Seismic data dimensionality reduction method and apparatus

Abstract

The application discloses a seismic data dimension reduction method and device, wherein the method comprises the following steps: constructing a neural network for seismic data dimension reduction; taking a seismic data set as an input sample point set and taking a dimension reduction result of the seismic data set as a final output sample point set; the neural network comprises a convolution network unit, a fully connected network unit and an output layer; a target function of the neural network is constructed according to a probability distribution of distance relationships of the input sample point set, output sample point sets of the convolution network unit and the fully connected network unit and the final output sample point set; the neural network is trained according to the target function of the neural network; and the trained neural network is used to perform dimension reduction processing on seismic data. The application can avoid the limitation of storage space and calculation amount and realize nonlinear seismic data dimension reduction.

Description

Technical Field

[0001] This invention relates to the field of geophysical oil and gas exploration technology, and in particular to a method and apparatus for seismic data dimensionality reduction. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] Seismic exploration is a crucial method for predicting and identifying oil and gas geological targets in the field of geophysical oil and gas exploration. Its main tasks include seismic data acquisition, processing, and interpretation. Seismic waves generated at the surface propagate to subsurface reflecting interfaces or geological targets, then reflect back to the surface and are received by geophones. The received seismic data thus contains important information about these subsurface reflecting interfaces or geological targets. In the early stages of exploration, drilling data is very limited; therefore, conducting unsupervised seismic data dimensionality reduction is of great significance for guiding subsequent oil and gas geological target identification.

[0004] Typical seismic data dimensionality reduction techniques include seismic waveform dimensionality reduction and seismic multi-attribute dimensionality reduction. A seismic waveform is composed of multiple sampled point values ​​and can be considered a high-dimensional data point. In seismic data processing and interpretation, it is often necessary to compare and analyze seismic waveforms at different spatial locations. Because the number of these seismic waveforms is enormous (often greater than 10), this process can be very complex. 6 Seismic waveform dimensionality reduction is a crucial method for comparing and analyzing seismic waveforms at different spatial locations, as they represent high-dimensional data points. Seismic attributes are measures of the geometric, kinematic, dynamic, and statistical characteristics of seismic data. They are components or subsets of seismic data reflecting different geological features (information), and are seismic characteristic quantities that characterize and describe geological information such as stratigraphic structure, lithology, and physical properties. Seismic attribute analysis is one of the important methods in the field of seismic oil and gas exploration for analyzing stratigraphic structural features and predicting and identifying oil and gas geological targets. An important task of seismic attribute analysis is to comprehensively analyze multiple seismic attributes and form comprehensive maps that are conducive to predicting and identifying oil and gas geological targets.

[0005] Existing nonlinear dimensionality reduction methods (manifold learning, t-distributed random neighborhood embedding) are optimized based on the probability distribution matrix of the distance relationships of all sample point sets. The size of the matrix is ​​directly related to the number of sample point sets. When the amount of seismic data is very large, it is difficult to directly use it for seismic data dimensionality reduction due to limitations in storage space and computational cost. Summary of the Invention

[0006] This invention provides a method for dimensionality reduction of seismic data to avoid limitations in storage space and computational load, thereby achieving dimensionality reduction of nonlinear seismic data. The method includes:

[0007] A neural network for dimensionality reduction of seismic data is constructed, taking the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer;

[0008] The objective function of the neural network is constructed based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set.

[0009] Train the neural network according to the objective function of the neural network;

[0010] The trained neural network is used to perform dimensionality reduction on the seismic data.

[0011] This invention also provides a seismic data dimensionality reduction device to avoid limitations in storage space and computational load, thereby achieving dimensionality reduction of nonlinear seismic data. The device includes:

[0012] A neural network construction module is used to construct a neural network for dimensionality reduction of seismic data. The seismic dataset is used as the input sample point set, and the dimensionality reduction result of the seismic dataset is used as the final output sample point set. The neural network includes: convolutional network units, fully connected network units, and an output layer.

[0013] The objective function construction module is used to construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set.

[0014] A training module is used to train the neural network according to the objective function of the neural network;

[0015] The dimensionality reduction module is used to perform dimensionality reduction processing on the seismic data using the trained neural network.

[0016] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described seismic data dimensionality reduction method.

[0017] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described seismic data dimensionality reduction method.

[0018] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described seismic data dimensionality reduction method.

[0019] In this embodiment of the invention, a neural network for seismic data dimensionality reduction is constructed. The seismic dataset is used as the input sample point set, and the dimensionality reduction result of the seismic dataset is used as the final output sample point set. The neural network includes convolutional network units, fully connected network units, and an output layer. An objective function of the neural network is constructed based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set. The neural network is trained based on the objective function. The trained neural network is then used to perform dimensionality reduction on the seismic data. Compared with existing technical solutions, constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set, and training the neural network based on the objective function, leverages the memory and feature extraction capabilities of the convolutional network units. This eliminates the need to train the neural network using the probability distribution of the distance relationships of all sample point sets, avoiding limitations in storage space and computational load, and enabling dimensionality reduction of nonlinear seismic data. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0021] Figure 1 This is a flowchart illustrating the seismic data dimensionality reduction method provided in this embodiment of the invention;

[0022] Figure 2 This is a specific example diagram of a neural network for seismic data dimensionality reduction provided in an embodiment of the present invention;

[0023] Figure 3 This is a specific example diagram illustrating the random selection of a predetermined number of seismic data points from a seismic dataset according to Embodiment 1 of the present invention.

[0024] Figure 4 This is a specific example diagram of drawing a dimensionality-reduced plane diagram of seismic data based on the dimensionality reduction value according to Embodiment 1 of the present invention;

[0025] Figure 5 This is the result of globally sorting seismic waveforms in the entire seismic waveform dataset based on the magnitude of the dimensionality reduction value, as provided in Embodiment 1 of the present invention, and it also displays... Figure 4 The correspondence between medium grayscale color scales and seismic waveforms;

[0026] Figure 6 This is a specific example diagram of clustering seismic data points according to Embodiment 1 of the present invention;

[0027] Figure 7 This is the result of globally sorting seismic waveforms in the entire seismic waveform dataset based on the magnitude of the dimensionality reduction value, as provided in Embodiment 1 of the present invention, and it also displays... Figure 6 The correspondence between black, gray, and white markers and earthquake waveforms;

[0028] Figure 8 This is a coherent seismic property planar diagram related to underground faults and cracks, provided according to Embodiment 2 of the present invention.

[0029] Figure 9 This is a plan view of the seismic properties of a curvature body related to underground faults and cracks, provided according to Embodiment 2 of the present invention.

[0030] Figure 10 This is an AVAZ seismic attribute planar map related to underground faults and cracks provided according to Embodiment 2 of the present invention;

[0031] Figure 11 This refers to the seismic multi-attribute clustering result obtained by combining seismic multi-attribute dimensionality reduction results, known geological information, and drilling information, as provided in Embodiment 2 of the present invention.

