A method for identifying a layer boundary in a geological envelope based on artificial intelligence

Abstract

The application provides a method for identifying a layer boundary in a geological envelope based on artificial intelligence, comprising: determining seismic exploration data and logging data of a target work area; extracting seismic reflection features from the seismic exploration data based on a first trained neural network model, and simultaneously extracting logging sequence features from the logging data based on a second trained neural network model; generating geological property representation of the target work area according to the seismic reflection features and the logging sequence features; determining implicit topological structure representation of the geological envelope in the target work area according to the geological property representation based on constraints of stratum contact relationship and lithology sequence for the target work area; and processing the implicit topological structure representation based on a third trained neural network model to predict the layer boundary of the geological envelope in the target work area. The application makes the prediction result of the layer boundary of the geological envelope more in line with the actual geological law.

Description

Technical Field

[0001] This invention relates to the fields of mineral resource exploration and data interpretation, and in particular to a method for identifying stratigraphic boundaries in geological envelopes based on artificial intelligence. Background Technology

[0002] In fields such as oil and gas exploration, mineral resource development, and engineering geological surveys, accurately determining the stratigraphic boundaries of geological envelopes within a target work area is one of the core technical requirements. The precise location of these stratigraphic boundaries directly impacts the scientific validity of stratigraphic division, resource reserve assessment, and development plan design. Especially in areas with complex geological structures, boundary prediction requires comprehensive analysis of multi-source geological data to meet the demands of practical engineering projects for refined geological information.

[0003] Existing schemes for predicting the stratigraphic boundaries of geological envelopes mostly use seismic exploration data as the main input, supplemented by a small amount of well logging data for auxiliary verification. After denoising the seismic data through a filtering algorithm with fixed parameters, a threshold segmentation method is used to extract seismic reflection phase axes as preliminary indicators of stratigraphic boundaries. Finally, relying on a traditional statistical regression model and combined with manually selected well logging lithology parameters, the preliminary indicated boundaries are fitted to obtain the stratigraphic boundary prediction results.

[0004] The existing solutions have obvious technical limitations: because they only use fixed algorithms to process single-type seismic features and do not establish an adaptation mechanism with the actual geological laws of the target area (such as stratigraphic contact relationships and lithological sequence distribution), the complementary information of multi-source data (seismic and well logging) cannot be effectively integrated. In areas with complex geological conditions (such as diverse stratigraphic contact relationships and complex vertical transitions of lithology), the predicted stratigraphic boundaries are prone to spatial continuity breaks or large deviations from the actual geological envelope boundaries, making it difficult to meet the needs of high-precision geological exploration. Summary of the Invention

[0005] (I) Purpose of the Invention

[0006] The main objective of this patent is to provide a method for identifying stratigraphic boundaries in geological envelopes based on artificial intelligence, so as to overcome or alleviate the aforementioned problems in the prior art.

[0007] (II) Technical Solution

[0008] This invention provides a method for identifying stratigraphic boundaries in geological envelopes based on artificial intelligence, comprising:

[0009] Step 1: Determine the seismic exploration data and well logging data for the target work area;

[0010] Step 2: Extract seismic reflection features from seismic exploration data based on the trained first neural network model, and simultaneously extract well logging sequence features from well logging data using the trained second neural network model;

[0011] Step 3: Generate a geological characteristic profile of the target work area based on seismic reflection characteristics and well logging sequence characteristics;

[0012] Step 4: Based on the stratigraphic contact relationship constraints and lithological sequence constraints for the target work area, and according to the geological characteristic characterization, determine the implicit topological structural characterization of the geological envelope in the target work area;

[0013] Step 5: Process the implicit topological structural representation based on the trained third neural network model to predict the stratigraphic boundaries of the geological envelope in the target work area.

[0014] (III) Beneficial Effects

[0015] 1. Addressing the limitations of "fixed algorithms for processing single-type seismic features and manual selection of logging lithology parameters": Step 2 of this application uses a trained first neural network model to extract seismic reflection features from seismic exploration data, and a trained second neural network model to extract logging sequence features from logging data. Unlike traditional fixed-parameter filtering, which can only process single seismic features, and manual selection of logging parameters, which is easily influenced by subjectivity, the two neural network models can dynamically learn the extraction rules of seismic reflection features and logging sequence features through training. Without relying on fixed algorithms or manual intervention, they can more comprehensively capture multi-dimensional reflection information in seismic data and vertical sequence information in logging data, providing richer basic features for multi-source data integration and avoiding the limitations of single feature extraction.

[0016] 2. Regarding the problem of "ineffective integration of complementary information from multi-source data (seismic and well logging): Step 3 of this application clearly defines the generation of geological characteristic representations of the target work area based on seismic reflection characteristics and well logging sequence characteristics. Unlike the passive role of well logging data in traditional schemes, which is only used for auxiliary verification, this step actively combines the two features, making the macroscopic stratigraphic reflection information reflected by seismic data complementary to the microscopic lithological sequence information reflected by well logging data, directly transforming it into a geological characteristic representation containing multi-source information, effectively solving the technical bottleneck of the difficulty in integrating complementary information from multi-source data.

[0017] 3. Addressing the deficiency of "not establishing an adaptation mechanism with the actual geological laws of the target work area (such as stratigraphic contact relationships and lithological sequence distribution)": Step 4 of this application, based on the constructed stratigraphic contact relationship constraints and lithological sequence constraints of the target work area, combines geological characteristic characterization to determine implicit topological structural characterization. Unlike the general processing method of traditional schemes that deviates from the actual geological laws of the target work area, this step ensures that the generated implicit topological structural characterization conforms to the real stratigraphic contact relationships and lithological sequence distribution of the target work area by constructing geological law constraints that match the target work area. This avoids boundary prediction deviations caused by deviating from actual geological laws and lays a geologically realistic foundation for subsequent accurate prediction of stratigraphic boundaries.

[0018] 4. Regarding the problem that "the predicted stratigraphic boundaries in complex geological areas are prone to spatial continuity breaks or large deviations": Step 5 of this application processes the implicit topological structural representation based on the trained third neural network model to predict stratigraphic boundaries. Unlike the limitations of traditional statistical regression models, which can only perform simple fitting of preliminary boundaries, the third neural network model can learn the spatial distribution law of stratigraphic boundaries more accurately based on the implicit topological structural representation that conforms to the geological laws of the target work area. In areas with complex geological conditions (such as diverse stratigraphic contact relationships and complex vertical transitions of lithology), it can make the predicted stratigraphic boundaries have high spatial continuity and low deviation from the actual geological envelope boundaries. This effectively solves the shortcomings of traditional schemes in boundary prediction in complex areas and meets the needs of high-precision geological exploration in the background technology. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for identifying stratigraphic boundaries in a geological envelope based on artificial intelligence, according to an embodiment of this application. Detailed Implementation

[0020] Figure 1 This is a flowchart illustrating a method for identifying stratigraphic boundaries in a geological envelope based on artificial intelligence, according to an embodiment of this application. Figure 1 As shown, it includes:

[0021] Step 1: Determine the seismic exploration data and well logging data for the target work area;

[0022] Step 2: Extract seismic reflection features from seismic exploration data based on the trained first neural network model, and simultaneously extract well logging sequence features from well logging data using the trained second neural network model;

[0023] Step 3: Generate a geological characteristic profile of the target work area based on seismic reflection characteristics and well logging sequence characteristics;

[0024] Step 4: Based on the stratigraphic contact relationship constraints and lithological sequence constraints for the target work area, and according to the geological characteristic characterization, determine the implicit topological structural characterization of the geological envelope in the target work area;

[0025] Step 5: Process the implicit topological structural representation based on the trained third neural network model to predict the stratigraphic boundaries of the geological envelope in the target work area.

[0026] Optionally, step 1 involves determining the seismic exploration data and well logging data for the target work area, specifically including:

[0027] Step 1.1: Perform adaptive noise suppression processing on the original seismic exploration data of the target work area to generate denoised seismic exploration data;

[0028] Step 1.2: Perform depth correction and outlier removal on the original logging data of the target work area to generate logging data correction results;

[0029] Step 1.3: Perform spatial coordinate matching on the denoised seismic exploration data and the corrected well logging data to generate seismic exploration data and well logging data for the target work area.

[0030] Preferably, the specific implementation process of step 1.2 is as follows: The processing object is the original logging data of the target work area (this data is stored in the form of a vertical sequence, and each data point contains a "original recorded depth" field and a "physical parameter value" field, where the "physical parameter value" field covers parameters reflecting the differences in formation lithology, such as resistivity, sonic transit time, and density. There are two types of problems in the data: one is that the "original recorded depth" deviates from the actual formation depth (due to wellbore inclination), and the other is that the "physical parameter value" has abnormal jumps (due to instrument interference or measurement error)). First, depth correction is performed: the wellbore trajectory data of the target work area is obtained (this data records the inclination angle and azimuth angle of each depth point of the wellbore, reflecting the degree of deviation between the actual wellbore direction and the vertical well), and a depth correction matrix is ​​constructed—the rows of this matrix represent the "original recorded depth point numbers" of the original logging data. (Each number corresponds to an original recorded depth value), the column represents the "actual depth point number" (each number corresponds to an actual formation depth value), and the element at the intersection of the matrix row and column is the "mapping coefficient from original recorded depth to actual depth" (obtained through wellbore trajectory geometry calculation; for example, in a deviated well scenario, the mapping coefficient = cos(wellbore inclination angle), and the actual depth = original recorded depth × mapping coefficient). Substitute each "original recorded depth value" in the original logging data into the depth correction matrix, calculate the corresponding "actual depth value" through matrix operations, and replace the "original recorded depth" field in the original logging data with the "actual depth value" to generate depth-corrected logging data (the depth field of this data is consistent with the actual formation depth, solving the depth deviation problem caused by relying solely on instrument-recorded depth in the traditional method, and laying the foundation for subsequent spatial matching with seismic data).

