A method for identifying thin interbedded lithologic structures based on deep learning

Abstract

The present invention discloses a method for identifying thin interbedded lithologic structures based on deep learning. The method comprises: manually collecting and processing seismic data; well-seismic calibration and deep-time conversion to obtain the lithologic structure and seismic data of the thin interbedded section at the well point; time-frequency analysis and sliding time windows to obtain time-frequency spectrum images within a local time window of the seismic trace near the well, which serve as input samples for the deep learning network; obtaining and classifying the lithologic structure of the thin interbedded lithologic structures within the local time window and encoding them as sample labels; completing training of a network model for identifying thin interbedded lithologic structures to obtain optimal network parameters; using the optimal network parameters, traversing all seismic traces to obtain lithologic classification results of the thin interbedded lithologic structures within all time windows; and processing the results within overlapping time windows to determine the final spatial distribution of the thin interbedded lithologic structures. Compared with existing methods, this method combines well logging and seismic data, can break through seismic resolution, obtain the spatially continuous distribution of the thin interbedded lithologic structures, and does not require wavelet information, resulting in high recognition accuracy and ease of promotion.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir prediction in artificial seismic exploration, and more specifically to the field of lithology and thin interbedded reservoir exploration. Specifically, the present invention relates to a method for identifying thin interbedded lithologic structures based on deep learning. Background Art

[0002] Artificial seismic exploration involves artificially stimulating seismic waves at the surface. As these waves propagate underground, they are reflected and transmitted by interfaces between rock layers with different properties. Surface detectors receive these waves, producing seismic records. The characteristics of these records are related to the properties and structure of the underlying rock formations. By processing and interpreting these records, the properties and morphology of the underlying rock formations can be inferred. Because artificial seismic exploration surpasses other geophysical exploration methods in terms of precision and accuracy, it plays a crucial role in oil and gas reservoir prediction.

[0003] With the continuous advancement of oil and gas exploration in recent decades, large, easy-to-find structural traps have been exhausted, and oil and gas exploration targets have gradually shifted to thin, complex lithologic traps with rapidly changing lithologies. Thin interbedded reservoirs are common in my country's oil and gas basins, and large-scale thin interbedded lithologic reservoirs are currently the focus of conventional oil and gas exploration to increase reserves and production. Thin interbedded strata have small individual thicknesses, frequent vertical interbedding, and rapid lateral changes in lithology and thickness, making accurate identification and prediction of their lithology challenging. Therefore, research on identifying thin interbedded lithologic structures is a highly significant task in oil and gas exploration.

[0004] In actual artificial seismic exploration, strata with a single layer thickness less than 1 / 4 of the main wavelength of the seismic wavelet are usually considered to be thin layers, and multiple thin layers stacked vertically form thin interlayers. Due to the limitation of the vertical resolution of seismic data, it is difficult to accurately identify the internal lithologic structure of thin interlayers. When using seismic data for lithologic structure identification, layer-by-layer seismic attributes and seismic inversion methods are usually used. The resolution of the seismic attribute method is equivalent to that of the seismic data. For thin interlayers, it cannot effectively solve the problem of internal lithologic structure identification; the seismic inversion method can integrate well data, effectively improve the resolution of the results, and provide useful assistance for thin interlayer identification, but the inversion method requires seismic wavelet information, and the accuracy of the wavelet directly affects the accuracy of the inversion. The present invention provides a thin interlayer lithologic structure identification method based on deep learning, which uses the two-dimensional time-frequency spectrum of seismic data as input, automatically extracts features related to the thin interlayer lithologic structure in the time-frequency spectrum with the help of deep learning, establishes a mapping relationship between the local time-frequency spectrum and the lithologic structure, and realizes the accurate identification of the thin interlayer lithologic structure. Compared with existing methods, this method combines artificial seismic data and well data to break through the resolution of seismic data and obtain the spatial continuous distribution of thin interbedded lithologic structures. It does not require seismic wavelet information, has high recognition accuracy, and is easy to promote and use. Summary of the Invention

[0005] The purpose of the present invention is to provide a thin interbed lithologic structure identification method based on deep learning, so as to improve the accuracy of identifying and predicting the internal lithologic structure of thin interbeds.