[0032] Figure 12 This is a schematic diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention;

[0033] Figure 13 This is a specific example diagram of the seismic data dimensionality reduction device provided in this embodiment of the invention;

[0034] Figure 14 This is a specific example diagram of the seismic data dimensionality reduction device provided in this embodiment of the invention;

[0035] Figure 15 This is a specific example diagram of the seismic data dimensionality reduction device provided in this embodiment of the invention;

[0036] Figure 16 This is a specific example diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0038] This invention provides a method for dimensionality reduction of seismic data. Figure 1This is a flowchart illustrating the seismic data dimensionality reduction method provided in this embodiment of the invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0039] Step 101: Construct a neural network for dimensionality reduction of seismic data, using the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer;

[0040] Step 102: Construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set;

[0041] Step 103: Train the neural network according to the objective function of the neural network;

[0042] Step 104: Use the trained neural network to perform dimensionality reduction on the seismic data.

[0043] Depend on Figure 1 As shown in the flowchart, in this embodiment of the invention, a neural network for seismic data dimensionality reduction is constructed, using the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set. The neural network includes convolutional network units, fully connected network units, and an output layer. Based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set, the objective function of the neural network is constructed. The neural network is trained according to the objective function. The trained neural network is then used to perform dimensionality reduction on the seismic data. Compared with existing technical solutions, constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set, and training the neural network based on the objective function, leverages the memory and feature extraction capabilities of the convolutional network units. This eliminates the need to train the neural network using the probability distribution of the distance relationships of all sample point sets, avoiding limitations in storage space and computational load, and enabling dimensionality reduction of nonlinear seismic data.

[0044] The execution process of steps 101 to 104 above will be described in detail below.

[0045] For step 101 above, a neural network for dimensionality reduction of seismic data is constructed, with the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer.

[0046] The input sample point set of the neural network described in step 101 is an earthquake dataset, that is, a high-dimensional earthquake dataset, hereinafter referred to as {x}. i |i=1,2,3,…,N} represents, where x i A sample point, or seismic data point, is a column vector, equivalent to a point in a high-dimensional space. The length of the vector is equal to the dimension of the high-dimensional space, hereinafter referred to as D. x In this context, i represents the sample point number in the dataset, and N represents the number of sample points in the dataset, which is also the batch size of the neural network. Generally, N is taken as an integer between 1000 and 5000.

[0047] The final output sample point set of the neural network described in step 101 is the dimensionality reduction result of the seismic dataset, that is, the dimensionality reduction result of the corresponding high-dimensional seismic dataset, as follows {y i |i=1,2,3,…,N} represents, where y i The dimensionality reduction result for a single sample point, i.e., a single seismic data point, is a column vector, equivalent to a point in the lower-dimensional space. The length of the vector is equal to the dimension of the lower-dimensional space, hereinafter referred to as D. y This means that, generally speaking, we take D. y =1 or D y =2, where i is the sample point number in the dataset, and N has the same meaning and value as above.

[0048] In one embodiment, the neural network includes: multiple convolutional network units, multiple fully connected network units, and an output layer; each convolutional network unit includes: a convolutional layer, a max-pooling layer, and a ReLU activation function; each fully connected network unit includes: a fully connected layer and a Sigmoid activation function; the output layer includes: a fully connected layer. The batch size of the neural network is N, and the meaning and value of N are the same as above.

[0049] Specifically, the number of convolutional network units can be, for example, 3 to 10. Each convolutional network unit can include, for example, a convolutional layer, a max pooling layer, and a ReLU activation function. The core parameters of each convolutional layer can include, for example, the kernel parameters, the kernel stride, and the zero-padding method. Among them, the kernel parameters include the kernel length, the number of input channels, and the number of output channels. Generally, the kernel length is set to 3, the number of input channels is equal to the number of output channels of the previous convolutional network unit, and the number of output channels is set to twice the number of input channels. The kernel stride can be set to, for example, 1. The zero-padding method can be set to, for example, no zero padding. The core parameters of each max pooling layer can include, for example, the pooling window length and the pooling window stride. Generally, the pooling window length is set to 2, and the pooling window stride is equal to the current pooling window length.

[0050] Specifically, the number of fully connected network units can be, for example, 3 to 10. Each fully connected network unit may include, for example, a fully connected layer and a sigmoid activation function. The core parameters of each fully connected layer may include, for example, the number of input nodes and the number of output nodes. Generally, the number of input nodes is equal to the number of output nodes of the previous fully connected network unit, and the number of output nodes is set to 0.5 or 0.25 times the number of input nodes, and if it is not an integer, it needs to be rounded down to the nearest integer.

[0051] Specifically, the output layer may include, for example, a fully connected layer. The output of the output layer may, for example, be {y} i |i=1,2,3,…,N}, the number of output nodes can be set to D, for example. y D y The meaning is the same as above.

[0052] Specifically, the first convolutional network unit among the multiple convolutional network units described above takes a high-dimensional seismic dataset {x} as its input. i |i=1,2,3,…,N}, where the number of input channels in the convolution kernel parameters is set to 1.

[0053] In one embodiment, a flattening layer is added after the last convolutional network unit in a plurality of convolutional network units. The flattening layer is used to concatenate vectors from different channels end to end to form a vector and provide input to the first fully connected network unit.

[0054] For step 102 above, the objective function of the neural network is constructed based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set.

[0055] The input sample point set mentioned in step 102 is {x} i |i=1,2,3,…,N}, the symbols have the same meaning as above;

[0056] The final output sample point set mentioned in step 102 is {y} i |i=1,2,3,…,N}, which has the same meaning as above;

[0057] The output sample point set of the convolutional network unit and the fully connected network unit mentioned in step 102 refers to the output sample point set of the multiple convolutional network units and the multiple fully connected network units mentioned in step 101, hereinafter referred to as... Let represent the set of output sample points for each convolutional network unit and each fully connected network unit, where Let a sample point be a column vector, and let the length of the vector be . but Equivalent to In the dimensional space, i is the sample point number in the dataset, N has the same meaning and value as above, and η is the unified number of the convolutional network unit and the fully connected network unit, η = 1, 2, 3, ..., L, where L is the number of network units, that is, the total number of multiple convolutional network units and multiple fully connected network units in the neural network.

[0058] Specifically, the above-mentioned If it is the output sample point set of a convolutional network unit, it means This is the output of a convolutional network unit after passing through a flattening layer. The function of the flattening layer is to connect the vectors from different channels end to end to form a single vector.

[0059] In one embodiment, before constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the method may, for example, include: determining the probability distribution of the distance relationships between the input sample point set according to the following formula:

[0060]

[0061] Where, x i x j x s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point x under investigation i When, sample point x j The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to a fixed constant value. For example, α i It can indicate that The information entropy is equal to a fixed constant H. For a specific dimensionality reduction problem, the value of H can range from 1.5 to 4.0.

[0062] In one embodiment, before constructing the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the method may, for example, include: determining the probability distribution of the distance relationship between the output sample point sets of the convolutional network unit and the fully connected network unit according to the following formula:

[0063]

[0064] in, η represents the sample points in the dataset; the subscripts i, j, and s are the sample point numbers in the dataset, and the sample point numbers traverse the range of 1, 2, 3, ..., N, where N is the number of sample points in the dataset; η is the number of multiple convolutional network units and multiple fully connected network units, η = 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; β η For the weighting coefficients, take Indicates the sample points under investigation At that time, sample points The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to the value of a fixed constant.

[0065] In one embodiment, before constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the method may, for example, include: determining the probability distribution of the distance relationships between the final output sample point set according to the following formula:

[0066]

[0067] Among them, y i y j y s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; the symbol |||| denotes the L2 norm of a vector.

