A coal seam thickness prediction method and device based on deep learning

Abstract

A coal seam thickness prediction method and device based on deep learning, the method comprising: constructing a sample data set for coal seam thickness prediction according to well data of a work area; establishing a coal seam thickness prediction model based on a deep learning network; training the coal seam thickness prediction model using the sample data set to obtain a trained coal seam thickness prediction model; and predicting the coal seam thickness of a target layer section using the trained coal seam thickness prediction model according to seismic data of the work area.

Description

Technical Field

[0001] This article relates to the field of geophysical oil and gas exploration technology, and in particular to a method and device for predicting coal seam thickness based on deep learning. Background Technology

[0002] Coal seams are typically thin, with thicknesses ranging from 0 to 20 meters in many representative regions of my country. Due to limitations in seismic data resolution, coal seam thickness prediction is a typical problem of thin-layer seismic prediction. Furthermore, influenced by geological conditions and other factors, the floor and caprock of coal seams in some areas often exhibit diverse characteristics. For example, the Benxi Formation No. 8 coal seam in the Ordos Basin exhibits multiple reservoir-caprock combinations, including coal-sand, coal-mud, and coal-ash combinations. This means the P-wave impedance characteristics of the caprock vary. Seismic reflections near the coal seam are affected not only by the coal seam thickness but also by the impedance variations of the caprock and floor, further increasing the difficulty of predicting coal seam thickness using seismic reflection information.

[0003] Deep coal seam gas is an important successor area for oil and gas exploration and development. Accurately predicting coal seam thickness is of great research significance for guiding coal seam gas exploration and development and well location deployment. Therefore, how to accurately predict coal seam thickness is an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to provide a method and apparatus for predicting coal seam thickness based on deep learning. This method utilizes real well data from the research area to establish and augment sample label data for predicting thin coal seam thickness using deep learning, and constructs a deep learning network for predicting coal seam thickness using multi-convolutional kernels. This enables a nonlinear mapping from seismic waveforms to coal seam thickness, thereby improving the accuracy of thin coal seam thickness prediction.

[0005] In a first aspect, this application provides a deep learning-based method for predicting coal seam thickness, the method comprising:

[0006] Based on well data from the work area, a sample dataset for coal seam thickness prediction is constructed; a coal seam thickness prediction model is established based on a deep learning network; the coal seam thickness prediction model is trained using the sample dataset to obtain a trained coal seam thickness prediction model; based on seismic data from the work area, the coal seam thickness of the target segment is predicted using the trained coal seam thickness prediction model.

[0007] Secondly, embodiments of the present invention also provide a coal seam thickness prediction device based on deep learning, the device comprising: a memory and a processor; the memory is used to store a program for performing coal seam thickness prediction based on deep learning, and the processor is used to read and execute the program for performing coal seam thickness prediction based on deep learning, and execute the method described in any one of the above embodiments.

[0008] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a data processing program, wherein the data processing program is executed by a processor using the deep learning-based coal seam thickness prediction method described in any of the above embodiments.

[0009] Compared with related technologies, this application provides a deep learning-based method for predicting coal seam thickness. The method includes: constructing a sample dataset for coal seam thickness prediction based on well data from the work area; establishing a coal seam thickness prediction model based on a deep learning network; training the coal seam thickness prediction model using the sample dataset to obtain a trained coal seam thickness prediction model; and predicting the coal seam thickness of the target segment using the trained coal seam thickness prediction model based on seismic data from the work area. By utilizing real well data from the research area to establish and augment sample label data for deep learning-based thin coal seam thickness prediction, and constructing a multi-convolutional kernel deep learning network for coal seam thickness prediction, a nonlinear mapping from seismic waveforms to coal seam thickness is achieved, thereby improving the accuracy of thin coal seam thickness prediction.

[0010] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the solutions described in the description and the accompanying drawings. Attached Figure Description

[0011] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0012] Figure 1 This is a flowchart of a deep learning-based coal seam thickness prediction method according to an embodiment of this application.