[0031] Next, outlier removal is performed: Using the "physical parameter value sequence" (such as resistivity sequence, sonic transit time sequence, each sequence arranged in order of actual depth) of the depth-corrected logging data as the processing object, a vertical statistical range is calculated for each physical parameter sequence (statistical range = mean of the parameter sequence ± 3 times the standard deviation, this range covers the reasonable fluctuation range of parameters under normal formation conditions); an outlier determination vector is constructed—the length of this vector is consistent with the "number of actual depth points" in the depth-corrected logging data. If the "physical parameter value" of an actual depth point exceeds the statistical range of the corresponding parameter sequence, the element at the corresponding position in the vector is marked as "outlier"; otherwise, it is marked as "normal"; the outlier determination vector... For depth points marked as "abnormal" in the logging data, the "interpolated physical parameter value" of the anomalous point is calculated using the adjacent "normal" depth points above and below it (usually 2-3 consecutive normal points are selected above and below). (By fitting the parameter change trend of adjacent normal points, the interpolation result is ensured to conform to the gradual change law of strata lithology). The original anomalous value is replaced with the "interpolated physical parameter value". Finally, the well logging data correction result is generated (the "actual depth" of this data is accurate, the "physical parameter value" has no abnormal jumps and is vertically continuous. Unlike the traditional processing method that only deletes anomalous values, this step avoids subsequent feature extraction faults caused by missing data through interpolation and ensures the vertical integrity of the well logging data).

[0032] Preferably, in the specific technical implementation of step 1.3, the processing objects are the denoised seismic exploration data and the corrected well logging data (wherein, the denoised seismic exploration data is three-dimensional grid data, and its coordinate system is "lateral survey line X-longitudinal survey line Y-depth Z", with each grid point recording "seismic amplitude value"; the corrected well logging data is vertical point data, and its coordinate system is "wellhead lateral X-wellhead longitudinal Y-actual depth Z", with each depth point recording "actual depth" and "physical parameter value". The two types of data have different coordinate systems due to different acquisition methods, and spatial consistency needs to be achieved through matching). First, a coordinate mapping table is constructed: using the unified spatial coordinate system of the target work area (such as the Gaussian plane coordinate system + elevation and depth coordinate system) as the benchmark, the "three-dimensional grid coordinate set" of the denoised seismic exploration data is extracted (each element is the "lateral survey line X value - depth Z" of a single grid point). The wellbore coordinate set consists of "longitudinal Y-value - depth Z-value" and "wellbore coordinate set" (each element being "wellhead lateral X-value - wellhead longitudinal Y-value - actual depth Z-value" for a single depth point). For each depth point in the wellbore coordinate set, the spatial Euclidean distance between it and all grid points in the three-dimensional grid coordinate set is calculated. Grid points with a distance less than 1 / 2 of the seismic data sampling interval are identified as "matching grid points" (because grid points within this distance range are spatially closest to the wellbore depth points, ensuring the rationality of parameter assignment). A coordinate mapping table is constructed—the rows of this table represent the "depth point number" of the wellbore data correction results, and the columns represent the "grid point number" of the seismic exploration denoised data. Each cell in the table records the "matching grid point number" and the "corresponding spatial Euclidean distance" (the smaller the distance, the stronger the spatial correlation between the grid point and the wellbore depth point).

[0033] Subsequently, parameter assignment and data integration are performed: Based on the coordinate mapping table, the "physical parameter values" (such as resistivity and sonic transit time) of each depth point in the well logging data correction results are assigned to the "physical parameter field" of the corresponding "matching grid point" according to the "spatial Euclidean distance weighting" rule (the weighting rule is: the weight of a matching grid point = 1 / the spatial Euclidean distance between the grid point and the well logging depth point, and the total weight is normalized to 1 to ensure that the closer the grid point is, the higher the proportion of parameter assignment), generating coordinate-matched well logging data (this data uses the three-dimensional grid of the seismic exploration denoised data as a carrier to achieve accurate mapping of well logging physical parameters to seismic space); the "seismic amplitude" of the seismic exploration denoised data is then... After the "value" is matched with the coordinate, the "physical parameter value" of the well logging data is integrated one by one according to the "three-dimensional grid coordinate" (that is, the "seismic amplitude value" and "physical parameter value" are retained at the same grid point. If a grid point does not have a corresponding "matching grid point", only the "seismic amplitude value" is retained. The relevant information is supplemented in the subsequent feature fusion in step 3). Finally, the seismic exploration data and well logging data of the target work area are generated (this data is a multi-source data set with completely consistent spatial coordinates, which solves the fusion error problem caused by coordinate disconnection in traditional multi-source data. Its spatial consistency provides a necessary premise for the correlation extraction of "seismic reflection features" and "well logging sequence features" in step 2, ensuring that the two types of features correspond to the same stratigraphic unit when the feature is fused in the future).

[0034] Optionally, step 1.1 involves adaptive noise suppression processing of the original seismic exploration data for the target area to generate denoised seismic exploration data, specifically including:

[0035] Step 1.1.1: Divide the original seismic exploration data of the target work area into spatially continuous windows, calculate the amplitude standard deviation, frequency distribution characteristics and amplitude variation rate between adjacent windows for each window, and determine the statistical characteristics of seismic data windowing accordingly.

[0036] Step 1.1.2: Based on the windowed statistical characteristics of seismic data, and combined with pre-labeled valid reflection signal samples and noise samples, the original seismic exploration data of the target work area is judged window by window for signal type determination. Dynamic weighted filtering is applied to windows marked as noise, and the original features are retained for windows marked as valid signals, so as to generate denoised seismic exploration data.

[0037] Preferably, the specific implementation process of step 1.1.1 is as follows: The processing object is the original seismic exploration data of the target work area (this data is stored in the form of a three-dimensional data volume, and each data point records the seismic amplitude value corresponding to the spatial location of "lateral survey line X-longitudinal survey line Y-depth Z". The data contains a mixture of effective reflection signals of strata (corresponding to the vibration response of continuous strata interfaces, with high spatial continuity), instrument electronic noise (randomly distributed, without spatial regularity), and surface environmental interference noise (such as surface waves, with concentrated energy but significant differences in spatial continuity from the effective signal)). First, according to the spatial sampling pattern of the original seismic exploration data (preset sampling intervals along the X-axis lateral survey line, Y-axis longitudinal survey line, and Z-axis depth direction), spatial continuous window division is performed - the windows are distributed in a continuous order in the X, Y, and Z dimensions, and adjacent windows retain overlapping areas in each dimension (the overlap ratio is determined based on the sampling interval, such as 50%, the design purpose is to avoid the loss of features due to the breakage of effective reflection signals at the window boundary), and the divided window data is obtained (each window data contains all seismic amplitude values ​​within the corresponding spatial range, and the window numbers are arranged in XYZ order for easy subsequent feature tracing).

[0038] Next, multi-dimensional feature calculations are performed on the data of each divided window: First, the amplitude standard deviation is calculated (the standard deviation of all earthquake amplitude values ​​within the window is statistically analyzed, reflecting the degree of signal fluctuation within the window—the effective reflected signal has a lower degree of fluctuation due to the continuity of the strata, while the noise component has a higher degree of fluctuation), resulting in a window amplitude standard deviation vector (the length of this vector is consistent with the total number of data in the divided windows, and each element in the vector corresponds to the amplitude standard deviation of a window, with the element subscript corresponding one-to-one with the window number); Second, the frequency distribution characteristics are calculated through time-domain to frequency-domain transformation (Fourier transform is performed on the window data to convert the time-domain amplitude signal into a frequency-domain energy signal, and the highest energy dominant frequency value and the bandwidth value with energy greater than 50% of the dominant frequency energy are statistically analyzed in the frequency domain), resulting in the window frequency distribution. The first step is to obtain the characteristic matrix (the rows of this matrix correspond to the window numbers, and the columns are "dominant frequency value" and "bandwidth value", respectively. The elements at the intersection of the matrix are the frequency parameters of the corresponding window - the effective reflected signal has a concentrated dominant frequency and a narrow bandwidth, while the noise component has a dispersed dominant frequency and a wide bandwidth); the second step is to calculate the amplitude change rate between adjacent windows (for each window, select its six adjacent windows above, below, left, right, front, and back, and calculate the ratio of the difference between the average amplitude of the adjacent windows and the average amplitude of the current window, i.e., (average amplitude of adjacent windows - average amplitude of the current window) / average amplitude of the current window), thus obtaining the amplitude change rate matrix between adjacent windows (the rows of this matrix correspond to the window numbers, and the columns correspond to the six adjacent directions. The matrix elements are the amplitude change rates of the corresponding directions - the effective reflected signal has a low rate of change due to the continuous change of the strata, while the noise component has a high rate of change).

[0039] Finally, the window amplitude standard deviation vector, window frequency distribution feature matrix, and amplitude change rate matrix between adjacent windows are associated and integrated according to the window number to ensure that the three types of features in the same window correspond one-to-one, and generate seismic data window statistical features (this feature fully records the signal fluctuation, frequency distribution, and spatial variation characteristics of each window, providing a quantitative basis for the signal type determination in the subsequent step 1.1.2, and each feature component can directly reflect the difference between effective signal and noise within the window).