[0006] To achieve the above objectives, the present invention provides a method for identifying thin interbedded lithologic structures based on deep learning, which includes:

[0007] Collect artificial seismic data in the field and obtain post-stack seismic data through indoor processing;

[0008] Perform well-seismic calibration and deep-time conversion to obtain thin interbedded lithologic structure and seismic data at the well point;

[0009] Perform time-frequency analysis on the seismic data of the target thin interbedded section to obtain a two-dimensional time-frequency spectrum of the seismic trace;

[0010] The two-dimensional time-frequency spectrum images in the local time window are obtained from the seismic traces near the well and used as input samples of the deep learning network after preprocessing;

[0011] Obtain thin interbedded lithologic structures within a local time window from well logging interpretation lithologic data, and classify and encode them as sample labels.

[0012] Build a deep learning network architecture, define the loss function, select the network model evaluation method, complete the thin interbedded lithologic structure identification network model training, and obtain the optimal network parameters;

[0013] Traverse the two-dimensional time-spectrum diagrams of all seismic traces to obtain the entire set of two-dimensional time-spectrum images within the local time window;

[0014] Using the optimal network parameters, all sets of pre-processed two-dimensional time-spectral images are converted into thin interbed lithologic classification results;

[0015] The lithologic classification results within the overlapping time windows are processed to determine the final spatial distribution of the thin interbedded lithologic structure.

[0016] The beneficial effects of this invention are that, compared with methods that use layer-by-layer seismic attributes for lithologic structure identification, the proposed method combines seismic and well data, achieving higher vertical resolution. It also eliminates the need for manual matching between seismic attributes and lithologic properties, reducing subjectivity and ambiguity in identification results. Compared with methods based on seismic inversion, this method does not require known seismic wavelet information, does not require the establishment of an initial inversion model, and is independent of the accuracy of the wavelet and initial model, making it easier to generalize and apply. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flow chart of a method for identifying thin interbedded lithologic structures based on deep learning according to an embodiment of the present invention;

[0019] Figure 2 Seismic data containing thin interbedded sections according to an embodiment of the present invention;

[0020] Figure 3 A seismic trace and a two-dimensional time-frequency spectrum image thereof according to an embodiment of the present invention;

[0021] Figure 4 The sample and label making method of the embodiment of the present invention;

[0022] Figure 5 The deep learning network architecture of the embodiment of the present invention;

[0023] Figure 6 A diagram showing a data set partitioning ratio according to an embodiment of the present invention;

[0024] Figure 7 This is a network hyperparameter table of an embodiment of the present invention;

[0025] Figure 8 1 is the loss curve and accuracy curve of the embodiment of the present invention;

[0026] Figure 9 This is the result of thin interbed identification using seismic relative wave impedance in this embodiment;

[0027] Figure 10 This is the result of thin interbedded lithologic structure recognition based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] Figure 1 FIG. 1 is a flow chart of a method for identifying thin interbedded lithologic structures based on deep learning according to an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0030] S101. Collect artificial seismic data on the surface, process seismic data indoors, and obtain post-stack seismic data. Specifically, in step S101, artificial earthquake refers to an earthquake that is artificially stimulated for the purpose of exploration. When conducting artificial seismic acquisition in the field, it can be three-dimensional acquisition along a surface on the surface, or two-dimensional acquisition along a line, and the obtained post-stack seismic data can be three-dimensional data or two-dimensional data. An underground exploration target in a certain work area at a certain place has a two-dimensional line of well-connected three-dimensional seismic data obtained by three-dimensional acquisition, and the application of an embodiment of the present invention is demonstrated. The underground exploration target in this area includes a set of thin interbedded sections, which are delta front deposits, sand and mudstone lithology, and a multi-layer sand and mudstone interbedded structure developed from bottom to top. The spatial relationship of the reservoir sand body is complex, the lateral changes are fast, and the heterogeneity is strong. Intercept the seismic data of the target thin interbedded section, Figure 2 This is seismic data containing thin interbedded sections in an embodiment of the present invention. The arrows above indicate the locations of the wells, which means that there are 5 wells drilled in the area and the well data are complete.