[0013] Figure 2 This is a schematic diagram of a deep learning-based coal seam thickness prediction device according to an embodiment of this application;

[0014] Figure 3 A schematic diagram of the longitudinal wave impedance curves of the pseudo-wells in the first augmented well dataset constructed in some exemplary embodiments;

[0015] Figure 4 A schematic diagram of the longitudinal wave impedance curves of a pseudo-well in a second augmented well dataset constructed in some exemplary embodiments;

[0016] Figure 5 This is a schematic diagram showing the correspondence between synthetic seismic records and P-wave impedance curves of sample labeled datasets in some exemplary embodiments;

[0017] Figure 6This is a schematic diagram of a neural network structure for seismic prediction of coal seam thickness in some exemplary embodiments;

[0018] Figure 7 This is a schematic diagram of a well-connected profile showing the predicted top and bottom coal seams in some exemplary embodiments. Detailed Implementation

[0019] This application describes several embodiments, but these descriptions are exemplary and not restrictive, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0020] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application may also be combined with any conventional features or elements to form a unique inventive scheme as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive schemes to form another unique inventive scheme as defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in this application may be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.

[0021] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.

[0022] Traditional methods for predicting thin layer thickness include peak amplitude (PA) and peak frequency (PF) methods. These methods are based on analyzing the relationship between PA or PF and thin layer thickness using a wedge model and forward seismic records to predict the thickness. However, these methods have several shortcomings:

[0023] 1. The variations in lithology and impedance of the caprock and basement were not considered;

[0024] 2. The wedge body model was not created using actual well curves;

[0025] 3. Only amplitude or frequency information was used for thin layer prediction, without using all the information of the seismic waveform for thickness prediction;

[0026] 4. The monotonicity of the amplitude-thickness relationship differs on both sides of the tuning point. Therefore, when predicting the thickness, it is necessary to divide the interval into multiple sub-intervals and consider different situations.

[0027] The inventors analyzed the aforementioned traditional methods and developed a method using deep learning to predict the thickness of thin coal seams, overcoming the shortcomings of traditional thin coal seam prediction methods.

[0028] This invention provides a deep learning-based method for predicting coal seam thickness, such as... Figure 1 As shown, the method includes steps S1-S4:

[0029] S1: Construct a sample dataset for coal seam thickness prediction based on well data from the work area;

[0030] S2: Establish a coal seam thickness prediction model based on deep learning networks;

[0031] S3: Use the sample dataset to train the coal seam thickness prediction model to obtain a trained coal seam thickness prediction model;

[0032] S4: Based on the seismic data of the work area, use the trained coal seam thickness prediction model to predict the coal seam thickness of the target section.

[0033] In one exemplary embodiment, step S1 involves constructing a sample dataset for coal seam thickness prediction based on well data from the work area; specifically, it requires converting the depth-domain well data into the time domain and establishing a sample label dataset for deep learning-based thin coal seam thickness prediction.

[0034] Step S11: Construct the first augmented well dataset;

[0035] The real wells in the work area are paired up in pairs. Under the constraint of the top and bottom positions of the coal seam, the first pseudo-well set of the transition area between the two wells is established by linear interpolation. The first pseudo-well set is combined with the real wells to form the first augmented well dataset.

[0036] In this step, the pairing method is to pair any two wells, regardless of their distance. For example, if there are three wells a, b, and c in the work area, pairing any two wells can result in three combinations: ab, ac, and bc.

[0037] Additionally, it should be noted that the pseudo-well lacks location information; the transition in the transition zone between the two wells only represents the transition of the impedance curve. That is, the impedance curves and coal seam thicknesses of the two randomly selected wells are different. Interpolation can be used to obtain the transition from one well to the other (the impedance curve and coal seam thickness change gradually), such as... Figure 2 .