[0068] In one embodiment, the objective function of the neural network is constructed according to the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, as follows:

[0069]

[0070] in, Represents sample point x i With x j The distance between them; Indicates the sample point x under investigation i When, sample point x j The conditional probability; Indicates the sample point x under investigation j When, sample point x i The conditional probability; Indicates the sample point x under investigation t When, sample point x s The conditional probability; Indicates the sample point x under investigation s When, sample point x t The conditional probability; Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Represents sample point y i With y j The distance between them; Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; Indicates the sample point y under investigation s At that time, sample point y t The conditional probability; subscripts i, j, s, t are the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset; the range of the subscript η is 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; the objective function J(k) and the weighting coefficient λ η (k) represents a function that trains the neural network for k iterations. Generally, we take... Where k = 1, 2, 3, ..., K, and K is the total number of times the neural network is trained.

[0071] In this embodiment of the invention, the objective function of the neural network constructed by the seismic data dimensionality reduction method is a function of the number of training iterations of the neural network. In the early stage of neural network training, the objective function requires that the output results of the first few layers of the neural network maintain similar information to the input data. As the number of training iterations increases, the objective function gradually requires that the output results of deeper layers of the neural network maintain similar information to the input data. Until the training is nearing its end, the objective function requires that the final output results of the neural network maintain similar information to the input data. The design of the objective function reflects the idea of ​​layer-by-layer dimensionality reduction and step-by-step training, ensuring the stability of the training process and the reliability of the training results.

[0072] In step 103 above, the neural network is trained according to the objective function of the neural network.

[0073] In one embodiment, training the neural network according to its objective function may include, for example, adjusting the neural network parameters using a stochastic gradient descent method based on the objective function of the neural network to obtain the trained neural network.

[0074] In one embodiment, the parameters of the neural network are adjusted using a stochastic gradient descent method based on the objective function of the neural network to obtain a trained neural network. This may include, for example, the following:

[0075] Repeat the following steps until the objective function value of the neural network converges to its minimum:

[0076] A predetermined number of earthquake data points are randomly selected from the earthquake dataset; the predetermined number of randomly selected earthquake data points are input into the neural network, and the dimensionality reduction result of the corresponding earthquake data points is output; the objective function value of the neural network is calculated based on the earthquake data points and the corresponding dimensionality reduction result.

[0077] Specifically, let the neural network be trained k = 1, 2, 3, ..., K times. Starting from k = 1, N earthquake data points {x} are randomly selected from the earthquake dataset each time. i Given a set of data points (i = 1, 2, 3, ..., N), input the data into the neural network and output the dimensionality reduction result of the corresponding seismic data points. Based on the seismic data points and the corresponding dimensionality reduction result, calculate the objective function value of the neural network, and adjust the neural network parameters according to the objective function J(k), completing one neural network parameter update. Repeat the above process until k = K. Generally, K = 1000~5000 is sufficient to make the objective function J(k) converge to its minimum value.

[0078] In step 104 above, the trained neural network is used to perform dimensionality reduction on the seismic data.

[0079] Specifically, the trained neural network is used to sequentially input the earthquake data point vectors from the earthquake dataset into the trained neural network to perform dimensionality reduction processing on the earthquake data, obtaining the dimensionality reduction result. If the earthquake dataset has lateral spatial distribution characteristics, the dimensionality reduction result can be used to draw a dimensionality-reduced planar map of the earthquake data.

[0080] In one embodiment, after performing dimensionality reduction on the seismic data using the trained neural network, the process may further include: obtaining the dimensionality reduction values ​​of the seismic data; sorting the seismic dataset according to the magnitude of the dimensionality reduction values ​​to obtain sorted seismic data points; determining multiple clustering thresholds based on the sorted seismic data points and combining geological and drilling information; and clustering the seismic data points whose dimensionality reduction values ​​fall between two adjacent clustering thresholds. The seismic data dimensionality reduction method provided in this embodiment reduces the seismic data to one dimension and then clusters it, overcoming the technical problems of traditional seismic data clustering methods, such as large workload, high uncertainty, and poor flexibility due to "multiple trial calculations."

[0081] Specifically, for example, D y When = 1, the earthquake data point vectors in the earthquake dataset are sequentially input into the trained neural network to perform dimensionality reduction processing on the earthquake data and obtain the dimensionality reduction value of the earthquake data; the earthquake dataset is globally sorted according to the magnitude of the dimensionality reduction value, and the global sorting result can ensure that similar earthquake data points have similar dimensionality reduction values; based on the sorted earthquake data points, combined with known geological information and drilling information, a clustering threshold is determined; earthquake data points whose dimensionality reduction value belongs to two adjacent clustering thresholds are clustered to obtain the earthquake data clustering result.

[0082] The following specific embodiments illustrate the application of the seismic data dimensionality reduction method of the present invention.

[0083] Example 1:

[0084] This invention provides an application example of a seismic data dimensionality reduction method in the dimensionality reduction of seismic waveforms in actual work areas. The core of this method is to achieve nonlinear seismic data dimensionality reduction by leveraging the memory and feature extraction capabilities of convolutional neural networks. This solves the problem that some existing nonlinear dimensionality reduction methods cannot be applied to seismic data dimensionality reduction due to limitations in storage space and computational load.

[0085] like Figure 1 As shown, the seismic data dimensionality reduction method includes the following steps:

[0086] Step 101: Construct a neural network for dimensionality reduction of seismic data, using the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer;

[0087] Step 102: Construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set;

[0088] Step 103: Train the neural network according to the objective function of the neural network;

[0089] Step 104: Use the trained neural network to perform dimensionality reduction on the seismic data.

[0090] In step 101, the input sample point set of the neural network is an earthquake dataset, i.e., a high-dimensional earthquake waveform dataset, hereinafter referred to as {x}. i |i=1,2,3,…,N} represents, where x i A sample point, or seismic waveform, is a column vector, equivalent to a point in a high-dimensional space. The length of the vector is equal to the dimension of the high-dimensional space, hereinafter referred to as D. x Let i represent the sample point number in the dataset, and N represent the number of sample points in the dataset, i.e., the batch size of the neural network. Generally, N is an integer between 1000 and 5000. In this embodiment of the invention, the number of sampling points for each seismic waveform can be, for example, 60, i.e., D. x =60, and take N=1000.

[0091] The final output sample point set of the neural network described in step 101 is the dimensionality reduction result of the seismic dataset, that is, the dimensionality reduction result of the corresponding high-dimensional seismic waveform dataset. Hereinafter, {y} i |i=1,2,3,…,N} represents, where y i For a sample point, i.e., the dimensionality reduction result of a seismic waveform, it is a column vector, equivalent to a point in the low-dimensional space. The length of the vector is the dimension of the low-dimensional space, hereinafter referred to as D. y This means that, generally speaking, we take D. y =1 or D y =2, where i is the sample point number in the dataset, and N has the same meaning and value as above. In this embodiment of the invention, D is taken as... y =1.