[0031] S102. Use well logging data and near-well seismic traces to perform well-seismic calibration and depth-time conversion on the target layer to obtain lithologic structural data and seismic data for the thin interbedded section at the well point. Specifically, in step S102, for this embodiment of the present invention, the well logging curves used include acoustic time difference curves and density curves, and the well logging interpretation data used refers to lithologic data. In this example, well-seismic calibration uses the method of creating artificial synthetic seismic records. First, the well logging acoustic time difference curve is used to obtain an acoustic velocity curve through an inverse operation. The acoustic velocity curve is multiplied by the density curve to obtain a longitudinal impedance curve. The longitudinal velocity curve is subjected to a point-by-point reflection coefficient calculation to obtain a discrete longitudinal reflection coefficient sequence. A wavelet matching the main frequency of the seismic data is convolved with the reflection coefficient sequence to obtain a synthetic seismic record. Seismic traces from the well or near-well are extracted from the actual seismic data. The near-well seismic traces are matched and calibrated with the synthetic seismic record to determine the time-depth relationship at the well. During well-seismic calibration, the degree of matching of the target thin interbedded section is emphasized. The same calibration process was performed for all wells within the work area. Using the obtained time-depth relationship, the interpreted lithologic data from each well was converted into lithologic-structural data with time coordinates. The converted lithologic-structural data matched the time coordinates of the seismic data and had the same temporal sampling interval. In this example, the temporal sampling interval of the seismic data was 1 ms.

[0032] S103: Perform time-frequency analysis on the target thin interbedded segment seismic data, converting the one-dimensional waveform data into a two-dimensional time-frequency spectrum. The time-frequency analysis in step S103 can be linear, such as short-time Fourier transform, continuous wavelet transform, and S-transform, or nonlinear, such as Wigner-Ville distribution, pseudo-Wigner-Ville distribution, and Cohen-type time-frequency distribution. In this example, wavelet transform time-frequency analysis is used.

[0033] In step S103, the frequency range of the time-frequency analysis is based on the main frequency f of the seismic data of the target thin interbedded section. m Its effective frequency band range [f1, f h ] OK, low frequency starts from 0hz and high frequency takes 2f m +(f h -f1) / 2 and above, the frequency domain sampling interval is not greater than 1 Hz. In this example, the main frequency of the seismic data is f m It is about 30 Hz, the effective frequency band range is from 8 Hz to 88 Hz, the frequency range of time-frequency analysis is 0 Hz to 100 Hz, and the frequency domain sampling interval is 1 Hz. Figure 3 A seismic trace and its two-dimensional time-frequency spectrum image according to an embodiment of the present invention are shown, wherein the left side is a seismic trace waveform diagram, and the right side is a two-dimensional time-frequency spectrum image obtained by wavelet transforming the seismic trace.

[0034] S104, using a sliding time window to obtain a two-dimensional time-frequency spectrum image in a local time window from the seismic trace near the well, and pre-processing it as an input sample for the deep learning network. Specifically, in step S104, the time window is slid, and its time window length is based on the main frequency f of the seismic data of the target thin interbedded section. m OK, the minimum is 1 / f m 2 times, the maximum is 1 / f m 4 times, and an integer multiple of the time sampling interval; the sliding step of the sliding window is at least one time sampling interval and at most one time window length. In this example, the main frequency of the seismic data f m It is about 30 Hz, the length of the sliding window is determined to be 100 ms, and the sliding step adopts the minimum step size, that is, 1 time sampling interval is 1 ms.