[0038] Step 12: Construct a second augmented well dataset based on the first augmented well dataset.

[0039] Using each well in the first augmented well dataset as a base well, and constrained by the top and bottom positions of the coal seam, a second set of pseudo-wells with different coal seam thicknesses is established by stretching and compressing the longitudinal wave impedance curves of the coal seam segments. The second set of pseudo-wells is combined with the first augmented well dataset to form the second augmented well dataset.

[0040] Step 13: Based on the second augmentation well dataset, establish a sample label dataset for deep learning thin coal seam thickness prediction.

[0041] In step 11, the process of converting well data in the depth domain to the time domain is as follows:

[0042] Step 01: Obtain data from actual wells within the work area. The well data includes longitudinal wave impedance curves, coal seam top interpretation results, and coal seam bottom interpretation results, etc.

[0043] Step 02: Convert the depth well curve into a time-domain well curve through well-seismic calibration.

[0044] In this step, we first determine whether the sampling rate of the seismic data is 1 millisecond. If it is not 1 millisecond, the seismic data needs to be resampled to 1 millisecond.

[0045] Then, well-seismic calibration is performed using resampled seismic data.

[0046] Step 03: Based on the well seismic calibration results, convert the interpretation results of the coal seam top and bottom to the time domain;

[0047] In step 11, the process of constructing the first pseudo-well set is as follows:

[0048] The actual wells within the work area are paired up in pairs. The paired actual wells are denoted as W0 and W1, respectively. N-1It is necessary to establish N-2 pseudo wells using the two paired real wells. The pseudo wells in the first set of pseudo wells are denoted as W1, W2, ..., W... N-2 The value of N can be changed as needed, but N = 10 is generally used. The specific setup process is as follows:

[0049] Step 01: Segment the wave impedance blocks of the two paired real wells;

[0050] The longitudinal wave impedance curve z0 of well W0 is divided into three segments: z0 = [a0; b0; c0];

[0051] Where a0 is the caprock of well W0, b0 is the coal seam of well W0, c0 is the bottom plate of well W0, and the coal seam time thickness is set to h0 milliseconds.

[0052] W N-1 Longitudinal wave impedance curve z of the well N-1 Divided into three segments z N-1 =[a N-1 b N-1 ;c N-1 ];

[0053] Among them, a N-1 For W N-1 The capping section of the well, b N-1 For W N-1 The coal seam section of the well, c N-1 For W N-1 The bottom section of the well, W N-1 The coal seam thickness of the well is h. N-1 millisecond.

[0054] Step 02, let W be the first pseudo-well concentration. i Longitudinal wave impedance curve z of well (i = 1, 2, ..., N-2) i It consists of three segments, z i =[a i b i ;c i ];

[0055] Among them, a i For the cap layer segment, b i For coal seam section, c i This is the bottom section, and its coal seam thickness is set to h. i millisecond;

[0056] Step 03: Determine W based on the parameter calculation formula of the pseudo-well. i The thickness of the caprock, coal seam, floor, and coal seam over time in the well;

[0057] The formula for calculating the parameters is:

[0058]

[0059] In the above formula, i is the number of the first pseudo-well, i = 1, 2, ..., N-2, a i For the cap layer segment, b i For coal seam section, c i The bottom section has a coal seam thickness of h. i The unit is milliseconds;