[0040] Preferably, in the specific technical implementation of step 1.1.2, the processing objects are the seismic data windowing statistical features and pre-labeled effective reflection signal samples and noise samples (wherein, both effective reflection signal samples and noise samples are selected from seismic profiles that have been verified by drilling in the target work area—effective reflection signal samples correspond to the seismic phase axis region at the stratigraphic interface revealed by drilling, and the seismic data windowing statistical features of this region are extracted as a reference; noise samples correspond to the blank area without stratigraphic reflection or the known interference source area revealed by drilling, and the seismic data windowing statistical features of this region are extracted as a reference); firstly, a signal type determination benchmark is constructed: for the seismic data windowing statistical features of effective reflection signal samples, the mean range of each feature component is calculated (such as the mean amplitude standard deviation ± 1 standard deviation, the mean frequency ± 1 standard deviation), to obtain the effective signal feature benchmark range; for the seismic data windowing statistical features of noise samples, the mean range of each feature component is also calculated to obtain the noise feature benchmark range;

[0041] Next, the window-by-window signal type determination is performed: using each window feature in the seismic data window statistical features as a processing unit, the amplitude standard deviation, dominant frequency value, and amplitude change rate between adjacent windows of the window are matched with the effective signal feature reference range and the noise feature reference range, respectively. The number of feature components that meet the effective signal feature reference range and the number of feature components that meet the noise feature reference range of the window are counted. If the number of components that meet the effective signal feature reference range is greater, the window is marked as "effective signal"; if the number of components that meet the noise feature reference range is greater, the window is marked as "noise". A window signal type identification vector is generated (the length of this vector is consistent with the number of windows, and the elements are "effective signal" or "noise", with the element subscripts corresponding one-to-one with the window number).

[0042] Then, dynamic weighted filtering is performed on the windows identified as "noise": based on the window signal type identifier vector, all windows identified as "noise" are filtered out, and the noise ratio of each noise window is calculated (the proportion of amplitude values ​​in the window that exceed the amplitude standard deviation range of the effective signal characteristic reference range is statistically analyzed—the higher the ratio, the higher the noise content in the window); the filtering weight is dynamically adjusted according to the noise ratio (the noise ratio is positively correlated with the filtering weight, for example, the filtering weight is 0.9 when the noise ratio is 80%, and the filtering weight is 0.5 when the noise ratio is 50%, the design reason is to avoid the fixed weight filtering from excessively suppressing the effective signal of windows with low noise ratios); for each noise window, a Gaussian filter kernel matched with the window size is used, and filtering is applied according to the dynamically adjusted filtering weight (the filtering weight of the filter kernel is realized through Gaussian function distribution, the larger the weight, the stronger the smooth suppression of the amplitude values ​​in the window), and the filtered noise window data is obtained;

[0043] For windows marked as "effective signal" in the window signal type identifier vector, the original amplitude characteristics are directly retained to avoid the effective reflection signal being weakened by filtering. Finally, the filtered noise window data and the effective signal window data with retained original characteristics are reintegrated according to the spatial coordinate order of the original three-dimensional data volume to ensure that the amplitude value at each spatial location corresponds to the corrected window data, generating the denoised seismic exploration data (in this data, the spatial continuity of the effective reflection phase axis of the strata is relatively high, and the energy of the noise component is significantly suppressed—unlike the limitation of traditional fixed parameter filtering which can only optimize for a single noise type, this step adapts the noise content of different windows through dynamic weights, which improves the retention of effective signals while suppressing noise, and the denoised seismic exploration data will be used as the basic seismic end data for data spatial coordinate matching in step 1.3).

[0044] Optionally, step 2 involves extracting seismic reflection features from seismic exploration data based on the trained first neural network model, and simultaneously extracting well logging sequence features from well logging data using the trained second neural network model. Specifically, this includes:

[0045] Step 2.1: Based on the first neural network model that has been trained, perform multi-scale feature extraction on the denoised seismic exploration data to obtain the initial seismic reflection features;

[0046] Step 2.2: Based on the trained second neural network model, extract vertical sequence features from the well logging data to generate initial well logging sequence features;

[0047] Step 2.3: Remove redundant features that are not related to the formation boundary from the initial seismic reflection features and initial well logging sequence features to generate the selected seismic reflection features and selected well logging sequence features.

[0048] Optionally, step 2.1, based on the trained first neural network model, involves multi-scale feature extraction of the denoised seismic exploration data to obtain initial seismic reflection features, specifically including:

[0049] Step 2.1.1: Based on the multi-scale convolutional feature extraction layer in the first neural network model that has been trained, perform feature extraction at different spatial scales on the denoised seismic exploration data to generate a preliminary multi-scale seismic feature map.

[0050] Step 2.1.2: Based on the cross-scale feature attention fusion layer in the first neural network model that has been trained, the attention weights of the preliminary multi-scale seismic feature map are calculated and the features are fused to generate the fused seismic feature map.

[0051] Step 2.1.3: Based on the residual noise suppression layer in the first neural network model that has been trained, perform residual calculation on the fused seismic feature map to generate a denoised seismic feature map.

[0052] Step 2.1.4: Based on the feature dimension compression and filtering layer in the first neural network model that has been trained, the denoised seismic feature map is subjected to dimension compression to remove redundant feature dimensions and effective feature filtering to retain features that are strongly correlated with the layer boundary, so as to generate initial seismic reflection features.

[0053] Preferably, the specific implementation process of step 2.1.1 is as follows: The processing object is the denoised seismic exploration data (this data is three-dimensional grid data, each grid point records the denoised seismic amplitude value corresponding to the spatial location of "lateral survey line X - longitudinal survey line Y - depth Z". The data has high spatial continuity of effective stratum reflection signals and low residual noise energy, providing a high-quality data foundation for feature extraction); relying on the multi-scale convolutional feature extraction layer in the trained first neural network model (this layer has three different sizes of three-dimensional convolution kernels built in, designed to capture seismic reflection features at different spatial scales respectively - small-sized convolution kernels are suitable for local reflection feature extraction of fine stratigraphic interfaces, medium-sized convolution kernels are suitable for medium-scale reflection feature extraction of stratigraphic segments, and large-sized convolution kernels are suitable for macroscopic reflection feature extraction of regional stratigraphic distribution), multi-scale feature extraction is performed on the denoised seismic exploration data:

[0054] First, a small-sized three-dimensional convolution kernel is used to perform convolution operations on the denoised seismic exploration data (the convolution kernel slides along the XYZ three-dimensional direction, and the sliding step size is matched with the seismic data sampling interval. Edge features are preserved during the operation to avoid loss of boundary information). The detailed features of the reflection phase axis at the stratigraphic interface (such as abrupt change points of reflection amplitude and inflection points of phase axis continuity) are extracted to obtain a small-scale seismic feature map.

[0055] Next, a medium-sized three-dimensional convolution kernel is used to perform convolution operations on the denoised seismic exploration data (the convolution kernel has a larger coverage area than the small-sized convolution kernel, and the sliding step size is the same as the small-sized kernel) to extract the reflection structure features within the stratigraphic segment (such as the combined reflection mode of multiple adjacent stratigraphic interfaces and the reflection energy distribution corresponding to stratigraphic thickness changes) to obtain a mesoscale seismic feature map.

[0056] Finally, a large-size three-dimensional convolution kernel is used to perform convolution operations on the denoised seismic exploration data (the convolution kernel has the largest coverage and is suitable for regional scale feature extraction) to extract the macroscopic distribution features of the strata in the target work area (such as the overall dip direction of the strata and the continuity trend of the reflection layer in the region) and obtain a large-scale seismic feature map.

[0057] Small-scale, medium-scale, and large-scale seismic feature maps are linked and integrated according to spatial coordinates (ensuring that features at different scales at the same spatial location correspond one-to-one, facilitating subsequent fusion) to generate a preliminary multi-scale seismic feature map (this feature map contains seismic reflection features at three scales, and each scale feature can reflect stratigraphic properties from different dimensions, which is different from the limitation of traditional single-scale feature extraction, which can only capture local or macroscopic single information).

[0058] Preferably, in the specific technical implementation of step 2.1.2, the processing object is a preliminary multi-scale seismic feature map (this feature map includes small-scale seismic feature maps, mesoscale seismic feature maps, and large-scale seismic feature maps; the spatial coordinate system of the three types of sub-maps is consistent, but the feature dimensions and the geological information reflected differ—small-scale features focus on interface details, mesoscale features focus on intra-segment structure, and large-scale features focus on regional distribution); relying on the cross-scale feature attention fusion layer in the trained first neural network model, the attention weights of the three types of sub-maps in the preliminary multi-scale seismic feature map are calculated respectively:

[0059] Using the labeled stratigraphic boundary samples of the target work area as a reference (the samples contain features of various scales corresponding to the known stratigraphic boundary locations), the similarity between each feature point in the preliminary multi-scale seismic feature map and the features of the stratigraphic boundary samples is calculated (the higher the similarity, the stronger the correlation between the feature point and the stratigraphic boundary). Attention weights are assigned according to the similarity (feature points with stronger correlation have higher weights, and those with weaker correlation have lower weights), generating small-scale feature attention weight maps, medium-scale feature attention weight maps, and large-scale feature attention weight maps.

[0060] Subsequently, the three types of sub-maps are weighted point by point with the corresponding attention weight maps (i.e., sub-map feature value × corresponding weight value) to strengthen features strongly correlated with the layer boundary and weaken irrelevant and redundant features, resulting in weighted small-scale earthquake feature maps, weighted meso-scale earthquake feature maps, and weighted large-scale earthquake feature maps.

[0061] Finally, feature fusion is performed on the three weighted submaps (the spatial coordinates are kept aligned during the fusion process, and the three weighted feature values ​​at the same spatial location are spliced ​​together according to the channel dimension to form a multi-channel feature vector) to generate a fused seismic feature map (this feature map integrates key reflection features at different scales and highlights the relevant features of the layer boundary through an attention mechanism, solving the problem of effective feature dilution caused by traditional simple feature splicing).