[0092] Figure 2 This is a specific example diagram of a neural network for seismic data dimensionality reduction provided in an embodiment of the present invention, as shown below. Figure 2As shown, the batch size of the neural network is N, with the meaning and value of N as described above. The neural network includes: multiple convolutional network units, multiple fully connected network units, and an output layer; the number of convolutional network units can be, for example, 3 to 10, and each convolutional network unit includes: a convolutional layer, a max-pooling layer, and a ReLU activation function; the core parameters of each convolutional layer can include, for example, kernel parameters, kernel stride, and zero-padding method; the kernel parameters include kernel length, number of input channels, and number of output channels. Generally, the kernel length is set to 3, the number of input channels is equal to the number of output channels of the previous convolutional network unit, and the number of output channels is set to twice the number of input channels; the kernel stride can be, for example, set to 1; the zero-padding method can be, for example, set to no zero-padding; the core parameters of each max-pooling layer can include, for example, pooling window length and pooling window stride. Generally, the pooling window length is set to 2, and the pooling window stride is equal to the current pooling window length.

[0093] The number of fully connected network units can be, for example, 3 to 10. Each fully connected network unit includes: a fully connected layer and a sigmoid activation function. The core parameters of each fully connected layer can include, for example, the number of input nodes and the number of output nodes. Generally, the number of input nodes is equal to the number of output nodes of the previous fully connected network unit, and the number of output nodes is set to 0.5 times or 0.25 times the number of input nodes. If the number of output nodes is not an integer, it needs to be rounded down to the nearest integer.

[0094] The output layer may include, for example, a fully connected layer; the output may be, for example, {y} i |i=1,2,3,…,N}, the number of output nodes can be set to D, for example. y D y The meaning is the same as above. In this embodiment of the invention, the neural network may include, for example, 5 convolutional network units and 4 fully connected network units.

[0095] Specifically, the first convolutional network unit among the multiple convolutional network units described above takes a high-dimensional seismic waveform dataset {x} as input. i |i=1,2,3,…,N}, where the number of input channels in the convolution kernel parameters is set to 1.

[0096] Specifically, after the last convolutional network unit in a plurality of convolutional network units, a flattening layer is added. The flattening layer is used to concatenate the vectors of different channels end to end to form a vector and provide input to the first fully connected network unit.

[0097] The input sample point set mentioned in step 102 is {x} i |i=1,2,3,…,N}, the symbols have the same meaning as above;

[0098] The final output sample point set mentioned in step 102 is {y} i |i=1,2,3,…,N}, which has the same meaning as above;

[0099] The output sample point set of the convolutional network unit and the fully connected network unit mentioned in step 102 refers to the output sample point set of the multiple convolutional network units and the multiple fully connected network units mentioned in step 101, hereinafter referred to as... Let represent the set of output sample points for each convolutional network unit and each fully connected network unit, where Let a sample point be a column vector, and let the length of the vector be . but Equivalent to In the dimensional space, i is a point, where i is the sample point number in the dataset, N has the same meaning and value as above, and η is the unified number of the convolutional network unit and the fully connected network unit, η = 1, 2, 3, ..., L, where L is the number of network units, that is, the total number of multiple convolutional network units and multiple fully connected network units in the neural network. In this embodiment of the invention, the neural network may include, for example, 5 convolutional network units and 4 fully connected network units, i.e., L = 5 + 4 = 9.

[0100] Specifically, the above-mentioned If it is the output sample point set of a convolutional network unit, it means This is the output of a convolutional network unit after passing through a flattening layer. The function of the flattening layer is to connect the vectors from different channels end to end to form a single vector.

[0101] Specifically, the probability distribution of distance relationships in the input sample point set is determined using the following formula:

[0102]

[0103] Where, x i x j x s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point x under investigation i When, sample point x j The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to a fixed constant value. For example, α i It can indicate that The information entropy is equal to a fixed constant H. For a specific dimensionality reduction problem, the value of H can range from 1.5 to 4.0.

[0104] Specifically, the probability distribution of the distance relationship between the output sample point sets of convolutional network units and fully connected network units is determined by the following formula:

[0105]

[0106] in, η represents the sample points in the dataset; the subscripts i, j, and s are the sample point numbers in the dataset, and the sample point numbers traverse the range of 1, 2, 3, ..., N, where N is the number of sample points in the dataset; η is the number of multiple convolutional network units and multiple fully connected network units, η = 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; β η For the weighting coefficients, take Indicates the sample points under investigation At that time, sample points The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to the value of a fixed constant.

[0107] Specifically, the probability distribution of distance relationships in the final output sample point set is determined using the following formula:

[0108]

[0109] Among them, y i y j y s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; the symbol |||| denotes the L2 norm of a vector.

[0110] Specifically, the objective function of the neural network is constructed according to the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, using the following formula:

[0111]

[0112] in, Represents sample point x i With x j The distance between them; Indicates the sample point x under investigation i When, sample point x j The conditional probability; Indicates the sample point x under investigation j When, sample point x i The conditional probability; Indicates the sample point x under investigation t When, sample point x s The conditional probability; Indicates the sample point x under investigation s When, sample point x t The conditional probability; Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Represents sample point y i With y j The distance between them; Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; Indicates the sample point y under investigation s At that time, sample point y t The conditional probability; subscripts i, j, s, t are the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset; the range of the subscript η is 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; the objective function J(k) and the weighting coefficient λ η (k) represents a function that trains the neural network for k iterations. Generally, we take... Where k = 1, 2, 3, ..., K, and K is the total number of times the neural network is trained.

[0113] In this embodiment, training the neural network according to its objective function may include, for example, adjusting the neural network parameters using a stochastic gradient descent method based on the objective function of the neural network to obtain the trained neural network.

[0114] In this embodiment, the parameters of the neural network are adjusted using a stochastic gradient descent method based on the objective function of the neural network to obtain the trained neural network. For example, this may include:

[0115] Repeat the following steps until the objective function value of the neural network converges to its minimum:

[0116] A predetermined number of earthquake data points are randomly selected from the earthquake dataset; the predetermined number of randomly selected earthquake data points are input into the neural network, and the dimensionality reduction result of the corresponding earthquake data points is output; the objective function value of the neural network is calculated based on the earthquake data points and the corresponding dimensionality reduction result.

[0117] Specifically, let the neural network be trained k = 1, 2, 3, ..., K times. Starting from k = 1, N earthquake data points {x} are randomly selected from the earthquake dataset each time. i Given a set of data points (i = 1, 2, 3, ..., N), input to the neural network. The network outputs the dimensionality reduction result for the corresponding seismic data points. Based on the seismic data points and their dimensionality reduction results, the objective function value of the neural network is calculated. The neural network parameters are then adjusted according to the objective function J(k), completing one neural network parameter update. This process is repeated until k = K. Generally, K = 1000 to 5000 is sufficient to minimize the objective function J(k). In this embodiment, K = 5000 is used.

[0118] In this embodiment, the earthquake data point vectors from the earthquake dataset are sequentially input into the trained neural network to perform dimensionality reduction on the earthquake data, resulting in a dimensionality reduction map. If the earthquake dataset has lateral spatial distribution characteristics, the dimensionality reduction result can be used to draw a dimensionality-reduced planar map of the earthquake data. In this embodiment, the earthquake dataset has lateral spatial distribution characteristics, containing a total of 530 × 841 = 445,730 earthquake data points on a two-dimensional plane, meaning the earthquake dataset contains 445,730 earthquake waveform vectors. Figure 3 This is a specific example diagram illustrating the random selection of a predetermined number of seismic data points from a seismic dataset according to Embodiment 1 of the present invention. Figure 3 As shown, this displays 50 earthquake data points randomly selected from the earthquake dataset. Figure 4 This is a specific example diagram illustrating the plotting of a dimensionality-reduced plane map of seismic data based on the dimensionality reduction value provided in Embodiment 1 of the present invention, as shown below. Figure 4 As shown, the exploration target is a sandy debris flow on a fore-edge slope break, which includes four known wells. The sandy debris flow characteristics are more obvious around the wells in the seismic data dimensionality reduction plane. Figure 5 This is the result of globally sorting seismic waveforms in the entire seismic waveform dataset based on the magnitude of the dimensionality reduction value, as provided in Embodiment 1 of the present invention, and it also displays... Figure 4 The correspondence between medium grayscale color scales and seismic waveforms.