[0060] In constructing the cap layer segment a i At that time, it is necessary to ensure that the caprock sections a0 and a of the two known wells are... N-1 The lowest sampling point (the bottom of the cap layer) is aligned, therefore, let This is a segment of the longitudinal wave impedance curve extracted from a0, with a time window size of T. w milliseconds, eye The lowest sampling point of a0 (the bottom of the cap layer) is aligned with the lowest sampling point of a0. From a N-1 The segment of the longitudinal wave impedance curve is extracted, and the time window size of the extracted segment is T. w milliseconds, and and a N-1 The lowest sampling point (bottom of the cap layer segment) is aligned; in constructing the cap bottom plate segment c i At that time, it is necessary to ensure that the bottom plate sections c0 and c of the two known wells are... N-1 The topmost sampling point (the top of the bottom plate segment) is aligned; therefore, let... This is a segment of the longitudinal wave impedance curve extracted from c0, with a time window size of T. w milliseconds, and It is aligned with the top sampling point of c0 (the top of the bottom plate segment). From c N-1 The segment of the longitudinal wave impedance curve is extracted, and the time window size of the extracted segment is T. w milliseconds, and and c N-1 The topmost sampling point (the top of the bottom plate segment) is aligned; It involves stretching or compressing the longitudinal wave impedance curve b0 with a time thickness of h0, so that its time thickness becomes h. i Longitudinal wave impedance curve at time, For a time thickness of h N-1 millisecond longitudinal wave impedance curve b N-1 Stretch or compress it so that its thickness becomes h over time. i Longitudinal wave impedance curve at milliseconds; T w It's a time window. w The value of T is set according to the specific work area conditions. w =100.

[0061] In step 12, the process of constructing the second augmented well dataset based on the first augmented well dataset is as follows:

[0062] Step 120: Divide the P-wave impedance curve of each well U in the first augmented well dataset into three segments;

[0063] The longitudinal wave impedance curve z of each well U in the first augmented well dataset is divided into three segments z = [a; b; c], where a is the caprock segment, b is the coal seam segment, and c is the floor segment, with the coal seam time thickness being h milliseconds.

[0064] Step 121: Use the U-well to establish N more dummy wells;

[0065] Using well U, N pseudo-wells will be established, denoted as U0, U1, ..., U N-1 Generally, N=10 is sufficient.

[0066] Step 122, determine the U of each pseudo-well. i Longitudinal wave impedance curve z′ i

[0067] Let U i Longitudinal wave impedance curve z′ of well (i = 0, 2, ..., N-1) i It consists of three segments, z′ i =[a′ i b′ i ;c′ i ]; where a′ i For the cap layer segment, b′ i For the coal seam section, c′ i The bottom section has a coal seam thickness of h′. i ;

[0068] a′i=a, c′i=c:

[0069] Time thickness changes h′ i for:

[0070]

[0071] Where, h′ min With h′ max The minimum and maximum values ​​of coal seam time thickness are set based on the actual geological conditions of the coal seam in the work area. Generally, h is taken as... min =0,h max =25 is sufficient.

[0072] It is the longitudinal wave impedance curve b with time thickness h stretched or compressed so that its time thickness becomes h′i.

[0073] In step 13, a sample label dataset for deep learning thin coal seam thickness prediction is established based on the second augmented well dataset.

[0074] For each well F in the second augmented well dataset, its longitudinal wave impedance curve z is divided into three segments z = [a; b; c], where a is the caprock segment, b is the coal seam segment, and c is the floor segment, and the coal seam time thickness is h milliseconds.

[0075] In one exemplary embodiment, in step S2, a coal seam thickness prediction model is established based on a deep learning network.

[0076] The input to the coal seam thickness prediction model is: seismic synthetic record s, which is a vector of length 2*H+1; where H is the time window, H=100.

[0077] The output of the coal seam thickness prediction model is the relative position of the top and bottom of the coal seam [p1; p2], which is a vector of length 2.

[0078] The network consists of a core network unit c(nL) and fully connected network units d(nH1, nH2) connected in series:

[0079] The fully connected network element d(nH1, nH2) is:

[0080] Among them, the fully connected network unit:

[0081] d(nH1,nH2)=Flatten()→Linear(n Flatten , nH1)→Relu()→Linear(nH1, nH2)→Relu()→Linear(nH2, 2)→Sigmod()

[0082] In this layer, Flatten() is the flattening layer, Linear() is the linear transformation layer, and ReLU() and Sigmod() are the activation functions. Generally, nH1 = 128 and nH1 = 32 can be chosen.