[0035] The preprocessing in step S104 is specifically to perform normalization and resampling on the two-dimensional time-frequency spectrum image within the local time window. Normalization is necessary, and resampling is performed selectively based on the requirements of recognition accuracy and computational efficiency. Resampling can be performed in both the vertical and horizontal directions of the two-dimensional time-frequency spectrum image within the local time window, or in only one direction. In this example, normalization and resampling preprocessing are performed. Normalization is based on the mean and standard deviation of pixel values, mapping pixels to the range of [-1, 1]. After normalization, the data distribution is centered around 0, which is conducive to gradient descent optimization. The formula is:

[0036]

[0037] Where x is the pixel value of the image; μ is the mean used for normalization; σ is the standard deviation used for normalization; x norm is the normalized pixel value.

[0038] In this embodiment, the resampling process uses a linear interpolation method to improve the resolution of the time-frequency spectrum. The specific operation is to perform three rounds of linear interpolation on each line of time-frequency data. Suppose the original data is x = [x1, x2, ..., x n ], after one interpolation, the interpolation data is:

[0039]

[0040] Repeat the above steps three times to obtain the final interpolated data.

[0041] S105: Using the same sliding window as in step S104, obtain the thin interbedded lithologic structure within the local time window from the well logging interpretation lithologic data, classify and encode its lithologic properties, and use it as the sample label. In step S105, the thin interbedded lithologic structure within the local time window is the well logging interpretation lithologic conclusion with the same sampling interval and number of samples as the seismic data within the local time window. The sliding window is the same as in step S104, with the same window length and sliding step size. Figure 4 The sample and labeling method of an embodiment of the present invention are shown. In this example, the time window length and sliding step length are 100ms and 1ms respectively.

[0042] In step S105, lithology classification coding is performed according to the following rules: 1st lithology: classification code value = 1; 2nd lithology: classification code value = 2; ..., nth lithology: classification code value = m; where n is the total lithology category interpreted by the well logging, and m is the total coding category, with m≤n. Lithologies with excessive indications are coded using adjacent values. For example, if the well logging interpretation indicates mudstone, muddy sandstone, and sandstone, the three lithologies can be coded using coding method 1 (mudstone = 1, muddy sandstone = 2, sandstone = 2) or coding method 2 (mudstone = 1, muddy sandstone = 2, sandstone = 3). The coding method used is determined by the required lithology identification accuracy. In this example, the well logging interpretation has two lithology conclusions, labeled mudstone and sandstone, and the lithology codes used are mudstone = 1 and sandstone = 2.

[0043] In step S105, the encoded sample label is a discrete one-dimensional vector consisting of values ​​1, 2, ..., m, and the length of the vector is the same as the number of time sampling points in the sample. In this example, each sample label is a one-dimensional vector consisting of the values ​​1 and 2, with a vector length of 100. In this example, a single label indicates whether the lithology at each time sampling point within its corresponding local time window of 100ms is mudstone or sandstone. As the time window slides, the changes in the values ​​of adjacent labels indicate the vertical changes in the lithology.

[0044] S106, match samples and labels, divide training set and validation set, build deep learning network architecture, define loss function, select network model evaluation method, test network hyperparameters, complete thin interbedded lithologic structure recognition network model training, and obtain optimal network parameters. Deep learning network refers to a deep network with three or more feature extraction layers, and its feature extraction layer uses two-dimensional convolution operation. In this example, a deep network with a U-shaped structure is used. Figure 5 This is a diagram of the deep learning network architecture of an embodiment of the present invention. The network structure is an encoder-decoder structure. The encoder consists of four step-by-step downsampling modules, each of which includes two consecutive 3×3 convolution operations and one 2×2 maximum pooling operation. The decoder consists of four step-by-step upsampling modules, each of which includes one 2×2 deconvolution and two 3×3 convolution operations. Skip connections are used to splice the output of each layer in the encoder with the upsampling result of the corresponding layer in the decoder, achieving the fusion of image features at different scales.

[0045] Figure 6 This is a diagram showing the data set division ratio of an embodiment of the present invention. In this example, there are 5 wells with a total of 3840 labeled samples. The samples of 4 wells are used as training sets. Figure 2 The middle well was used as a blind well for verification, and its samples were used as the verification set. The training set and verification set accounted for 80% and 20% respectively.

