity image, a black and white mask image of artificially picked data is generated.

[0021] Specifically, in order to adapt to the seismic phase picking based on the deep learning model of the present application, the collected microseismic data is preprocessed. During the processing, the original microseismic data ( Figure 2 ) is converted to a chromaticity image normalized to the interval [-1, 1] ( Figure 3 ), and generate a black and white mask image of the corresponding artificially picked data, such as Figure 4 shown.

[0022] S2: Construct a seismic phase picking model by introducing coordinate channels and attention mechanism; Step S2 includes: The seismic phase picking model includes: a CoordConv coordinate convolution layer, an encoder and a decoder; The CoordConv coordinate convolution layer is connected to the encoder; the encoder is connected to the decoder; The encoder consists of four layers, each of which includes: a maximum pooling layer, a first convolutional layer, a first LeakyRelu activation function, a second convolutional layer, a second LeakyRelu activation function and a residual block; The maximum pooling layer, the first convolutional layer, the first LeakyRelu activation function, the second convolutional layer, and the second LeakyRelu activation function are connected in sequence; The decoder includes: a transposed convolution layer, an upsampling layer, a maximum pooling layer, a residual block, a Sigmoid activation function, and an attention module.

[0023] Specifically, the core model is the Attentive-Coord-UNet (ACU) model based on multi-channel features, and its structure is as follows: Figure 5 As shown in the figure. Before the phase picking model starts convolution, a CoordConv coordinate convolution layer is added in the initial stage. It combines the input feature map with its corresponding coordinate enhancement before applying the convolution operation. The core innovation of the coordinate convolution layer is that it introduces spatial coordinates as an additional input channel before the traditional convolution operation. This design enables the convolution network to perceive the position information of each pixel, thereby more effectively processing image boundaries and shape changes.

[0024] The model first goes through the encoder path for downsampling. The encoder part consists of four main layers, each of which includes a layer with two convolutional layers and a residual block. The MaxPooling layer is used for downsampling after the convolution to reduce the amount of computation and increase the receptive field. The Dropout layer at each downsampling step is used to reduce the risk of overfitting during the training process.

[0025] The convolution processing of the model mainly adopts the residual block design to deepen the network structure without introducing additional training difficulty. Through its unique jump connection method that allows gradients to propagate directly, it solves the common gradient vanishing problem in deep networks. At the bottom of the encoder, there is a central residual block to process the deepest and most compressed feature representation, extract highly abstract features, and lay the foundation for subsequent feature reconstruction and fine analysis.

[0026] The decoder part of the upsampling stage is designed to gradually reconstruct the detailed feature information of the image. This part increases the resolution of the feature map layer by layer through transposed convolution (ConvTranspose2d). After each upsampling step, attention blocks are used to enhance the decoded features and focus on key areas in the image. These attention blocks are implemented through a gated convolution layer switch, which gradually restores image details and structure by adjusting the feature strength passed to subsequent layers to better integrate features from the encoder. At the end of each decoding stage, the output of the decoder is concatenated with the attention-weighted encoder features along the feature channel dimension. During the upsampling process, the network not only restores the spatial dimension, but also refines the semantic information encoded by the deep network.

[0027] At the end of the entire network, a simple Sigmoid activation function is set to convert the multi-channel feature map into the final output channel. The output of the convolutional layer is mapped to the [0,1] interval to generate pixel-level prediction probabilities, which are suitable for binary classification problems. In this way, the model can output accurate and probabilistic segmentation results.

[0028] Traditional methods cannot effectively utilize the multi-channel correlation characteristics of borehole microseismic data. In order to learn the regular "step-like" multi-channel correlation characteristics of borehole microseismic observation data in images, and better utilize this special characteristic to enhance the accuracy of first arrival wave picking, this study introduces an attention module based on deep learning, with the following structure: Figure 7 As shown in the figure, the module can dynamically focus on the most informative part of the input features. By calculating the weight distribution between the input features, it explicitly simulates the relationship between the features, so that the model can better understand and utilize the spatial and temporal dependencies in the data. The attention module of this study uses a convolutional layer to perform feature transformation on the input microseismic data image, extract the necessary feature representations, calculate the interactions and dependencies between different features, and generate an attention map. This process involves weighted summation of the features, and the weights are dynamically determined by the relationship between the features themselves. Multiplying the attention map with the original features achieves recalibration of the original input features, strengthening useful features and suppressing unimportant features.