[0062] Preferably, in a scenario, when step 2.1.3 is specifically implemented, the processing object is the fused seismic feature map (this feature map is a multi-channel three-dimensional feature map, each channel corresponds to a weighted scale feature, and there may still be a small amount of residual noise superimposed on the effective reflection features in the feature map—such as abnormal fluctuations in feature values ​​in local areas, which need to be further suppressed to improve feature purity); relying on the residual noise suppression layer in the trained first neural network model (this layer contains residual connections and noise suppression branches, and the design logic is to separate effective features and residual noise through residual operations to avoid excessive suppression of effective features by traditional denoising methods), residual denoising processing is performed on the fused seismic feature map:

[0063] First, the fused seismic feature map is input into the noise suppression branch (this branch learns the residual noise distribution pattern in the fused seismic feature map through convolution and activation function operations, and outputs a noise feature map—the spatial dimension of the noise feature map is the same as that of the fused seismic feature map). Figure 1 (Each feature point value corresponds to the predicted residual noise intensity);

[0064] Next, the residual operation relationship is constructed: based on the fused seismic feature map, the corresponding feature values ​​in the noise feature map are subtracted (subtracting point by point to ensure that the effective features and noise at the same spatial location are accurately separated) to obtain the preliminary denoised feature map;

[0065] Finally, the effective feature components (such as feature points strongly correlated with layer boundaries) in the fused seismic feature map are directly transferred to the preliminary denoised feature map through residual connection to compensate for the weak effective features that may be lost in the residual calculation, and generate a denoised seismic feature map (the residual noise energy in this feature map is significantly reduced, and the integrity and clarity of the effective reflection features are high, providing high-purity feature data for subsequent dimensionality compression and screening).

[0066] Preferably, in the specific technical implementation of step 2.1.4

[0067] The object being processed is the denoised seismic feature map (this feature map is a multi-channel 3D feature map, the number of channels is related to the fused seismic features). Figure 1To address the redundancy in feature dimensions in some channels—for example, features from different channels reflecting the same geological information—feature effectiveness needs to be improved through compression and filtering. Based on the feature dimension compression and filtering layer in the trained first neural network model (this layer integrates a dimension compression module and a feature filtering module to achieve integrated "compression-filtering" processing, avoiding feature loss caused by traditional staged processing), feature optimization processing is performed on the denoised seismic feature map.

[0068] First, the dimension compression module is activated: dimensionality reduction operation is performed on the multi-channel features of the denoised seismic feature map (by analyzing the variance contribution of each channel feature, channels with high variance contribution are retained, and redundant channels with extremely low variance contribution are removed—variance contribution reflects the information content of channel features; the higher the contribution, the richer the geological information contained), resulting in a compressed seismic feature map (this feature map has fewer channels than the denoised seismic feature map, eliminating meaningless redundant dimensions and reducing the complexity of subsequent processing);

[0069] Next, the feature filtering module is activated: taking the prior geological laws of the stratigraphic boundary of the target work area (such as the variance range of reflection features at the stratigraphic boundary and the range of differences with adjacent stratigraphic features) as a reference, the correlation between each feature point in the compressed seismic feature map and the stratigraphic boundary is calculated (the correlation is quantified by the matching of feature values ​​with prior laws; the higher the matching degree, the stronger the correlation).

[0070] Finally, the feature points with strong correlation in the compressed seismic feature map are retained, while the feature points with weak correlation are removed (the removal threshold is set based on prior geological laws to ensure that the retained features are strongly correlated with the stratigraphic boundaries), and the initial seismic reflection features are generated (this feature is a three-dimensional spatially distributed feature set, each feature point corresponds to a clear "spatial coordinate-reflection feature value" information, directly providing the processing object for the redundant feature removal in step 2.3, and the feature purity is high, which can accurately reflect the seismic reflection attributes related to the stratigraphic boundaries).

[0071] The first neural network model training aims to accurately extract seismic reflection features strongly correlated with stratigraphic boundaries. Training data consists of seismic exploration data samples (including denoised seismic exploration data) from similar target work areas and corresponding labeled stratigraphic boundary samples (labeled samples include known stratigraphic boundary locations and corresponding seismic reflection feature labels). During training, the denoised seismic exploration data samples are first input into the model's multi-scale convolutional feature extraction layer. Different sized convolutional kernels are used to initially learn reflection features at local, intra-segment, and regional scales, generating preliminary multi-scale seismic feature map samples. Next, a cross-scale feature attention fusion layer is input. Using the labeled samples as a reference, the model learns the correlation weights between features at different scales and stratigraphic boundaries, optimizing the feature fusion strategy to highlight boundary-related features in the fused seismic feature map. Subsequently, a residual noise suppression layer learns residual mapping to separate residual noise from the fused feature samples, generating denoised seismic feature map samples. Finally, a feature dimension compression and filtering layer learns redundancy rules and effective feature selection criteria for feature dimensions, retaining features strongly correlated with stratigraphic boundaries. During training, the mean squared error loss function is used. The initial seismic reflection feature samples output by the model are compared with the reflection feature labels in the labeled samples. The parameters of each convolution kernel, attention weights and residual mapping parameters are adjusted iteratively through backpropagation until the loss function value converges to the preset threshold, thus completing the model training and enabling the model to extract high-quality seismic reflection features from seismic data.

[0072] Optionally, step 2.2, based on the trained second neural network model, extracts vertical sequence features from the logging data to generate initial logging sequence features, specifically including:

[0073] Step 2.2.1: Based on the vertical window serialization layer in the trained second neural network model, perform vertical window partitioning and sequence transformation on the logging data to generate a set of vertical logging sequence fragments;

[0074] Step 2.2.2: Based on the sequence residual denoising layer in the trained second neural network model, perform residual operations on the vertical logging sequence fragment set to generate a denoised vertical logging sequence fragment set.

[0075] Step 2.2.3: Based on the multi-feature attention enhancement layer in the trained second neural network model, calculate the attention weights for the denoised vertical logging sequence fragment set to generate the attention-enhanced logging sequence feature set;

[0076] Step 2.2.4: Based on the sequence feature filtering layer in the trained second neural network model, perform feature correlation analysis and filtering on the attention-enhanced logging sequence feature set to generate initial logging sequence features.

[0077] Preferably, the specific implementation process of step 2.2.1 is as follows: the processing object is well logging data (this data is the target area well logging data generated in step 1.3, which includes the spatial coordinates of "lateral logging line X - longitudinal logging line Y - actual depth Z" and the corresponding physical parameter sequence of "resistivity - sonic transit time - density". The physical parameters are continuously distributed along the actual depth direction, and the parameter value of each depth point reflects the lithological properties of the corresponding stratum. It is necessary to capture the vertical lithological variation law through serialization processing); relying on the vertical window serialization layer in the trained second neural network model, the vertical window division is first performed:

[0078] Based on the actual depth sampling interval of the logging data (e.g., each preset depth interval along the Z-axis is a basic unit), continuous vertical windows are divided. The windows are distributed only along the actual depth direction, and the overlapping area of ​​adjacent windows is retained (the overlap ratio is determined based on the sampling interval, such as 50%, the design purpose is to avoid the loss of the sequence characteristics of the lithological transition section due to window breakage). The window numbers are arranged sequentially from shallow to deep according to the actual depth to obtain the divided vertical windows (each window contains all physical parameter values ​​within the corresponding depth range, such as a window covering resistivity, sonic transit time, and density data of "depth Z1-Z2").

[0079] Next, sequence transformation is performed: the physical parameter values ​​in each divided vertical window are arranged in ascending order of actual depth to form a one-dimensional parameter sequence (the sequence length is the same as the number of depth points in the window, and the sequence elements are a combination vector of "resistivity value - acoustic time difference value - density value" in sequence), and the window number is associated with the corresponding one-dimensional parameter sequence to ensure that each sequence can be traced back to the original depth range;

[0080] Finally, all the one-dimensional parameter sequences of the divided vertical windows are integrated to generate a set of vertical logging sequence fragments (this set of fragments has a three-dimensional structure, with dimensions of "window number - sequence length - parameter type"). Each fragment can reflect the changes in vertical lithological parameters within a local depth range, which is different from the problem of local lithological details being diluted due to traditional full-depth sequence processing.

[0081] Preferably, in the specific technical implementation of step 2.2.2, the processing object is a set of vertical logging sequence segments (this set of segments contains multiple vertical logging sequence segments, and there may still be a small amount of measurement noise in the segments—such as abnormal jumps in resistivity values ​​at a certain depth point, which contradict the lithological variation patterns at adjacent depth points, and denoising is required to preserve the true vertical lithological sequence characteristics); relying on the sequence residual denoising layer in the trained second neural network model, a residual processing branch is first constructed:

[0082] The vertical logging sequence fragment set is input into the noise extraction branch (this branch learns the distribution pattern of noise in the fragments through convolution and recurrent activation function operations—such as the features of randomly changing parameter values, and outputs a noise sequence fragment set. The dimension of the noise sequence fragment set is consistent with that of the vertical logging sequence fragment set, and each element corresponds to the predicted noise value).

[0083] Next, residual calculation is performed: using the vertical logging sequence fragment set as the base sequence, the noise value corresponding to the noise sequence fragment set (i.e., the base sequence parameter value - noise value) is subtracted element by element to separate the preliminary denoised sequence fragment set (this fragment set has eliminated most of the random noise, but may lose slight lithological transition characteristics);

[0084] Finally, residual connection is used to compensate for effective features: features related to lithological transitions (such as the gradual trend of parameter values ​​at adjacent depth points) in the vertical logging sequence fragment set are directly transferred to the preliminary denoised sequence fragment set to correct the feature deviation caused by residual calculation and generate a denoised vertical logging sequence fragment set (the physical parameter sequence of this fragment set has high vertical continuity and clear lithological change law, solving the problem that traditional fixed threshold denoising is prone to accidentally deleting weak lithological features).