[0119] In this embodiment, after using the trained neural network to perform dimensionality reduction on the seismic data, the process may further include: obtaining the dimensionality reduction value of the seismic data; sorting the seismic dataset according to the size of the dimensionality reduction value to obtain sorted seismic data points; determining multiple clustering thresholds based on the sorted seismic data points and combining geological information and drilling information; and clustering the seismic data points whose dimensionality reduction value falls between two adjacent clustering thresholds. Figure 6 This is a specific example diagram of clustering seismic data points according to Embodiment 1 of the present invention, as shown in the figure. Figure 6 As shown in the embodiment of the present invention, there are 4 known well points. The exploration target is a sandy debris flow on a certain fore-edge slope break. Well w3 did not hit the sandy debris flow, while wells w1, w2, and w4 did hit the sandy debris flow. Two thresholds are selected from -20 to 20. Seismic data points whose dimensionality reduction values ​​are between two adjacent clustering thresholds are clustered. For example, they can be divided into three categories to obtain the seismic data clustering results. Figure 7 This is the result of globally sorting seismic waveforms in the entire seismic waveform dataset based on the magnitude of the dimensionality reduction value, as provided in Embodiment 1 of the present invention, and it also displays... Figure 6 The correspondence between black, gray, and white markers and earthquake waveforms.

[0120] Example 2:

[0121] This invention provides an application example of a seismic data dimensionality reduction method in multi-attribute seismic dimensionality reduction in actual work areas. The core of this method is to achieve nonlinear seismic data dimensionality reduction by leveraging the memory and feature extraction capabilities of convolutional neural networks. This solves the problem that some existing nonlinear dimensionality reduction methods cannot be applied to seismic data dimensionality reduction due to limitations in storage space and computational load.

[0122] like Figure 1 As shown, the seismic data dimensionality reduction method includes the following steps:

[0123] Step 101: Construct a neural network for dimensionality reduction of seismic data, using the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer;

[0124] Step 102: Construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set;

[0125] Step 103: Train the neural network according to the objective function of the neural network;

[0126] Step 104: Use the trained neural network to perform dimensionality reduction on the seismic data.

[0127] In step 101, the input sample point set of the neural network is an earthquake dataset, i.e., a high-dimensional earthquake multi-attribute dataset, hereinafter referred to as {x}. i |i=1,2,3,…,N} represents, where x i A sample point is a column vector composed of multiple seismic attribute values ​​at the same spatiotemporal location. It is equivalent to a point in a high-dimensional space, and the length of the vector is the number of input seismic attributes, i.e., the dimension of the high-dimensional space, hereinafter referred to as D. x In this context, i represents the sample point number in the dataset, and N represents the number of sample points in the dataset, i.e., the batch size of the neural network. Generally, N is taken as an integer between 1000 and 5000. In this embodiment of the invention, the input can be, for example, three types of seismic attribute data, namely D... x =3, and take N=1000.

[0128] The final output sample point set of the neural network described in step 101 is the dimensionality reduction result of the earthquake dataset, that is, the dimensionality reduction result of the corresponding high-dimensional earthquake multi-attribute dataset, hereinafter referred to as {y i |i=1,2,3,…,N} represents, where y i For a single sample point, i.e., the dimensionality reduction result of a multi-attribute seismic event, it is a column vector, equivalent to a point in the low-dimensional space. The length of the vector is the dimension of the low-dimensional space, hereinafter referred to as D. y This means that, generally speaking, we take D. y =1 or D y =2, where i is the sample point number in the dataset, and N has the same meaning and value as above. In this embodiment of the invention, D is taken as... y =1.

[0129] like Figure 2As shown, the batch size of the neural network is N, with the meaning and value of N as described above. The neural network includes: multiple convolutional network units, multiple fully connected network units, and an output layer; the number of convolutional network units can be, for example, 1 to 10, and each convolutional network unit includes: a convolutional layer, a max-pooling layer, and a ReLU activation function; the core parameters of each convolutional layer can include, for example, kernel parameters, kernel stride, and zero-padding method; the kernel parameters include kernel length, number of input channels, and number of output channels. Generally, the kernel length is set to 3, the number of input channels is equal to the number of output channels of the previous convolutional network unit, and the number of output channels is set to twice the number of input channels; the kernel stride can be, for example, set to 1; the zero-padding method can be, for example, set to no zero-padding; the core parameters of each max-pooling layer can include, for example, pooling window length and pooling window stride. Generally, the pooling window length is set to 2, and the pooling window stride is equal to the current pooling window length.

[0130] The number of fully connected network units can be, for example, 3 to 10. Each fully connected network unit includes: a fully connected layer and a sigmoid activation function. The core parameters of each fully connected layer can include, for example, the number of input nodes and the number of output nodes. Generally, the number of input nodes is equal to the number of output nodes of the previous fully connected network unit, and the number of output nodes is set to 0.5 times or 0.25 times the number of input nodes. If the number of output nodes is not an integer, it needs to be rounded down to the nearest integer.

[0131] The output layer may include, for example, a fully connected layer; the output may be, for example, {y} i |i=1,2,3,…,N}, the number of output nodes can be set to D, for example. y D y The meaning is the same as above. In this embodiment of the invention, the neural network may include, for example, one convolutional network unit and four fully connected network units.

[0132] Specifically, the first convolutional network unit among the multiple convolutional network units described above takes a high-dimensional seismic waveform dataset {x} as input. i |i=1,2,3,…,N}, where the number of input channels in the convolution kernel parameters is set to 1.

[0133] Specifically, after the last convolutional network unit in a plurality of convolutional network units, a flattening layer is added. The flattening layer is used to concatenate the vectors of different channels end to end to form a vector and provide input to the first fully connected network unit.

[0134] The input sample point set mentioned in step 102 is {x} i |i=1,2,3,…,N}, the symbols have the same meaning as above;

[0135] The final output sample point set mentioned in step 102 is {y} i |i=1,2,3,…,N}, which has the same meaning as above;

[0136] The output sample point set of the convolutional network unit and the fully connected network unit mentioned in step 102 refers to the output sample point set of the multiple convolutional network units and the multiple fully connected network units mentioned in step 101, hereinafter referred to as... Let represent the set of output sample points for each convolutional network unit and each fully connected network unit, where Let a sample point be a column vector, and let the length of the vector be . but Equivalent to In the dimensional space, i is a point, where i is the sample point number in the dataset, N has the same meaning and value as above, and η is the unified number of the convolutional network unit and the fully connected network unit, η = 1, 2, 3, ..., L, where L is the number of network units, that is, the total number of multiple convolutional network units and multiple fully connected network units in the neural network. In this embodiment of the invention, the neural network may include, for example, 1 convolutional network unit and 4 fully connected network units, i.e., L = 1 + 4 = 5.