[0029] The CoordConv coordinate convolution layer specifically includes: By introducing the coordinate channel, the CoordConv coordinate convolution layer is constructed; For a given input feature map, generate two coordinate maps of the same size, one for the x-coordinate and the other for the y-coordinate; Normalize the coordinate values ​​to the range of [-1, 1], and the coordinate value calculation formula is:

[0030]

[0031] in, and The pixels are respectively and height The coordinate index on .

[0032] Specifically, Figure 6 As shown in the figure, the core of coordinate convolution is to add two additional coordinate channels on the basis of the original input features, which represent the x-coordinate and y-coordinate of each pixel respectively, allowing the network to directly access the absolute position information of each pixel, so that the network can explicitly learn the spatial position. By adding the AddCoords module before the first convolution layer, the coordinate channel is attached to the original input image, so that subsequent convolution operations are performed on this extended image, so that each convolution can utilize the spatial position information. Coordinate convolution enables the network to more easily learn the shape and size of objects, as well as their relative positions in the image, which can not only locate the feature areas in the image more accurately, but also more effectively restore the edge details of the object.

[0033] S3: The phase picking model is trained using chromaticity images and black and white mask images of artificially picked data; S4: Acquire the microseismic data to be measured in a single microseismic event and preprocess them; input the preprocessed microseismic data to be measured into the trained phase picking model to obtain phase picking data with time and space characteristics.

[0034] Specifically, the method realizes high-precision seismic phase picking by combining feature extraction and fusion of multi-channel signals of microseismic data. The method also combines hardware equipment to realize integrated operation of data processing, storage and playback viewing.

[0035] Step S4 includes: S41: inputting the chromaticity image of the microseismic data to be measured into the trained seismic phase picking model to generate a corresponding black and white mask image; S42: Analyze the pixel positions of the boundary edge lines between the black area and the white area in the black-and-white mask image to obtain the first arrival picking data in a single microseismic event: Step S42 includes: Design a post-processing algorithm; the post-processing algorithm is used to ignore black noise in the black and white mask image and identify the longest white chain of pixels in each column of the chromaticity image column by column; Through the post-processing algorithm, the longest white chain of pixels in each column of the black-and-white mask image is identified column by column; The starting position of the longest white chain in each column is picked as the first arrival time to obtain the first arrival picking data.

[0036] Specifically, in the recognition stage, in order to reduce the impact of noise on the picking results, a post-processing algorithm is designed such as Figure 8 As shown in the figure, the algorithm mainly looks for the longest continuous chain of white pixels in each column. Under the premise of tolerating a certain amount of black noise, this algorithm effectively identifies the longest white chain of pixels in each column of the image column by column, and picks the starting position of the chain as the first arrival time. Subsequently, the number of channels is checked and averaged to generate an accurate value for each data channel. According to the principle of microseismic signal processing, the first arrival picking data in three directions of a single microseismic event are averaged to finally obtain accurate picking data.

[0037] Specifically, all data after seismic phase picking is saved through SSD or HDD storage devices, supporting long-term data storage for later analysis. The seismic phase picking results in the storage medium can be retrospectively analyzed through the playback module, and users can select the seismic phase picking conditions of a specific time period and specific location for viewing and further research.

[0038] S43: Repeat steps S41 to S42 to obtain the first arrival picking data in three directions in a single microseismic event and perform an average operation to obtain seismic phase picking data with time and space characteristics.

[0039] like Fig.10 As shown, the present invention has carried out a large number of tests using actual data from Sichuan shale gas fields and geothermal projects. The experimental data show that the phase picking excellence rate of the present invention reaches 96.2901%, which is significantly better than the 82.3195% of the traditional algorithm.