[0083] The core network unit c(nL) is:

[0084] c(nL) consists of multiple multi-convolutional kernel network units b(nC) in nC out Series connection configuration:

[0085] c(nL)=b(1,16)→b(3*16,2 4+1 → b(3*2) 4+1 ,2 4+2 )→…→b(3*2 4+nL-2 ,2 4+nL-1 ),

[0086] Where nL is the number of network units, and nL = 5 is usually sufficient.

[0087] The above multi-convolutional kernel network unit b(nC) in nC out It consists of the following network units:

[0088]

[0089] in,

[0090]

[0091] In the above formula, a(nC) in nC out k) is composed of one-dimensional convolutional units connected in series with one-dimensional BatchNorm units, and then connected in series with ReLU activation functions:

[0092] a(nC in nC out ,k)=Conv1d(nC in nC out ,k)→BatchNorm1d(nC out → Relu(), nC in nC represents the number of input channels. out is the number of input channels, k is the kernel size, all convolutions are not padded with zeros, AvgPool1d(2) is a one-dimensional average pooling layer with a kernel size of 2, and cat(x, 1) means merging several single outputs in x along the channel dimension.

[0093] In this embodiment, the objective function of the coal seam thickness prediction model is:

[0094]

[0095] Where n is the number of samples input into the neural network each time, y i For the output of the neural network, This is the label data.

[0096] In step S3, the coal seam thickness prediction model is trained using the sample data in the sample label dataset to obtain a trained deep learning network model for coal seam thickness.

[0097] In step S3, the coal seam thickness prediction model can be trained using the stochastic gradient descent method. The specific steps are as follows:

[0098] Step S31: Set the number of training iterations for the coal seam thickness prediction model to k = 1, 2, 3, ..., K;

[0099] Step S32: Starting from k=1, extract L samples from the second augmentation well dataset and input them into the deep learning network model for coal seam thickness to obtain the output of the deep learning network model for coal seam thickness; generally, L=8 is sufficient.

[0100] Step S33: Based on the output of the deep learning network model for coal seam thickness, optimize the parameters of the deep learning network model for coal seam thickness using the objective function to complete one parameter update of the deep learning network model for coal seam thickness.

[0101] Step S34: Repeat the above process until k = K, and obtain the trained deep learning network model of coal seam thickness.

[0102] Generally, taking K = 1000 to 5000 will allow the objective function to converge to its minimum value.

[0103] S4: Predict coal seam thickness using a trained deep learning network model of coal seam thickness, combined with seismic data from the work area.

[0104] S41. Process the seismic data in the work area and extract the seismic trace data.

[0105] S411. Determine the location of the coal seam center based on the seismic data of the work area.

[0106] Interpret the seismic data in the work area to determine the location of the coal seam center.

[0107] S412. Extract seismic trace data using the center of the coal seam as the window center.

[0108] Use the center of the coal seam as the window center and a window size of 2*H+1 to extract seismic trace data; H=100 is sufficient.

[0109] S42. Input the extracted seismic trace data into the trained deep learning network model of coal seam thickness to determine the coal seam thickness.

[0110] S421. Input the extracted seismic trace data into the trained deep learning network model of coal seam thickness, and obtain the output results as [p1; p2];

[0111] S422. Determine the positions of the top and bottom of the coal seam based on the output results p1 and p2, respectively.