[0085] Preferably, in a scenario, when step 2.2.3 is specifically implemented, the processing object is a set of denoised vertical logging sequence segments (this set of segments contains multiple denoised vertical logging sequence segments, and different physical parameters in the segments have different indicative meanings for lithological boundaries—for example, abrupt changes in resistivity often correspond to sandstone-mudstone boundaries, and gradual changes in acoustic transit time often correspond to changes in the density of the same lithology; attention mechanisms are needed to strengthen parameter features strongly correlated with lithological boundaries); relying on the multi-feature attention enhancement layer in the trained second neural network model, the attention weights are first calculated:

[0086] Using the lithological boundary samples already marked in the target work area as a reference (the samples contain the physical parameter sequence characteristics corresponding to the known lithological boundary depth points), the correlation between each sequence element in the denoised vertical logging sequence fragment set and the lithological boundary sample characteristics is calculated (the correlation is quantified by matching the parameter value change rate with the sample change rate; the higher the matching degree, the stronger the correlation between the element and the lithological boundary).

[0087] Attention weights are assigned based on relevance (the stronger the correlation, the higher the weight of the sequence element, and the weaker the correlation, the lower the weight), and a parameter attention weight matrix is ​​generated (the dimension of this matrix is ​​consistent with the denoised vertical logging sequence fragment set, the rows correspond to the window number, the columns correspond to the sequence length, the channels correspond to the parameter type, and the elements are the attention weights at the corresponding positions).

[0088] Next, weighted enhancement is performed: the set of vertical logging sequence fragments after noise reduction is multiplied element by element with the parameter attention weight matrix (i.e., sequence parameter value × corresponding weight value), which enhances features related to lithological boundaries such as resistivity abrupt changes and density sudden changes, while weakening irrelevant features such as small fluctuations in sonic transit time, to obtain the set of vertical logging sequence fragments after weighting.

[0089] Finally, the weighted vertical logging sequence fragment set was integrated to generate an attention-enhanced logging sequence feature set (this feature set highlights the key parameter features of vertical lithological changes, laying the foundation for subsequent screening of sequence features related to stratigraphic boundaries).

[0090] Preferably, in the specific technical implementation of step 2.2.4, the processing object is the attention-enhanced logging sequence feature set (this feature set contains multiple attention-enhanced logging sequence features, some of which are not related to formation boundaries—such as small density fluctuation features within the same lithological section, which need to be screened to improve feature effectiveness); relying on the sequence feature screening layer in the trained second neural network model, feature correlation analysis is first performed:

[0091] Using the labeled stratigraphic boundary samples of the target work area as a reference (the samples contain sequence features corresponding to known stratigraphic boundary depth points), calculate the cosine similarity between each feature in the attention-enhanced logging sequence feature set and the stratigraphic boundary sample features (the higher the similarity, the stronger the correlation between the feature and the stratigraphic boundary), and set a similarity threshold (the threshold is determined based on sample statistics; if the similarity is greater than the preset threshold, it is judged as a "correlated feature"; otherwise, it is a "redundant feature").

[0092] Next, feature filtering is performed: retain the associated features with similarity greater than the threshold in the feature set of the well logging sequence after attention enhancement, and remove redundant features—such as removing the resistivity stationary sequence features within the same sandstone layer, and retaining the resistivity abrupt sequence features at the sandstone-mudstone interface.

[0093] Finally, the filtered associated features are integrated according to "window number - actual depth" to ensure that each feature can be traced back to its original spatial location, generating an initial logging sequence feature (this feature is a mapping set of "spatial coordinates - vertical lithological change sequence", which only contains vertical sequence features strongly correlated with stratigraphic boundaries, directly providing processing objects for redundant feature removal in step 2.3, and the features are highly targeted and can accurately reflect the lithological sequence changes at the stratigraphic boundaries).

[0094] Preferably, the specific implementation process of step 2.3 is as follows: the processing objects are the initial seismic reflection features and the initial well logging sequence features (wherein, the initial seismic reflection features are the three-dimensional feature set generated in step 2.1.4, each feature point contains the coordinates of "lateral survey line X - longitudinal survey line Y - depth Z" and the corresponding reflection feature value, reflecting the spatial reflection attributes of the formation interface; the initial well logging sequence features are the vertical sequence feature set generated in step 2.2.4, each feature segment contains the coordinates of "wellhead X - wellhead Y - actual depth Z" and the corresponding physical parameter sequence value, reflecting the vertical lithological change attributes of the formation, and both types of features contain redundant features that are not related to the formation boundary - such as the background noise residual features in the seismic reflection features and the stable parameter features of the same lithological segment in the well logging sequence features).

[0095] First, redundant features are removed from the initial seismic reflection features: using the labeled stratigraphic boundary samples of the target work area (the samples contain known stratigraphic boundary locations and corresponding reflection features) as a reference, the correlation degree between each feature point in the initial seismic reflection features and the stratigraphic boundary sample features is calculated (the correlation degree is quantified by the matching trend of feature value changes with the sample change trend; the higher the matching degree, the higher the correlation degree), and a seismic feature correlation degree matrix is ​​generated (the rows of this matrix correspond to the feature point numbers of the initial seismic reflection features, the columns correspond to the stratigraphic boundary sample numbers, and the elements at the intersection of the matrices are the correlation degree values ​​between the corresponding feature points and the samples); a correlation degree threshold is set (the threshold is determined based on sample statistics; if the correlation degree is greater than the preset threshold, it is judged as "features associated with stratigraphic boundaries," otherwise it is "redundant features"), feature points with a correlation degree greater than the threshold in the seismic feature correlation degree matrix are retained, redundant feature points are removed, and a preliminary filtered seismic reflection feature is generated (this feature set has removed most reflection features unrelated to stratigraphic boundaries, but still needs to be cross-validated with well logging sequence features to avoid cross-feature redundancy);

[0096] Next, redundant feature removal is performed on the initial well logging sequence features: again, using the formation boundary sample as a reference (the sample includes known formation boundary depth points and corresponding well logging parameter sequences), the similarity between each feature segment in the initial well logging sequence features and the formation boundary sample sequence is calculated (the similarity is quantified by the fit between the vertical parameter change rate and the sample change rate; the higher the fit, the higher the similarity), generating a well logging feature similarity vector (the length of this vector is consistent with the number of feature segments in the initial well logging sequence features, and the vector elements are the similarity values ​​between the corresponding feature segments and the samples, with the element subscripts corresponding one-to-one with the feature segment numbers); a similarity threshold is set (the threshold is determined based on sample statistics; if the similarity is greater than the preset threshold, it is judged as "feature associated with the formation boundary," otherwise it is "redundant feature"), retaining feature segments in the well logging feature similarity vector with similarity greater than the threshold, removing redundant feature segments, and generating the pre-screened well logging sequence features;

[0097] Finally, cross-feature redundancy verification is performed: the mutual information value between the pre-screened seismic reflection features and the pre-screened well logging sequence features is calculated (the mutual information value reflects the degree of information overlap between the two types of features; the higher the value, the more similar the geological information reflected by the two types of features, indicating redundancy), and a mutual information threshold is set (if the mutual information value is greater than the preset threshold, it is judged as "cross-feature redundancy"); for features judged as cross-feature redundancy—such as a seismic reflection feature point and a well logging sequence feature segment both reflecting the same sandstone-mudstone boundary, the feature with richer information is retained (well logging sequence features are retained first because well logging data directly reflects lithology and provides a more direct indication of stratigraphic boundaries), and another type of redundant features is removed;

[0098] The features after cross-feature redundancy verification are integrated to generate filtered seismic reflection features and filtered well logging sequence features (both types of features retain only those strongly correlated with stratigraphic boundaries and have no cross-feature information redundancy, unlike the information duplication problem that is easily caused by traditional single-class feature screening. This step significantly improves the effectiveness of features through the dual processing of "single-class screening + cross-class verification", providing high-quality input for feature fusion in the subsequent step 3).

[0099] The second neural network model training focuses on "enabling the model to efficiently extract logging sequence features reflecting vertical lithological changes." The training data consists of logging data samples from the same target work area (including logging data correction results) and corresponding labeled lithological sequence tags (labeled samples include known lithological boundary depths and corresponding logging sequence feature tags). During training, the logging data samples are first input into the vertical window serialization layer. The model learns the optimal partition size of the vertical window and sequence transformation rules, transforming the logging data samples into a set of vertical logging sequence fragments. Then, the samples are input into the sequence residual denoising layer. By learning the sequence residual mapping, measurement noise in the fragment set samples is removed, generating a denoised set of vertical logging sequence fragments. The multi-feature attention enhancement layer, based on the labeled samples, learns the correlation weights between different logging parameter samples (resistivity, sonic transit time, etc.) and lithological boundaries, enhancing the sequence features of key parameters and generating a set of attention-enhanced logging sequence features. Finally, the sequence feature selection layer learns the correlation criteria between features and lithological boundaries, selecting effective features. Training employs the cross-entropy loss function, comparing the initial well logging sequence feature samples output by the model with the lithological sequence labels in the labeled samples. The window partitioning parameters, residual mapping coefficients, and attention weights are iteratively optimized through backpropagation. Training is completed when the loss function value is stable and the model's accuracy in extracting lithological sequences meets the target, enabling the model to extract accurate vertical well logging sequence features from well logging data.

[0100] Optionally, step 3, based on seismic reflection characteristics and well logging sequence characteristics, generates a geological characteristic profile of the target work area, specifically including:

[0101] Step 3.1: Based on the spatial coordinate system of the target work area, spatial coordinate anchoring is performed on the seismic reflection features and well logging sequence features. Then, the correlation weights are calculated based on the degree of correlation between the spatially anchored seismic reflection features and well logging sequence features and the formation boundary, so as to generate a feature correlation mapping table.

[0102] Step 3.2: Based on the feature association mapping table, the seismic reflection features and well logging sequence features are weighted and fused, and contradictory features after fusion are eliminated by stratigraphic topology verification to generate a fused geological feature set;

[0103] Step 3.3: Based on the geological background of the target work area, perform spatial continuity processing on the fused geological feature set and label the geological significance of each feature to generate a geological characteristic representation of the target work area.