[0137] Specifically, the above-mentioned If it is the output sample point set of a convolutional network unit, it means This is the output of a convolutional network unit after passing through a flattening layer. The function of the flattening layer is to connect the vectors from different channels end to end to form a single vector.

[0138] Specifically, the probability distribution of distance relationships in the input sample point set is determined using the following formula:

[0139]

[0140] Where, x i x j x s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point x under investigation i When, sample point x j The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to a fixed constant value. For example, α i It can indicate that The information entropy is equal to a fixed constant H. For a specific dimensionality reduction problem, the value of H can range from 1.5 to 4.0.

[0141] Specifically, the probability distribution of the distance relationship between the output sample point sets of convolutional network units and fully connected network units is determined by the following formula:

[0142]

[0143] in, η represents the sample points in the dataset; the subscripts i, j, and s are the sample point numbers in the dataset, and the sample point numbers traverse the range of 1, 2, 3, ..., N, where N is the number of sample points in the dataset; η is the number of multiple convolutional network units and multiple fully connected network units, η = 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; β η For the weighting coefficients, take Indicates the sample points under investigation At that time, sample points The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to the value of a fixed constant.

[0144] Specifically, the probability distribution of distance relationships in the final output sample point set is determined using the following formula:

[0145]

[0146] Among them, y i y j y s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; the symbol |||| denotes the L2 norm of a vector.

[0147] Specifically, the objective function of the neural network is constructed according to the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, using the following formula:

[0148]

[0149] in, Represents sample point x i With x j The distance between them; Indicates the sample point x under investigation i When, sample point x j The conditional probability; Indicates the sample point x under investigation j When, sample point x i The conditional probability; Indicates the sample point x under investigation t When, sample point x s The conditional probability; Indicates the sample point x under investigation s When, sample point x t The conditional probability; Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Represents sample point y i With y j The distance between them; Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; Indicates the sample point y under investigation s At that time, sample point y t The conditional probability; subscripts i, j, s, t are the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset; the range of the subscript η is 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; the objective function J(k) and the weighting coefficient λ η (k) represents a function that trains the neural network for k iterations. Generally, we take... Where k = 1, 2, 3, ..., K, and K is the total number of times the neural network is trained.

[0150] In this embodiment, training the neural network according to its objective function may include, for example, adjusting the neural network parameters using a stochastic gradient descent method based on the objective function of the neural network to obtain the trained neural network.

[0151] In this embodiment, the parameters of the neural network are adjusted using a stochastic gradient descent method based on the objective function of the neural network to obtain the trained neural network. For example, this may include:

[0152] Repeat the following steps until the objective function value of the neural network converges to its minimum:

[0153] A predetermined number of earthquake data points are randomly selected from the earthquake dataset; the predetermined number of randomly selected earthquake data points are input into the neural network, and the dimensionality reduction result of the corresponding earthquake data points is output; the objective function value of the neural network is calculated based on the earthquake data points and the corresponding dimensionality reduction result.

[0154] Specifically, let the neural network be trained k = 1, 2, 3, ..., K times. Starting from k = 1, N earthquake data points {x} are randomly selected from the earthquake dataset each time. i The neural network is input with the sequence |i=1,2,3,…,N}, and outputs the dimensionality reduction result of the corresponding seismic data points. Based on the seismic data points and the corresponding dimensionality reduction results, the objective function value of the neural network is calculated, and the neural network parameters are adjusted according to the objective function J(k), completing one neural network parameter update. This process is repeated until k=K. Generally, K=1000~5000 is sufficient to make the objective function J(k) converge to its minimum value. In this embodiment of the invention, K=5000 is used.

[0155] In this embodiment, the trained neural network is used to sequentially input the earthquake data point vectors from the earthquake dataset into the trained neural network to perform dimensionality reduction processing on the earthquake data, obtaining the dimensionality reduction result. If the earthquake dataset has lateral spatial distribution characteristics, the dimensionality reduction result can be used to draw a dimensionality-reduced planar map of the earthquake data. Generally, earthquake multi-attribute dimensionality reduction results need to be combined with known geological information and drilling information to form clustering results to have greater geological significance. Therefore, this embodiment of the invention does not provide a dimensionality-reduced planar map of earthquakes, but directly provides a dimensionality-reduced planar map of earthquakes.

[0156] In this embodiment, after using the trained neural network to perform dimensionality reduction on the seismic data, the process may further include: obtaining the dimensionality reduction value of the seismic data; sorting the seismic dataset according to the size of the dimensionality reduction value to obtain sorted seismic data points; determining multiple clustering thresholds based on the sorted seismic data points and combining geological information and drilling information; and clustering the seismic data points whose dimensionality reduction value falls between two adjacent clustering thresholds.

[0157] In this embodiment of the invention, the earthquake multi-attribute dataset has a lateral spatial distribution feature. On a two-dimensional plane, it contains a total of 500×600=300,000 earthquake multi-attribute data points, that is, the earthquake multi-attribute dataset contains 300,000 earthquake multi-attribute vectors. Figure 8 This is a coherent seismic property planar diagram related to underground faults and cracks, provided according to Embodiment 2 of the present invention. Figure 9 This is a plan view of the seismic properties of a curvature body related to underground faults and cracks, provided according to Embodiment 2 of the present invention. Figure 10This is an AVAZ seismic attribute planar map related to underground faults and fissures, provided according to Embodiment 2 of the present invention. The seismic multi-attribute dataset of this embodiment includes data from... Figure 8 , Figure 9 , Figure 10 The seismic attributes shown are coherence volume seismic attributes, curvature volume seismic attributes, and AVAZ (fractured reservoir characterization) seismic attributes. All three types of seismic attributes can characterize subsurface faults and fractures, but they have different scale characteristics. Each attribute is only sensitive to faults and fractures within a certain scale range. Therefore, conducting multi-attribute seismic dimensionality reduction and clustering work is of great significance for comprehensively evaluating the fracture development of the target layer in the work area. Figure 11 The seismic multi-attribute clustering result obtained by combining seismic multi-attribute dimensionality reduction results, known geological information, and drilling information, as provided in Embodiment 2 of the present invention, is as follows: Figure 11 As shown, there are 3 wells in the work area. Drilling information indicates that well w1 has moderately developed fractures, well w2 is a horizontal well with the most developed fractures, and well w3 has no developed fractures. Based on this geological information and drilling information, this embodiment of the invention selects a clustering threshold and clusters seismic data points whose dimensionality reduction values ​​belong to two adjacent clustering thresholds, thereby obtaining the seismic data clustering results.

[0158] This invention also provides a seismic data dimensionality reduction device, as described in the following embodiments. Since the principle behind this device's problem-solving is similar to that of the seismic data dimensionality reduction method, its implementation can be found in the implementation of the seismic data dimensionality reduction method; repeated details will not be elaborated further.

[0159] This invention provides a seismic data dimensionality reduction device. Figure 12 This is a schematic diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention, as shown below. Figure 12 As shown, the device includes the following modules:

[0160] The neural network construction module 1201 is used to construct a neural network for seismic data dimensionality reduction, taking the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set; the neural network includes: convolutional network units, fully connected network units, and an output layer;

[0161] The objective function construction module 1202 is used to construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set.