[0112] The position of the coal seam top relative to the topmost point of the extracted seismic trace (i.e., the topmost point of the window with a size of 2*H+1) is:

[0113] The position of the bottom of the coal seam relative to the top of the extracted seismic trace (i.e., the top of the window with a window size of 2*H+1) is:

[0114] S423. Determine the thickness of the coal seam;

[0115] Coal seam thickness is

[0116] In this embodiment, firstly, based on the convolution model, the reflection coefficient r is obtained using the P-wave impedance curve z. The reflection coefficient r is then convolved with the seismic wavelet w to obtain the synthetic seismic record S0, where the seismic wavelet w is a statistical wavelet extracted from actual seismic data. Next, a sample label truncation window is set, with a window size of 2*H+1 milliseconds. Generally, H=100 is sufficient. The center of the window is aligned with the center of the coal seam, and the window is randomly perturbed within a range of R milliseconds above and below. A portion s from the synthetic seismic record s0 is truncated using the perturbed window and used as the input to the neural network. Here, the time difference between the top and bottom of the coal seam and the uppermost edge of the window is set as follows: milliseconds and Milliseconds can be used to calculate the relative positions of the top and bottom of the coal seam with respect to the top edge of the window. and Use [p1; p2] as the output labels of the neural network. Generally, the perturbation amount R = 10 is sufficient. This perturbation amount is used to correct the error caused by manual picking of the coal seam center in actual seismic data, so that the training and learning of the neural network is not limited by the inaccuracy of picking the coal seam center.

[0117] Setting the output labels of the neural network to [p1; p2] has the advantage that both p1 and p2 are numbers between 0 and 1, which is beneficial for the neural network to work with the activation function to establish the loss function. Using p1 and p2, not only can the thickness of the coal seam be calculated, but also the position of the top and bottom of the coal seam can be located, so as to achieve high-precision interpretation of the top and bottom of the coal seam on the actual seismic profile.

[0118] This invention proposes a deep learning-based method for predicting coal seam thickness in seismic events, which has the following technical advantages:

[0119] 1) Considering the variations in lithology and impedance of the caprock and floor, it can adapt to coal seam thickness prediction under more complex geological conditions;

[0120] 2) Using actual well curves to create wedge-shaped models results in synthetic seismic records that more closely resemble actual conditions.

[0121] 3) Using seismic waveform information that simultaneously contains amplitude and frequency information for thickness prediction enhances information utilization;

[0122] 4) It utilizes the nonlinear information extraction capability of deep learning networks, eliminating the need to divide the data into multiple sub-intervals when performing thickness prediction;

[0123] 5) A deep learning network for predicting coal seam thickness using multiple convolutional kernels, which is more conducive to extracting features of different scales from seismic waveform information by using convolutional kernels of different lengths.

[0124] 6) It can not only predict the thickness of coal seams, but also locate the top and bottom of coal seams, achieving high-precision interpretation of the top and bottom of coal seams on actual seismic profiles.

[0125] Secondly, embodiments of the present invention also provide a deep learning-based seismic prediction device for coal seam thickness, such as... Figure 2 As shown, the device includes: a memory 200 and a processor 210; the memory is used to store a deep learning-based seismic prediction program for coal seam thickness, and the processor is used to read and execute the deep learning-based seismic prediction program for coal seam thickness, and to execute the method described in any of the above embodiments.

[0126] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a data processing program, wherein the data processing program is executed by a processor using the deep learning-based seismic prediction method for coal seam thickness as described in any of the above embodiments.

[0127] Example 1

[0128] This example demonstrates the implementation process of a deep learning-based method for predicting coal seam thickness in seismic earthquakes. The specific implementation process is as follows:

[0129] Step 101: Construct a sample dataset for a deep learning network model of coal seam thickness.

[0130] It includes three sub-steps:

[0131] Step 101-1: Pair the real wells in the work area into pairs. Under the constraint of the top and bottom positions of the coal seam, establish pseudo wells in the transition area between the two wells through linear interpolation. Combine these pseudo wells with the real wells to construct the first augmented well dataset.

[0132] like Figure 3 As shown, for two real wells W0 and W9 in the work area, under the constraint of the top and bottom positions of the coal seam, the first pseudo-well set W1 to W8 of the transition zone between the two wells is established by linear interpolation.