[0104] Preferably, the specific implementation process of step 3.1 is as follows: the processing objects are the spatial coordinate system of the target work area, the selected seismic reflection features, and the selected well logging sequence features (wherein, the spatial coordinate system of the target work area is defined as a three-dimensional rectangular coordinate system of "lateral survey line X - longitudinal survey line Y - depth Z", the X-axis corresponds to the number of the lateral exploration survey line in the work area, the Y-axis corresponds to the number of the longitudinal exploration survey line, and the Z-axis corresponds to the vertical depth of the strata, with values ​​increasing from the surface downwards; the selected seismic reflection features are the three-dimensional feature set generated in step 2.3, containing "XYZ" coordinates and reflection feature values, reflecting the spatial reflection attributes of the strata interface; the selected well logging sequence features are the vertical feature set generated in step 2.3, containing "wellhead X - wellhead Y - actual depth Z" coordinates and sequence feature values, reflecting the vertical lithological change attributes of the strata, and the two types of features need to be spatially unified through coordinate anchoring).

[0105] First, spatial coordinate anchoring is performed: for the selected seismic reflection features, their original "XYZ" coordinates are directly matched to the spatial coordinate system of the target work area, ensuring that each reflection feature point accurately corresponds to the specific spatial location of the work area, generating coordinate-anchored seismic reflection features; for the selected well logging sequence features, their "wellhead X - wellhead Y" coordinates are matched to the XY plane of the coordinate system, and their "actual depth Z" coordinates are matched to the Z-axis, realizing the alignment of the vertical sequence features with the coordinates of the three-dimensional space of the work area, generating coordinate-anchored well logging sequence features (this process solves the problem of coordinate disconnection between the two types of features due to different acquisition methods, laying a spatial foundation for subsequent fusion);

[0106] Next, the correlation weights are calculated: using the formation boundary samples of the target work area that have passed drilling verification as a reference (the samples contain "XYZ" coordinates and corresponding "seismic reflection standard features - well logging sequence standard features", such as "high amplitude continuous phase axis" and "resistivity surge" at a certain formation boundary), for each feature point in the seismic reflection features after coordinate anchoring, the similarity between its reflection feature value and the seismic reflection standard feature of the corresponding spatial location in the sample is calculated (the similarity is quantified by matching the trend of feature value changes, such as the closer the slope of the reflection amplitude change is to the standard feature, the higher the matching degree and the higher the correlation degree), and the seismic feature correlation weights are assigned according to the matching degree (the higher the correlation degree, the higher the weight value); for each feature segment in the well logging sequence features after coordinate anchoring, the similarity between its sequence feature value and the standard feature of the corresponding depth well logging sequence in the sample is calculated, and the well logging feature correlation weights are assigned.

[0107] Finally, the “feature number (including seismic / well logging identifier) ​​- spatial coordinates (XYZ) - seismic feature association weight - well logging feature association weight” are arranged and integrated by row to generate a feature association mapping table (the rows of this table correspond to the feature number, the columns correspond to the spatial coordinates, seismic feature association weight, and well logging feature association weight, and the elements at the intersection are the specific parameters of the corresponding features, ensuring that the weight of each feature can be traced back to the sample reference process and avoiding the subjectivity of weight allocation).

[0108] Preferably, in the specific technical implementation of step 3.2, the processing objects are the feature association mapping table, the seismic reflection features after coordinate anchoring, and the well logging sequence features after coordinate anchoring (the three types of objects need to work together to achieve feature fusion and verification to ensure that the fused features conform to the natural laws of the strata).

[0109] First, weighted fusion is performed: traversing each spatial coordinate (XYZ) in the feature association mapping table, if both the seismic reflection feature point after coordinate anchoring and the well logging sequence feature segment after coordinate anchoring exist under that coordinate, the fusion feature value of that coordinate is calculated according to the corresponding weight in the table as "fusion feature value = seismic reflection feature value × seismic feature association weight + well logging sequence feature value × well logging feature association weight"; if only one type of feature exists under a certain coordinate (such as only the seismic reflection feature after coordinate anchoring), then that feature value is directly used as the fusion feature value to ensure that no feature is missing; the fusion feature values ​​of all spatial coordinates are integrated according to "XYZ - fusion feature value" to generate a preliminary fusion geological feature set (this set covers most spatial locations in the work area, but may contain a small number of features that contradict the stratigraphic regularity due to feature errors);

[0110] Next, a stratigraphic topology verification is performed: the known topological relationships of the strata in the target work area are used as the verification basis (known topological relationships include vertical sequence rules, such as the fixed vertical order of "upper mudstone - middle sandstone - lower limestone"; and lateral continuity rules, such as the distribution range and continuity of the same stratum in the XY plane). The preliminary fusion geological feature set is traversed: if the stratigraphic attribute indicated by the fusion feature value of a certain spatial coordinate (such as "limestone") violates the vertical sequence rule with the attribute indicated by the fusion feature value of the coordinate directly above it (such as "sandstone"), or if a stratigraphic feature is suddenly interrupted in lateral extension without tectonic fracture interpretation, it is marked as a contradictory feature; all contradictory features are removed, and the remaining features after removal are reordered according to the "XYZ" coordinates to ensure that the spatial distribution of features conforms to the stratigraphic topology, generating a fusion geological feature set (this set has no topological contradictions, which is different from the traditional processing method of fusion only by weight and ignores stratigraphic rules, thus improving the authenticity of features).

[0111] Preferably, in a scenario, when step 3.3 is specifically implemented, the processing objects are the geological background of the target work area and the fused geological feature set (wherein, the geological background of the target work area includes sedimentary environment type, such as "delta front sedimentation"; tectonic movement influence, such as "gentle folds with small strata dip angles"; stratigraphic development patterns, such as "continuous horizontal distribution of sandstone and interlayered mudstone", providing a basis for feature continuity and geological significance labeling; although the fused geological feature set has no topological contradictions, it may have spatial gaps due to insufficient density of the original data collection).

[0112] First, spatial continuity processing is performed: spatial gaps in the fused geological feature set are identified (i.e., coordinate points without "XYZ-fused feature value" records, such as the edge of the work area or sparse logging areas). Based on the regularity in the geological background of the target work area, an interpolation method is selected—if the gap is located in a continuously distributed sedimentary facies area (such as delta front sand bodies), trend interpolation is used (the gap feature value is calculated based on the trend of the fused feature value changes of the surrounding effective feature points); if the gap is affected by tectonics (such as gentle folds), tectonic-guided interpolation is used (combined with stratigraphic dip parameters to ensure that the interpolation features conform to the tectonic morphology); the interpolated fused feature values ​​are added to the spatial gaps and integrated with the original fused geological feature set to generate a continuous fused geological feature set (this set covers the entire spatial range of the work area, with no feature gaps, solving the subsequent modeling fault problem caused by the spatial discontinuity of traditional fused features).

[0113] Next, the geological significance is marked: Based on the correspondence between "fusion characteristic value - geological attribute" in the geological background of the target work area (e.g., "high amplitude fusion characteristic value" corresponds to "sandstone stratigraphic interface", based on the difference in wave impedance between sandstone and mudstone in deltaic sedimentation; "sudden increase in resistivity fusion characteristic value" corresponds to "sandstone-mudstone boundary", based on the lithological indication law of well logging sequence), for each fusion characteristic value in the continuous fusion geological feature set, its corresponding geological attribute (e.g., "stratigraphic interface", "lithological boundary", "porosity anomaly zone") and interpretation basis (e.g., "based on the distribution law of sand bodies in delta front sedimentation, this high amplitude feature is determined to be a sandstone interface") is marked);

[0114] Finally, the four elements of "XYZ coordinates - continuous fusion feature value - geological attributes - interpretation basis" are linked and integrated to generate a geological characteristic representation of the target work area (this representation achieves a precise mapping between the spatial location of the work area and its geological attributes, and each feature has a clear geological meaning, providing a clear geological reference for the construction of stratigraphic contact relationship constraints and lithological sequence constraints in step 4, and ensuring that the subsequent implicit topological structural representation can accurately reflect the geological envelope attributes).

[0115] Optionally, step 3, generating a geological characteristic profile of the target work area based on seismic reflection characteristics and well logging sequence characteristics, further includes:

[0116] Step 3.0: Generate the regional geological background of the target work area based on the existing geological outcrop data and borehole core description data of the target work area and its surroundings.

[0117] Optionally, step 3.0 generates the regional geological background of the target work area based on existing geological outcrop data and borehole core description data of the target work area and its surroundings, specifically including:

[0118] Step 3.0.1: Classify and label the existing geological outcrop data and borehole core description data of the target work area and its surroundings according to lithology, sedimentary structure and fossil assemblage, so as to generate a classified geological basic dataset.

[0119] Step 3.0.2: Perform vertical sequence correlation analysis on the lithological assemblages and sedimentary structures in the classified geological basic dataset to extract key features reflecting the sedimentary environment in order to generate a sedimentary environment feature set;

[0120] Step 3.0.3: Based on the sedimentary environment feature set and combined with the existing tectonic movement records around the target work area, establish the temporal correlation between sedimentary environment features and tectonic movement stages to generate a tectonic evolution time series table.

[0121] Step 3.0.4: Based on the sedimentary environment feature set and the tectonic evolution time series table, comprehensively verify the sedimentary environment features and tectonic evolution stages, and supplement the description of stratigraphic development patterns to generate the regional geological background of the target work area.

[0122] Preferably, the specific implementation process of step 3.0.1 is as follows: the processing objects are the existing geological outcrop data and borehole core description data of the target work area and its surroundings (wherein, the geological outcrop data includes the "spatial coordinates (XYZ) - lithological appearance description - sedimentary structure photo - fossil type record" of the outcrop point, reflecting the intuitive geological attributes of the exposed strata on the surface; the borehole core description data includes the "hole number - depth range - lithological name - core color - sedimentary structure type - fossil assemblage list" of the borehole, reflecting the continuous geological information of the vertical strata below ground. Both types of data need to be classified and labeled to achieve information structuring).