[0162] Training module 1203 is used to train the neural network according to the objective function of the neural network;

[0163] The dimensionality reduction module 1204 is used to perform dimensionality reduction processing on the seismic data using the trained neural network.

[0164] In one embodiment, the neural network includes: a plurality of convolutional network units, a plurality of fully connected network units, and an output layer; each convolutional network unit includes: a convolutional layer, a max pooling layer, and a ReLU activation function; each fully connected network unit includes: a fully connected layer and a Sigmoid activation function; the output layer includes: a fully connected layer.

[0165] In one embodiment, a flattening layer is added after the last convolutional network unit in a plurality of convolutional network units. The flattening layer is used to concatenate vectors from different channels end to end to form a vector and provide input to the first fully connected network unit.

[0166] Figure 13 This is a specific example diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention, as shown below. Figure 13 As shown, in this example, the seismic data dimensionality reduction device also includes:

[0167] The first probability distribution determination module 1301 is used to determine the probability distribution of the distance relationship of the input sample point set according to the following formula before the objective function construction module 1202 constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set:

[0168]

[0169] Where, x i x j x s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point x under investigation i When, sample point x j The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to the value of a fixed constant.

[0170] Figure 14 This is a specific example diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention, as shown below. Figure 14 As shown, in this example, the seismic data dimensionality reduction device also includes:

[0171] The second probability distribution determination module 1401 is used to determine the probability distribution of the distance relationship between the output sample point sets of the convolutional network units and the fully connected network units according to the following formula before the objective function construction module 1202 constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network units and the fully connected network units, and the final output sample point set:

[0172]

[0173] in, η represents the sample points in the dataset; the subscripts i, j, and s are the sample point numbers in the dataset, and the sample point numbers traverse the range of 1, 2, 3, ..., N, where N is the number of sample points in the dataset; η is the number of multiple convolutional network units and multiple fully connected network units, η = 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; β η For the weighting coefficients, take Indicates the sample points under investigation At that time, sample points The conditional probability; the symbol |||| denotes the L2 norm of a vector; α i Indicates to make The information entropy is equal to the value of a fixed constant.

[0174] Figure 15 This is a specific example diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention, as shown below. Figure 15 As shown, in this example, the seismic data dimensionality reduction device also includes:

[0175] The third probability distribution determination module 1501 is used to determine the probability distribution of the distance relationship of the final output sample point set according to the following formula before the objective function construction module 1202 constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set:

[0176]

[0177] Among them, y i y j y s The subscripts i, j, and s represent the sample points in the dataset; the subscripts i, j, and s represent the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset. Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; the symbol |||| denotes the L2 norm of a vector.

[0178] In one embodiment, the objective function construction module 1202 is specifically used to construct the objective function of the neural network according to the following formula:

[0179]

[0180] in, Represents sample point x i With x j The distance between them; Indicates the sample point x under investigation i When, sample point x j The conditional probability; Indicates the sample point x under investigation j When, sample point x i The conditional probability; Indicates the sample point x under investigation t When, sample point x s The conditional probability; Indicates the sample point x under investigation s When, sample point x t The conditional probability; Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Represents sample point y i With y j The distance between them; Indicates the sample point y under investigation i At that time, sample point y j The conditional probability; Indicates the sample point y under investigation s At that time, sample point y t The conditional probability; subscripts i, j, s, t are the sample point numbers in the dataset, and the range of the sample point numbers is 1, 2, 3, ..., N, where N is the number of sample points in the dataset; the range of the subscript η is 1, 2, 3, ..., L, where L is the total number of multiple convolutional network units and multiple fully connected network units in the neural network; the objective function J(k) and the weighting coefficient λ η (k) represents a function that trains the neural network a total of k times.

[0181] In one embodiment, the training module 1203 is specifically used to: adjust the parameters of the neural network using a stochastic gradient descent method according to the objective function of the neural network, so as to obtain the trained neural network.

[0182] In one embodiment, the training module 1203 is specifically used to: repeat the following steps until the objective function value of the neural network converges to a minimum value: randomly select a set number of earthquake data points from the earthquake dataset; input the randomly selected set number of earthquake data points into the neural network and output the dimensionality reduction result of the corresponding earthquake data points; calculate the objective function value of the neural network based on the earthquake data points and the dimensionality reduction result of the corresponding data points.

[0183] Figure 16 This is a specific example diagram of the seismic data dimensionality reduction device provided in an embodiment of the present invention, as shown below. Figure 16 As shown, in this example, the seismic data dimensionality reduction device also includes:

[0184] The clustering module 1601 is used after the dimensionality reduction module 1204 uses the trained neural network to perform dimensionality reduction on the seismic data: obtaining the dimensionality reduction value of the seismic data; sorting the seismic dataset according to the size of the dimensionality reduction value to obtain sorted seismic data points; determining multiple clustering thresholds based on the sorted seismic data points and combining geological information and drilling information; and clustering the seismic data points whose dimensionality reduction value belongs to two adjacent clustering thresholds.

[0185] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described seismic data dimensionality reduction method.

[0186] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described seismic data dimensionality reduction method.

[0187] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described seismic data dimensionality reduction method.

[0188] In this embodiment of the invention, a neural network for seismic data dimensionality reduction is constructed. The seismic dataset is used as the input sample point set, and the dimensionality reduction result of the seismic dataset is used as the final output sample point set. The neural network includes convolutional network units, fully connected network units, and an output layer. An objective function of the neural network is constructed based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set. The neural network is trained based on the objective function. The trained neural network is then used to perform dimensionality reduction on the seismic data. Compared with existing technical solutions, constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point sets of the convolutional network units and fully connected network units, and the final output sample point set, and training the neural network based on the objective function, leverages the memory and feature extraction capabilities of the convolutional network units. This eliminates the need to train the neural network using the probability distribution of the distance relationships of all sample point sets, avoiding limitations in storage space and computational load, and enabling dimensionality reduction of nonlinear seismic data.

[0189] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0190] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0191] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0192] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0193] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dimensionality reduction of seismic data, characterized in that, include: A neural network for dimensionality reduction of seismic data is constructed, with the seismic dataset as the input sample point set and the dimensionality reduction result of the seismic dataset as the final output sample point set. The neural network includes: a convolutional network unit, a fully connected network unit, and an output layer; The objective function of the neural network is constructed based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set. Train the neural network according to the objective function of the neural network; The trained neural network is used to perform dimensionality reduction on the seismic data; Before constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the following steps are also included: The probability distribution of the distance relationship between the output sample point sets of convolutional network units and fully connected network units is determined by the following formula: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. This refers to the numbering of multiple convolutional network units and multiple fully connected network units. =1, 2, 3, ... L , L This represents the total number of convolutional network units and fully connected network units in the neural network. For the weighting coefficients, take ; Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This indicates the computational norm of a vector; Indicates to make The information entropy is equal to the value of a fixed constant.

2. The seismic data dimensionality reduction method as described in claim 1, characterized in that, The neural network includes: multiple convolutional network units, multiple fully connected network units, and an output layer; Each convolutional network unit includes: a convolutional layer, a max pooling layer, and a ReLU activation function; Each fully connected network unit includes: a fully connected layer and a Sigmoid activation function; The output layer consists of a fully connected layer.