[0133] Step 101-2: Construct a second augmented well dataset using each well in the first augmented well dataset as a base well.

[0134] like Figure 4 As shown, for each base well in augmented well dataset 1, under the constraint of the top and bottom positions of the coal seam, a second pseudo-well set U0 to U9 with different coal seam thicknesses is established by stretching and compressing the longitudinal wave impedance curves of the coal seam segment.

[0135] Step 101-3: Based on the second augmented well dataset, establish a sample dataset for a deep learning network model of coal seam thickness.

[0136] like Figure 5 As shown, in the sample dataset of the deep learning network model for coal seam thickness, the synthetic seismic record and the P-wave impedance curve of each sample data are in one-to-one correspondence.

[0137] Step 102: Establish a deep learning network model for coal seam thickness.

[0138] like Figure 6 The diagram shows the structure of a deep learning network model for coal seam thickness. Solid arrows connect the network output to the backbone network units, while dashed arrows illustrate the specific composition of each network unit.

[0139] Step 103: Train the neural network using the sample dataset;

[0140] Step 104: Apply the trained deep learning network model of coal seam thickness to predict coal seam thickness.

[0141] Figure 7 This is a well-connected profile of the predicted top and bottom coal seams in the example work area, which includes two training wells and one verification well. It can be seen that the top and bottom coal seams of the verification well match the low-resistivity boundary of the coal seam in the longitudinal wave impedance well.

[0142] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

Claims

1. A deep learning-based method for predicting coal seam thickness, characterized in that, The method includes: Based on the well data of the work area, a sample dataset for coal seam thickness prediction was constructed; A coal seam thickness prediction model based on deep learning networks; The coal seam thickness prediction model is trained using the sample dataset to obtain a trained coal seam thickness prediction model. Based on the seismic data of the work area, the coal seam thickness of the target section is predicted using a trained coal seam thickness prediction model.

2. The deep learning-based coal seam thickness prediction method according to claim 1, characterized in that, The sample dataset for predicting coal seam thickness, constructed based on well data from the work area, includes: Convert the depth domain data of real wells in the work area into the time domain data of real wells; The first augmented well dataset was constructed using real well time-domain data; Based on the P-wave impedance curves of each well in the first augmented well dataset, a second augmented well dataset is constructed. A sample dataset for predicting coal seam thickness was established based on time-domain data from real wells in the work area and data from the second augmentation well.

3. The deep learning-based coal seam thickness prediction method according to claim 2, characterized in that, The construction of the first augmented well dataset using real well time-domain data includes: The actual wells in the work area were paired up in pairs; For each pair of real wells, under the constraints of the top and bottom of the coal seam, the first set of pseudo-wells between each pair of real wells is established by linear interpolation; The first augmented well dataset is constructed based on the first pseudo-well set and the real wells in the work area.

4. The deep learning-based coal seam thickness prediction method according to claim 3, characterized in that, For each pair of real wells, under the constraints of the coal seam top and bottom, a first set of pseudo-wells is established between every two real wells through linear interpolation, including: The longitudinal wave impedance curves of the two real wells were segmented to determine the caprock, coal seam, floor, and coal seam time thickness of each real well. The caprock, coal seam, floor, and coal seam time thickness of each pseudo-well are determined using parameter calculation formulas based on the actual well's caprock, coal seam, floor, and coal seam time thickness. The formula for calculating the parameters is as follows: In the above formula, i is the number of the pseudo-well in the first pseudo-well set, i = 1, 2, ..., N-2, a i For the cap layer segment, b i For coal seam section, c i The bottom section has a coal seam thickness of h. i ; T is taken from the bottom of the a0 cap layer segment upwards. w The wave impedance curve segment in milliseconds, From a N-1 T is cut upwards from the bottom of the cap layer. w The wave impedance curve segment in milliseconds; T is cut downwards from the top of the c0 base plate segment. w The wave impedance curve segment in milliseconds, From c N-1 Cut T downwards from the top of the base plate section w The wave impedance curve segment in milliseconds; It involves stretching or compressing the longitudinal wave impedance curve b0 with a time thickness of h0, so that its time thickness becomes h. i Longitudinal wave impedance curve at time, For a time thickness of h N-1 millisecond longitudinal wave impedance curve b N-1 Stretch or compress it so that its thickness becomes h over time. i Longitudinal wave impedance curve at milliseconds; T w It is a time window.