[0123] First, the data dimensions were analyzed: for geological outcrop data, four core dimensions were extracted: "spatial coordinates, lithology type, sedimentary structure, and fossil assemblage"; for borehole core description data, four core dimensions were extracted: "hole number-depth range, lithology type, sedimentary structure, and fossil assemblage" to ensure that the labeling dimensions of the two types of data are consistent, which will facilitate subsequent integration.

[0124] Next, classification and labeling are performed: For the "lithology type" dimension, classification standards are set according to common lithologies in the target work area (such as sandstone, mudstone, limestone, and shale), and the lithological appearance descriptions of geological outcrops and the lithological names of borehole core descriptions are matched to the corresponding lithology types (e.g., the description "gray fine-grained, well-sorted grains" is labeled as "sandstone"); For the "sedimentary structure" dimension, classification standards are set according to structural morphology (such as horizontal bedding, cross-bedding, ripple marks, and mud cracks), and the sedimentary structure photos of geological outcrops and the sedimentary structure types of borehole core descriptions are labeled as the corresponding structural types; For the "fossil assemblage" dimension, classification standards are set according to fossil phyla (such as foraminifera, ostracods, and plant pollen), and the fossil species records of geological outcrops and the fossil assemblage lists of borehole core descriptions are labeled as the corresponding fossil assemblage types;

[0125] Finally, the annotation results are integrated: the "spatial coordinates-lithological type-sedimentary structure-fossil assemblage" of geological outcrop data and the "hole number-depth segment-lithological type-sedimentary structure-fossil assemblage" of borehole core description data are integrated by row to generate a classified geological basic dataset (the rows of this dataset correspond to a single data sample, the columns correspond to the above core dimensions, and the elements at the intersection are the specific annotation results, such as a row record of "outcrop point A(X1,Y1,Z1)-sandstone-cross-bedding-foraminiferal assemblage", to ensure that the annotation of each sample can be traced back to the original data description and to avoid annotation ambiguity).

[0126] Preferably, in the specific technical implementation of step 3.0.2, the processing object is the classified geological basic dataset (this dataset contains a large number of labeled geological samples, and it is necessary to mine the intrinsic relationship between the samples through vertical sequence correlation analysis and extract key features that can reflect the sedimentary environment).

[0127] First, vertically continuous samples are selected: From the classified geological dataset, borehole core description data are selected (because borehole core data are continuously distributed according to depth segments, they are more suitable for vertical sequence analysis). They are arranged in the order of "bore number-depth segment increment" to form a single-hole vertical geological sequence (e.g., the sequence of borehole 1 is "depth 0-5m: mudstone-horizontal bedding-no fossils → depth 5-~12m: sandstone-cross-bedding-foraminifera → depth 12-~20m: limestone-horizontal bedding-ostracods"). Each borehole corresponds to a single-hole vertical geological sequence.

[0128] Next, vertical sequence correlation analysis was performed: for each single-hole vertical geological sequence, the "lithological type combination pattern" and "sedimentary structure evolution trend" were analyzed sequentially according to depth segment: if the lithological type in the vertical sequence of a certain borehole shows a repeated combination of "mudstone → sandstone → mudstone", and the corresponding sedimentary structure transitions from "horizontal bedding" to "cross-bedding" and then back to "horizontal bedding", it is determined to be a typical vertical combination pattern of "delta front underwater distributary channel - interdistributary bay"; if the lithological type shows a gradual change of "limestone → shale", and the sedimentary structure is always "horizontal bedding", it is determined to be a vertical evolution pattern of "shallow sea and semi-deep sea".

[0129] Finally, key features are extracted: the representative "lithological type combinations (such as mudstone-sandstone-mudstone), sedimentary structural sequences (such as horizontal bedding-cross-bedding-horizontal bedding), and fossil assemblages (such as foraminiferal assemblages)" from the above vertical combination patterns and evolution patterns are taken as key features reflecting the sedimentary environment. They are integrated according to "sedimentary environment type - lithological combination features - sedimentary structural features - fossil assemblages" to generate a sedimentary environment feature set (each sedimentary environment type in this set corresponds to clear multi-dimensional features, which is different from the traditional method of judging the sedimentary environment by relying on a single lithology, thus improving the reliability of environment determination).

[0130] Preferably, in a scenario, when step 3.0.3 is specifically implemented, the processing objects are the sedimentary environment feature set and the existing tectonic movement records around the target work area (wherein, the tectonic movement record includes "tectonic movement name - occurrence time - movement nature (such as folding, faulting, uplift) - influence range", such as "Late Yanshanian tectonic movement - 100-80 Ma ago - regional uplift - covering the target work area and adjacent areas", and the correspondence between sedimentary environment and tectonic movement needs to be established through temporal correlation).

[0131] First, sort out the tectonic movement timeline: sort the existing tectonic movement records around the target work area according to "occurrence time from early to late" to form a tectonic movement timeline (such as "Indosinian tectonic movement → early Yanshanian tectonic movement → late Yanshanian tectonic movement → Himalayan tectonic movement"), and clarify the time range and movement nature of each tectonic movement stage;

[0132] Next, establish the correlation: traverse each sedimentary environment type in the sedimentary environment feature set, combine it with the corresponding geological sample time information (such as the geological age corresponding to the depth of the borehole core, determined by fossil assemblage dating), and match it to the corresponding stage in the tectonic movement time series: if the geological sample dating result of a certain sedimentary environment type (such as "delta front") is "90-85 Ma ago", it corresponds to the "Late Yanshanian Tectonic Movement" (100-80 Ma ago) in the tectonic movement time series, and the nature of this tectonic movement is "regional uplift" - uplift leads to sea level drop, which is conducive to the development of delta front sedimentation, and it is determined that there is a temporal correlation between this sedimentary environment type and the Late Yanshanian Tectonic Movement stage;

[0133] Finally, the results were integrated and correlated: the results were integrated according to the "tectonic movement stage - occurrence time - movement nature - associated sedimentary environment characteristics" to generate a tectonic evolution time series table (the rows of this table correspond to the tectonic movement stage, the columns correspond to the occurrence time, movement nature, and associated sedimentary environment characteristics, and the elements at the intersection are the specific correlation content, such as "Late Yanshanian tectonic movement - 100-80 Ma ago - regional uplift - delta front, fluvial sedimentary environment", to ensure that the correlation between sedimentary environment characteristics and tectonic movement has clear temporal and geological mechanism support).

[0134] Preferably, in the specific technical implementation of step 3.0.4, the processing objects are the sedimentary environment feature set and the tectonic evolution time series table (the two types of objects need to be verified to ensure consistency, and then the stratigraphic development law is added to generate the regional geological background).

[0135] First, a comprehensive verification is performed: traversing each tectonic movement stage in the tectonic evolution time series table, checking whether the associated sedimentary environment characteristics are consistent with the descriptions in the sedimentary environment characteristic set: if the associated sedimentary environment characteristic marked for a certain tectonic movement stage is "fluvial facies", but the lithological assemblage characteristics corresponding to "fluvial facies" in the sedimentary environment characteristic set (such as "conglomerate-sandstone-mudstone") contradict the lithological assemblage of the geological sample of that tectonic movement stage (the actual sample is "limestone-shale"), then the association in the tectonic evolution time series table is corrected, and the contradictory sedimentary environment characteristics are removed; if a certain environment type in the sedimentary environment characteristic set has no corresponding tectonic movement stage association, then supplementary dating analysis is performed to match it to a suitable tectonic movement stage, ensuring that there are no logical contradictions between the sedimentary environment characteristic set and the tectonic evolution time series table;

[0136] Next, we supplemented the description of stratigraphic development patterns: Based on the verified sedimentary environment feature set and tectonic evolution time series table, we extracted the vertical development patterns of the strata in the target work area (e.g., "from early to late, the vertical sequence of strata is 'shallow marine limestone → deltaic sandstone → fluvial conglomerate', corresponding to tectonic movements from 'regional subsidence' to 'regional uplift'"), lateral development patterns (e.g., "deltaic sandstone is thicker in the middle of the work area, gradually transitioning to mudstone towards the edge, controlled by paleotopography"), and lithological distribution patterns (e.g., "limestone is mainly distributed in the deep part of the work area, controlled by the range of early marine transgression").

[0137] Finally, all information is integrated: the “verified sedimentary environment feature set - verified tectonic evolution time series table - vertical development law - lateral development law - lithological distribution law” are integrated to generate the regional geological background of the target work area (this background covers multi-dimensional information on sedimentation, structure and stratigraphy, providing a comprehensive basis for the feature space continuity and geological significance annotation in the subsequent step 3.3, which is different from the traditional regional background that only contains a simple description of sedimentary environment, and the information is more complete).

[0138] Optionally, step 4, based on the stratigraphic contact relationship constraints and lithological sequence constraints for the target work area, and according to the geological characteristic characterization, determines the implicit topological structural characterization of the geological envelope in the target work area, specifically as follows:

[0139] Based on the seismic exploration interpretation profile data of the target work area, stratigraphic contact relationship constraints and lithological sequence constraints are constructed for the target work area. Based on the geological characteristics, the implicit topological structural characterization of the geological envelope in the target work area is determined.

[0140] Optionally, step 4 involves constructing stratigraphic contact relationship constraints and lithological sequence constraints for the target work area based on seismic exploration interpretation profile data. According to geological characteristic characterization, the implicit topological structural characterization of the geological envelope in the target work area is determined, specifically including:

[0141] Step 4.1: Based on the seismic exploration interpretation profile data of the target work area, generate contact relationship analogy rules for the target work area;

[0142] Step 4.2: Based on the contact relationship analogy rules of the target work area, generate stratigraphic contact relationship constraints and lithological sequence constraints of the target work area;

[0143] Step 4.3: Based on the stratigraphic contact relationship constraints and lithological sequence constraints of the target work area, determine the implicit topological structural characterization of the geological envelope in the target work area according to the geological characteristics.