[0046] In step S106, the loss function adopts the cross entropy function. If the target thin interbed section has two lithologies, the binary cross entropy function is adopted. If the target thin interbed section has three or more lithologies, the multivariate cross entropy function is adopted. In this example, the target thin interbed section has two lithologies: mudstone and sandstone, so the binary cross entropy function is adopted. The binary cross entropy loss function is:

[0047]

[0048] Where X is the training sample; Y is the label of X; is the probability matrix predicted by the network model; p x Represents a sampling point in X; For p x The label value of is the p predicted by the network model x probability.

[0049] In step S106, the network model evaluation method uses a dual evaluation method of loss curve and accuracy curve. The loss curve is generated by the loss function, and the accuracy curve is generated by the classification result confusion matrix. The loss curve is generated by the change of the loss function value with the number of training iterations, reflecting the optimization process of the model. The confusion matrix of multi-classification is a K×K matrix, where K represents the number of classification categories. The elements X in the matrix are ij Indicates the number of samples whose true category is i and the model predicts j. ii Indicates the number of correctly classified samples, non-diagonal elements X ij (i≠j) represents the number of samples that are misclassified. The confusion matrix can be used to calculate the classification accuracy, and its formula is:

[0050]

[0051] The numerator is the total number of correctly classified samples, and the denominator is the total number of samples. Accuracy is an important indicator of classification model performance. The higher the accuracy, the stronger the model's ability to distinguish between categories. Figure 7 This is a table of network hyperparameters for an embodiment of the present invention. In this example, the Adam optimization method is used. The ideal batch size obtained through testing is 1, epoch size is 150, and learning rate is 0.001. Figure 8 These are the loss curve and accuracy curve of the embodiment of the present invention. The loss curve and accuracy curve of the training set and the validation set can converge normally, indicating that the network training has achieved the expected effect, the thin interbedded lithologic structure identification network model training has been completed, and the optimal network parameters have been obtained.

[0052] S107. Use a sliding time window of the same length as in step S104 to traverse the two-dimensional time-spectrum diagrams of all seismic traces, obtain the entire set of two-dimensional time-spectrum images within the local time window, and preprocess all images in the set. In step S107, the length of the sliding time window is the same as that in step S104; the sliding step size can be the same as or different from the sliding step size in step S104. In this example, a sliding time window identical to that in step S104 is used, with a time window length and a sliding step size of 100ms and 1ms, respectively. In step S107, the same preprocessing method as in step S104 is used. In this example, normalization and resampling preprocessing are performed.

[0053] S108. Using the optimal network parameters, convert the entire set of preprocessed two-dimensional time-spectral images within the local time window into a lithologic classification result for the thin interbedded layers within the local time window. Specifically, in step S108, the optimal network is the deep network with the optimal weights obtained in step S106. The conversion process involves inputting each image in the set of preprocessed two-dimensional time-spectral images within the local time window into the optimal network, and the output is the lithologic classification result for the thin interbedded layers within the corresponding local time window.

[0054] S109: Process the lithologic classification results within the overlapping time windows according to the time position of the center point of each seismic trace sliding window to determine the final spatial distribution of the thin interbedded lithologic structure. Specifically, in step S109, the sliding time window is the same sliding time window as in step S107, having the same time window length and sliding step length. The overlapping time window refers to the overlapping portion of adjacent sliding time windows when the sliding step length is less than the time window length. The lithologic classification results within the overlapping time windows are processed as follows:

[0055]