5. The deep learning-based coal seam thickness prediction method according to claim 2, characterized in that, The construction of a second augmented well dataset based on the P-wave impedance curves of each well in the first augmented well dataset includes: The longitudinal wave impedance curves of each well in the first augmentation well dataset are divided into three segments, and the caprock segment, coal seam segment, floor segment, and coal seam time thickness of each well are determined respectively. Construct a second pseudo-well using each well in the first augmented well dataset; A second augmented well dataset is constructed using the second pseudo-well set and the first augmented well dataset.

6. The deep learning-based coal seam thickness prediction method according to claim 5, characterized in that, The construction of a second pseudo-well using each well in the first augmented well dataset includes: Based on the caprock, coal seam, floor, and coal seam time thickness of the first augmented well data set, the longitudinal wave impedance curve z′i of each pseudo well in the second pseudo well set is determined. z′ i =[a′ i ;b′ i ;c′ i ]; Where, a′ i For the cap layer segment, b′ i For the coal seam section, c′ i The bottom section has a coal seam thickness of h′. i ; a′i=a, c′i=c; The coal seam thickness over time is h′ i for: In the above formula, a represents the caprock segment of U, c represents the base plate segment, and h′ min It is the minimum time thickness of the coal seam, h′ max It is the maximum time thickness of the coal seam.

7. The deep learning-based coal seam thickness prediction method according to claim 1, characterized in that, The coal seam thickness prediction model is composed of a core network unit c(nL) and a fully connected network unit d(nH1, nH2) connected in series. Wherein, the core network unit c(nL) is: c(nL)=b(1,16)→b(3*16,2 4+1 )→b(3*2 4+1 ,2 4+2 )→…→b(3*2 4+nL-2 ,2 4+nL-1 ), In the above formula, nL is the number of network units, and b(nC) is the number of network units. in nC out ) is a multi-convolutional kernel network unit.

8. The deep learning-based coal seam thickness prediction method according to claim 7, characterized in that, The objective function of the coal seam thickness prediction model is: Where MSELoss is the objective function value, n is the number of samples input to the neural network each time, and y i For the output of the neural network, This is the label data.

9. The deep learning-based coal seam thickness prediction method according to claim 8, characterized in that, The step of predicting the coal seam thickness of the target section based on seismic data from the work area and using a trained coal seam thickness prediction model includes: The location of the coal seam center was determined based on the seismic data of the work area; Seismic trace data is extracted using the center of the coal seam as the window center. The extracted seismic trace data is input into the trained coal seam thickness prediction model to obtain the coal seam thickness.

10. The deep learning-based coal seam thickness prediction method according to claim 9, characterized in that, The step of inputting the extracted seismic trace data into the trained coal seam thickness prediction model to obtain the coal seam thickness includes: The extracted seismic trace data is input into the trained coal seam thickness prediction model to determine the location of the top and bottom of the coal seam. The thickness of the coal seam is determined by the location of the top and bottom of the coal seam.

11. A coal seam thickness prediction device based on deep learning, characterized in that, The apparatus includes a memory and a processor; the memory is used to store a program for performing deep learning-based coal seam thickness prediction, and the processor is used to read and execute the program for performing deep learning-based coal seam thickness prediction, and to perform the method according to any one of claims 1-10.

12. A computer-readable storage medium storing a data processing program, the data processing program being executed by a processor according to any one of claims 1-10, the deep learning-based coal seam thickness prediction method.

Images