3. The seismic data dimensionality reduction method as described in claim 2, characterized in that, After the last convolutional network unit in a series of convolutional network units, a flattening layer is added. This flattening layer is used to concatenate vectors from different channels to form a single vector and provide input to the first fully connected network unit.

4. The seismic data dimensionality reduction method as described in claim 1, characterized in that, Before constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the method further includes: determining the probability distribution of the distance relationships between the input sample point set according to the following formula: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This indicates the computational norm of a vector; Indicates to make The information entropy is equal to the value of a fixed constant.

5. The seismic data dimensionality reduction method as described in claim 1, characterized in that, Before constructing the objective function of the neural network based on the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, the following steps are also included: The probability distribution of distance relationships in the final output sample point set is determined by the following formula: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This represents the calculation of the L2 norm of a vector.

6. The seismic data dimensionality reduction method as described in claim 1, characterized in that, The objective function of the neural network is constructed according to the probability distribution of the distance relationships between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set, using the following formula: ； in, , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points Conditional probability; subscript i, j, s, t The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N The number of sample points in the dataset; subscript The traversal range is 1, 2, 3, ... L , L The total number of convolutional network units and fully connected network units in the neural network; the objective function. with weighting coefficients This is represented as the number of training iterations for the neural network. k The function.

7. The seismic data dimensionality reduction method as described in claim 1, characterized in that, Training the neural network according to its objective function includes: Based on the objective function of the neural network, the parameters of the neural network are adjusted using the stochastic gradient descent method to obtain the trained neural network.

8. The seismic data dimensionality reduction method as described in claim 7, characterized in that, Based on the objective function of the neural network, the parameters of the neural network are adjusted using the stochastic gradient descent method to obtain the trained neural network, including: Repeat the following steps until the objective function value of the neural network converges to its minimum: A predetermined number of earthquake data points are randomly selected from the earthquake dataset; A predetermined number of randomly selected seismic data points are input into the neural network, which outputs the dimensionality reduction result of the corresponding seismic data points. The objective function value of the neural network is calculated based on the seismic data points and the corresponding dimensionality reduction results.

9. The seismic data dimensionality reduction method as described in claim 1, characterized in that, After using the trained neural network to perform dimensionality reduction on the seismic data, the process also includes: Obtain the dimensionality reduction value of the earthquake data, sort the earthquake dataset according to the size of the dimensionality reduction value, and obtain the sorted earthquake data points; Based on the sorted seismic data points, combined with geological and drilling information, multiple clustering thresholds were determined. Seismic data points whose dimensionality reduction values ​​fall between two adjacent clustering thresholds are clustered.

10. A seismic data dimensionality reduction device, characterized in that, include: A neural network construction module is used to construct a neural network for dimensionality reduction of seismic data. The seismic dataset is used as the input sample point set, and the dimensionality reduction result of the seismic dataset is used as the final output sample point set. The neural network includes: convolutional network units, fully connected network units, and an output layer. The objective function construction module is used to construct the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set. A training module is used to train the neural network according to the objective function of the neural network; The dimensionality reduction module is used to perform dimensionality reduction processing on the seismic data using the trained neural network. It also includes a second probability distribution determination module, used to determine the probability distribution of the distance relationship between the output sample point sets of the convolutional network units and the fully connected network units according to the following formula before the objective function construction module constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network units and the fully connected network units, and the final output sample point set: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. This refers to the numbering of multiple convolutional network units and multiple fully connected network units. =1, 2, 3, ... L , L This represents the total number of convolutional network units and fully connected network units in the neural network. For the weighting coefficients, take ; Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This indicates the computational norm of a vector; Indicates to make The information entropy is equal to the value of a fixed constant.

11. The seismic data dimensionality reduction device as described in claim 10, characterized in that, The neural network includes: multiple convolutional network units, multiple fully connected network units, and an output layer; Each convolutional network unit includes: a convolutional layer, a max pooling layer, and a ReLU activation function; Each fully connected network unit includes: a fully connected layer and a Sigmoid activation function; The output layer consists of a fully connected layer.

12. The seismic data dimensionality reduction device as described in claim 11, characterized in that, After the last convolutional network unit in a series of convolutional network units, a flattening layer is added. This flattening layer is used to concatenate vectors from different channels to form a single vector and provide input to the first fully connected network unit.

13. The seismic data dimensionality reduction device as described in claim 10, characterized in that, It also includes a first probability distribution determination module, used to determine the probability distribution of the distance relationship of the input sample point set according to the following formula before the objective function construction module constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This indicates the computational norm of a vector; Indicates to make The information entropy is equal to the value of a fixed constant.

14. The seismic data dimensionality reduction device as described in claim 10, characterized in that, It also includes a third probability distribution determination module, used to determine the probability distribution of the distance relationship of the final output sample point set according to the following formula before the objective function construction module constructs the objective function of the neural network based on the probability distribution of the distance relationship between the input sample point set, the output sample point set of the convolutional network unit and the fully connected network unit, and the final output sample point set: ； in, , , For sample points in the dataset; subscript i, j, s The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N This represents the number of sample points in the dataset. Indicates the sample points under investigation At that time, sample points Conditional probability; symbol This represents the calculation of the L2 norm of a vector.

15. The seismic data dimensionality reduction device as described in claim 10, characterized in that, The objective function construction module is specifically used to construct the objective function of the neural network according to the following formula: ； in, , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points The conditional probability; , Represents sample points and The distance between them; Indicates the sample points under investigation At that time, sample points The conditional probability; Indicates the sample points under investigation At that time, sample points Conditional probability; subscript i, j, s, t The sample points in the dataset are numbered, and the sample point numbers are iterated through in the range of 1, 2, 3, ... N , N The number of sample points in the dataset; subscript The traversal range is 1, 2, 3, ... L , L The total number of convolutional network units and fully connected network units in the neural network; the objective function. with weighting coefficients This is represented as the number of training iterations for the neural network. k The function.

16. The seismic data dimensionality reduction device as described in claim 10, characterized in that, The training module is specifically used for: Based on the objective function of the neural network, the parameters of the neural network are adjusted using the stochastic gradient descent method to obtain the trained neural network.

17. The seismic data dimensionality reduction device as described in claim 16, characterized in that, The training module is specifically used for: Repeat the following steps until the objective function value of the neural network converges to its minimum: A predetermined number of earthquake data points are randomly selected from the earthquake dataset; A predetermined number of randomly selected seismic data points are input into the neural network, which outputs the dimensionality reduction result of the corresponding seismic data points. The objective function value of the neural network is calculated based on the seismic data points and the corresponding dimensionality reduction results.

18. The seismic data dimensionality reduction device as described in claim 10, characterized in that, It also includes a clustering module, used after the dimensionality reduction module uses the trained neural network to perform dimensionality reduction processing on the seismic data: Obtain the dimensionality reduction value of the earthquake data, sort the earthquake dataset according to the size of the dimensionality reduction value, and obtain the sorted earthquake data points; Based on the sorted seismic data points, combined with geological and drilling information, multiple clustering thresholds were determined. Seismic data points whose dimensionality reduction values ​​fall between two adjacent clustering thresholds are clustered.

19. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the seismic data dimensionality reduction method according to any one of claims 1 to 9.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the seismic data dimensionality reduction method according to any one of claims 1 to 9.

21. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the seismic data dimensionality reduction method according to any one of claims 1 to 9.

Images