[0056] Where M is the total number of overlapping time windows, which is related to the length of the sliding time window and the sliding step; N is the total number of overlapping sampling points, which is related to the time length of the seismic trace and the length of the sliding time window; ij is the lithology classification prediction value of the i-th sampling point in the j-th overlapping time window, l iis the final lithologic classification prediction value of the i-th sampling point, and Round(·) is the nearest integer (rounding off) function. The existence of overlapping time windows enables multiple lithologic predictions of the same underground location. Multiple lithologic identification results are integrated to determine the final lithology, avoiding the uncertainty of a single identification result and improving the accuracy and robustness of lithologic identification. In this example, M=101, N=450. In step S109, the final spatial distribution of the thin interbedded lithologic structure refers to the lithologic classification result that is consistent with the lithologic conclusion of the logging interpretation obtained by reverse-encoding the final lithologic classification prediction value of each sampling point in space according to the lithologic classification encoding rules described in step S105. In this example, it refers to converting the one-dimensional vector composed of the values ​​1 and 2 predicted by the network into mudstone and sandstone lithologies. In step S109, the obtained lithologic spatial distribution has the same dimension as the post-stack seismic data in step S101, that is, when the post-stack seismic data is three-dimensional, the obtained final spatial distribution of the thin interbedded lithologic structure is three-dimensional; when the post-stack seismic data is two-dimensional, the obtained final spatial distribution of the thin interbedded lithologic structure is two-dimensional. In this example, Figure 10 This is the result of thin interbedded lithologic structure recognition based on deep learning in an embodiment of the present invention, showing a well line. Figure 9 Compared with the traditional method of using seismic relative wave impedance to identify thin interbed effects, Figure 9 and Figure 10 It can be seen that the method proposed in the present invention has obvious advantages in identifying thin interbeds. The spatial lithologic structure of the thin interbeds is clearly depicted, and the vertical superposition relationship of the sand bodies and the horizontal annihilation points are clearly revealed.

[0057] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. A method for identifying thin interbedded lithologic structures based on deep learning, characterized by The steps include: (1) Collect and process artificial seismic data on the surface to obtain post-stack seismic data; (2) Using well logging data and wellside seismic traces, the target layer is calibrated for well-seismic calibration and deep-time conversion to obtain lithologic structural data and seismic data of the thin interbedded layer at the well point; (3) Perform time-frequency analysis on the seismic data of the target thin interbedded section, convert the one-dimensional waveform data of the seismic trace into a two-dimensional time-frequency spectrum of the seismic trace, and the frequency range of the time-frequency analysis is based on the main frequency f of the seismic data of the target thin interbedded section. m Its effective frequency band [f l ,f h ] OK, low frequency starts from 0hz and high frequency takes 2f m +(f h -f l ) / 2 or above, the frequency domain sampling interval is not greater than 1 Hz; (4) A two-dimensional time-frequency spectrum image in a local time window is obtained from the seismic trace near the well using a sliding time window, and is preprocessed as an input sample of the deep learning network. The time window length of the sliding time window is determined by the main frequency f of the seismic data of the target thin interbedded section. m OK, the minimum is 1 / f m 2 times, the maximum is 1 / f m 4 times of the sliding step; the minimum sliding step is one time sampling interval and the maximum is one time window length; (5) Using the same sliding time window as in step (4), the thin interbedded lithologic structure within the local time window is obtained from the well logging interpretation lithologic data, and its lithologic classification and coding is performed as the sample label. The lithologic classification and coding is performed according to the following rules: The first lithology: classification code value = 1; The second lithology: classification code value = 2; … The nth lithology: classification code value = m; Where n is the total lithologic category interpreted by logging, m is the total coded category, m≤n, and lithologic categories with excessive indication are coded using adjacent values. The encoded sample label is a discrete one-dimensional vector, where the values ​​in the vector are composed of 1, 2, …, m, and the length of the vector is the same as the number of time sampling points of the sample. (6) Match samples and labels, divide the training set and validation set, build a deep learning network architecture, define the loss function, select the network model evaluation method, test the network hyperparameters, complete the thin interbedded lithologic structure recognition network model training, and obtain the optimal network parameters; (7) Using a sliding time window of the same length as in step (4), traverse the two-dimensional time-spectrum images of all seismic traces to obtain a complete set of two-dimensional time-spectrum images within the local time window, and preprocess all images in the set; (8) Using the optimal network parameters, the entire set of preprocessed two-dimensional time-frequency spectrum images within the local time window is converted into the lithologic classification results of the thin interbeds within the local time window; (9) According to the time position of the center point of the sliding window of each seismic trace, the lithologic classification results in the overlapping time window are processed to determine the final spatial distribution of the thin interbedded lithologic structure. The overlapping time window refers to the overlapping part of the adjacent sliding time windows when the sliding step length is less than the time window length. The processing method of the lithologic classification results in the overlapping time window is as follows: Among them, M is the number of overlapping time windows, which is related to the length of the sliding time window and the sliding step; N is the number of overlapping sampling points, which is related to the time length of the seismic trace and the length of the sliding time window; l ij is the lithology classification prediction value of the i-th sampling point in the j-th overlapping time window, l i is the final lithologic classification prediction value of the i-th sampling point, and Round(˙) is the rounding function.