[0040] The present application also discloses an electronic device. Fig. 9 FIG. 1 is a schematic diagram of the structure of an electronic device for implementing a method for improving the resolution of seismic data provided by an embodiment of the present invention. The electronic device may include a processor memory, a communication bus, and a communication interface.

[0041] The communication interface is used to realize the connection and communication between these components, including standard wired interface and wireless interface. The graphics computing hardware can use a conventional graphics processor.

[0042] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned microseismic phase picking method based on multi-channel features.

[0043] The above are only exemplary embodiments of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0044] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A microseismic phase picking method based on multi-channel features, characterized in that: The method comprises the following steps: S1: Acquire microseismic data and preprocess them to obtain chromaticity images and black and white mask images of artificially picked data; S2: Construct a seismic phase picking model by introducing coordinate channels and attention mechanism; S3: The phase picking model is trained using chromaticity images and black and white mask images of artificially picked data; S4: obtaining the microseismic data to be measured in a single microseismic event and performing preprocessing; The preprocessed microseismic data to be measured are input into the trained phase picking model to obtain phase picking data with temporal and spatial characteristics.

2. A microseismic phase picking method based on multi-channel features as claimed in claim 1, characterized in that: Step S1 includes: The microseismic data is normalized to obtain a chromaticity image; Through the method of artificial picking and combining with the chromaticity image, a black and white mask image of artificially picked data is generated.

3. The microseismic phase picking method based on multi-channel features as claimed in claim 1, characterized in that: Step S2 includes: The seismic phase picking model includes: a CoordConv coordinate convolution layer, an encoder and a decoder; The CoordConv coordinate convolution layer is connected to the encoder; the encoder is connected to the decoder; The encoder consists of four layers, each of which includes: a maximum pooling layer, a first convolutional layer, a first LeakyRelu activation function, a second convolutional layer, a second LeakyRelu activation function and a residual block; The maximum pooling layer, the first convolutional layer, the first LeakyRelu activation function, the second convolutional layer, and the second LeakyRelu activation function are connected in sequence; The decoder includes: a transposed convolution layer, an upsampling layer, a maximum pooling layer, a residual block, a Sigmoid activation function, and an attention module.

4. A microseismic phase picking method based on multi-channel features as claimed in claim 3, characterized in that: The CoordConv coordinate convolution layer specifically includes: By introducing the coordinate channel, the CoordConv coordinate convolution layer is constructed; For a given input feature map, generate two coordinate maps of the same size, one for the x-coordinate and the other for the y-coordinate; Normalize the coordinate values ​​to the range of [-1, 1], and the coordinate value calculation formula is: in, and The pixels are respectively and height The coordinate index on .

5. The microseismic phase picking method based on multi-channel features as claimed in claim 1, characterized in that: Step S4 includes: S41: inputting the chromaticity image of the microseismic data to be measured into the trained seismic phase picking model to generate a corresponding black and white mask image; S42: Analyze the pixel positions of the boundary edge lines between the black area and the white area in the black-and-white mask image to obtain the first arrival picking data in a single microseismic event: S43: Repeat steps S41 to S42 to obtain the first arrival picking data in three directions in a single microseismic event and perform an average operation to obtain seismic phase picking data with time and space characteristics.

6. A microseismic phase picking method based on multi-channel features as claimed in claim 5, characterized in that: Step S42 includes: Design a post-processing algorithm; the post-processing algorithm is used to ignore black noise in the black and white mask image and identify the longest white chain of pixels in each column of the chromaticity image column by column; Through the post-processing algorithm, the longest white chain of pixels in each column of the black-and-white mask image is identified column by column; The starting position of the longest white chain in each column is picked as the first arrival time to obtain the first arrival picking data.

7. An electronic device, characterized in that: include: A processor, a memory, a communication interface and graphic computing hardware, wherein the memory is used to store instructions, the communication interface is used to communicate with other devices, the graphic computing hardware is used to calculate images, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 6 is executed.

Images