2. The method according to claim 1, characterized in that The artificial earthquake in step (1) is an earthquake artificially stimulated for the purpose of exploration; When conducting artificial seismic acquisition in the field, three-dimensional acquisition can be performed along a surface on the surface, or two-dimensional acquisition can be performed along a line. The post-stack seismic data obtained can be either three-dimensional or two-dimensional.

3. The method according to claim 1, characterized in that The logging data in step (2) include acoustic wave time difference curves, density curves and lithologic data obtained from logging interpretation; The lithologic structural data of the thin interbedded section at the well point refers to the lithologic data in the time domain after depth-time conversion at the well point, which is consistent with the time coordinate of the seismic data and has the same time sampling interval.

4. The method according to claim 1, characterized in that The time-frequency analysis described in step (3) can be either linear time-frequency analysis or nonlinear time-frequency analysis.

5. The method according to claim 1, characterized in that The sliding time window in steps (4) and (5) is the same, having the same time window length and the same sliding step size; The preprocessing described in step (4) refers to normalization and resampling of the two-dimensional time-frequency spectrum image within the local time window. The resampling can be performed in both the vertical and horizontal directions of the two-dimensional time-frequency spectrum image within the local time window, or in only one direction.

6. The method according to claim 1, characterized in that The thin interbedded lithologic structure within the local time window described in step (5) is the lithologic conclusion of the well logging interpretation with the same time sampling interval and number of samples as the seismic data within the local time window.

7. The method according to claim 1, characterized in that The deep learning network described in step (6) refers to a deep network with more than three feature extraction layers. Its feature extraction layer is performed using two-dimensional convolution operations; The loss function adopts a cross entropy function. If the target thin interbed section has two lithologies, a binary cross entropy function is used. If the target thin interbed section has three or more lithologies, a multivariate cross entropy function is used. The network model evaluation method adopts a dual evaluation method of loss curve and accuracy curve, the loss curve is generated by the loss function, and the accuracy curve is generated by the classification result confusion matrix.

8. The method according to claim 1, characterized in that The sliding time window described in step (7) has the same length as the sliding time window length in step (4), and its sliding step size is the same as or different from the sliding step size in step (4); the preprocessing adopts the same method as the preprocessing described in step (4).

9. The method according to claim 1, characterized in that The optimal network described in step (8) is the deep network with optimal weights obtained in step (6); The conversion refers to inputting each image in the pre-processed two-dimensional time-spectrum image set within the local time window into the optimal network, and the output is the lithology classification result of the thin interbed in the corresponding local time window.

10. The method according to claim 1, characterized in that The existence of the overlapping time windows described in step (9) enables multiple lithology predictions for the same underground location. Multiple lithology identification results are integrated to determine the final lithology, avoiding the uncertainty of a single identification result and improving the accuracy and robustness of lithology identification. The final spatial distribution of the thin interbedded lithologic structure refers to the lithologic classification result that is consistent with the lithologic conclusion of the logging interpretation obtained by inversely encoding the final lithologic classification prediction value of each sampling point in space according to the lithologic classification coding rule described in step (5); The spatial distribution has the same dimension as the post-stack seismic data described in step (1), that is, when the post-stack seismic data is three-dimensional data, the final spatial distribution of the thin interbedded lithologic structure obtained is three-dimensional; when the post-stack seismic data is two-dimensional data, the final spatial distribution of the thin interbedded lithologic structure obtained is two-dimensional.

Images