[0144] Optionally, step 4.1, based on the seismic exploration interpretation profile data of the target work area, generates contact relationship analogy rules for the target work area, including the following steps:

[0145] Step 4.1.1: Based on the seismic exploration interpretation profile data of the target work area, perform parametric extraction of the spatial location, dip angle, and interface continuity of the stratigraphic contact interface in the seismic exploration interpretation profile to generate the contact feature set of the target work area profile.

[0146] Step 4.1.2: Based on the contact feature set of the target work area profile, and combined with the verified analog geological model data of the neighboring areas of the target work area, establish an analog mapping between the profile contact features and the contact relationship of the neighboring areas to generate the contact relationship analogy rules of the target work area.

[0147] Preferably, the specific implementation process of step 4.1.1 is as follows: the processing object is the seismic exploration interpretation profile data of the target work area (this data includes two-dimensional or three-dimensional visualization profiles. The coordinate system of the two-dimensional profile is defined as "X-axis - transverse survey line direction, Z-axis - vertical depth direction". The three-dimensional profile additionally adds "Y-axis - longitudinal survey line direction". Each profile is marked with "survey line number - depth range" information. The profile clearly presents the reflection phase axis morphology of the stratigraphic contact interface. The interface is represented by the interruption of the phase axis, the change of attitude or the sudden change of amplitude, which provides a visual basis for parameter extraction).

[0148] First, the spatial location of the stratigraphic contact interface is extracted: using the coordinate system of the seismic exploration interpretation profile data as a reference, the stratigraphic contact interface is identified point by point along the XZ (or XYZ) trajectory of the profile. For two-dimensional profiles, key points with obvious reflection characteristics on the interface (such as amplitude peak points, phase axis inflection points) are selected, and the coordinates of each key point are recorded (X value - lateral survey line position, Z value - vertical depth). For three-dimensional profiles, the Y value - longitudinal survey line position is additionally recorded to form an interface coordinate sequence (e.g., the coordinate sequence of interface 1 in a two-dimensional profile is "(X1,Z1), (X2,Z2)...(X...Z1)"). n Z n )”, where X1-X n Representing different positions of the transverse survey line, Z1-Z n (Representing the interface depth at the corresponding location), this interface coordinate sequence directly reflects the spatial orientation of the interface;

[0149] Next, tilt angle extraction is performed: for the interface coordinate sequence of each interface, the positions of adjacent key points (such as (X)) are calculated. i Z i ) and (X i+1 Z i+1 The slope of the tangent line (slope k = (Z)) i+1 -Z i ) / (X i+1 -X iIn the 3D profile, the slope of the XZ plane is taken. Then, the angle between the tangent line and the horizontal direction (positive X-axis direction) of the profile is calculated using the arctangent function to obtain the interface tilt angle (a positive angle value indicates that the interface tilts downward along the positive X-axis direction, and a negative value indicates that it tilts upward). All tilt angles are integrated according to the interface number to generate an interface tilt angle vector (the length of this vector is consistent with the total number of interfaces, and the i-th element in the vector corresponds to the average tilt angle of the i-th interface, such as interface 1 corresponding to tilt angle α1, and interface 2 corresponding to α2).

[0150] Then, interface continuity extraction is performed: for the interface coordinate sequence of each interface, two parameters are calculated: total length of continuous segment (referring to the serial length of coordinate points without interruption of the same phase axis, such as (X1,Z1) to (X5,Z5) without interruption, the continuous segment length is X5-X1) and number of interruptions (referring to the number of times the same phase axis is obviously blank or the attitude changes abruptly in the coordinate sequence); the continuity ratio is calculated according to "continuity ratio = total length of continuous segment / total length of transverse section (two-dimensional) or total range of XY plane of section (three-dimensional)", and the interruption frequency is calculated according to "interruption frequency = number of interruptions / total length of interface coordinate sequence". The two together constitute the interface continuity parameter (for example, the continuity parameter of interface 1 is "continuity ratio 82%, interruption frequency 0.15 times / unit length", where "unit length" is the basic measurement unit of transverse section, such as meter).

[0151] Finally, the "interface number - interface coordinate sequence - interface tilt angle - interface continuity parameters" are integrated row by row to generate the contact feature set of the target work area profile (the rows of this feature set correspond to the interface number, the columns correspond to the four dimensions mentioned above, and the elements at the intersection are specific parameter values, such as a row record "interface 1-[(X1,Z1)…(X n Z n The continuity ratio is 82% and the interruption frequency is 0.15 times per unit length. This ensures that the parameters of each interface can be traced back to the original visualization information of the seismic exploration interpretation profile data, thus avoiding the subjectivity of parameter extraction.

[0152] Preferably, in the specific technical implementation of step 4.1.2, the processing objects are the contact feature set of the target work area profile and the verified analog geological model data of the neighboring areas of the target work area (wherein, the analog geological model data includes the mapping relationship of "formation contact relationship type - corresponding profile contact feature parameters" verified by drilling in the neighboring areas, such as the parameters corresponding to the "conformable contact" type in the neighboring areas are "interface tilt angle deviation <12°, continuity ratio >78%, and the interface coordinate sequence is a gentle curve", and the parameters corresponding to "unconformable contact" are "interface tilt angle deviation >28°, continuity ratio <65%, and the interface coordinate sequence has obvious discontinuity". This data has high reliability because it has been verified by drilling).

[0153] First, feature similarity calculation is performed: traversing each interface in the contact feature set of the target work area profile (processed sequentially according to interface number 1 to n), and comparing each interface's three core features—interface tilt angle, interface continuity parameters (including continuity ratio and interruption frequency), and interface coordinate sequence morphology—with the corresponding parameters of all stratigraphic contact relationship types in the analog geological model data item by item:

[0154] 1. Calculate the similarity of tilt angles: Based on the standard tilt angle range of a certain contact relationship type in the neighboring region (e.g., the standard range for integrated contact is -10° to 10°), if the tilt angle of the target interface falls within this range, the similarity is 1; if it exceeds the range, it is calculated according to "similarity = 1 - |target angle - midpoint of standard range| / half width of standard range" (e.g., if the target angle is 12°, the midpoint of the standard range is 0°, and the half width is 10°, then the similarity = 1 - 12 / 10 = 0, to avoid negative similarity).

[0155] 2. Calculate continuity similarity: Continuity ratio similarity = Target interface continuity ratio / Neighboring area standard continuity ratio (the ratio should be controlled between 0 and 1. For example, if the target ratio is 82% and the standard ratio is 78%, then the similarity = 82 / 78≈1, take 1); Interruption frequency similarity = 1 - Target interface interruption frequency / Neighboring area standard interruption frequency (for example, if the target frequency is 0.15 and the standard frequency is 0.3, the similarity = 1 - 0.15 / 0.3 = 0.5); Continuity similarity = (Continuity ratio similarity × 0.6 + Interruption frequency similarity × 0.4);

[0156] 3. Calculate morphological matching degree: Compare the target interface coordinate sequence with the standard sequence of neighboring regions (e.g., a smooth quadratic curve for integrated contact) through curve fitting. If the goodness of fit (e.g., R²) is satisfactory, the morphological matching degree is determined. 2 If the score is ≥0.85, the matching degree is 1; otherwise, it is calculated as "matching degree = goodness of fit".

[0157] The total similarity between the target interface and each contact relationship type of the neighboring area is calculated by weighting the similarity as follows: "Total similarity = tilt angle similarity × 0.4 + continuity similarity × 0.3 + morphological matching degree × 0.3".

[0158] Next, an analogy mapping is established: For each target interface, the stratigraphic contact relationship type of the neighboring area with the highest total similarity is selected. If the highest similarity is ≥60%, the corresponding mapping of "target interface feature - neighboring area contact relationship type" is directly established (e.g., if the total similarity between target interface 1 and the neighboring area "conformable contact" is 76%, then it is mapped to "conformable contact"); if the highest similarity is <60%, it is corrected in combination with the geological background of the target work area (e.g., it is known that the tectonic activity in the work area was weak in a certain period and there was no obvious evidence of unconformity), and it is prioritized to map it to the contact relationship type under the scenario of weak tectonic activity (e.g., conformable contact) to avoid misjudgment due to data errors.

[0159] Finally, all analogy mapping relationships are integrated and sorted according to "target interface feature matching conditions - corresponding contact relationship types" to generate contact relationship analogy rules for the target work area (rule example: "If the interface tilt angle falls within the range of -15° to 15°, the continuity ratio is ≥75%, the interruption frequency is ≤0.2 times / unit length, and the goodness of fit of the coordinate sequence is ≥0.8, then the corresponding contact relationship type is integrated contact"). Each matching condition in the rule corresponds to a clear parameter threshold, ensuring that when applying the rule in the subsequent step 4.2.1, the contact relationship type can be directly determined by parameter comparison. Unlike the traditional coarse approach of directly applying the neighboring area model, this step achieves dynamic mapping through multi-dimensional weighted similarity calculation, which significantly improves the adaptability of the rule to the target work area.

[0160] Optionally, in step 4.2, based on the contact relationship analogy rules of the target work area, stratigraphic contact relationship constraints and lithological sequence constraints of the target work area are generated;

[0161] Step 4.2.1: Generate stratigraphic contact relationship constraints for the target work area based on the contact relationship analogy rules.

[0162] Step 4.2.2: Based on the formation contact relationship constraints of the target work area, and combined with the lithological vertical transition records in the well logging data of the target work area, establish the corresponding relationship between the formation contact interface and the upper and lower lithologies to generate the lithological sequence constraints of the target work area.

[0163] Preferably, the specific implementation process of step 4.2.1 is as follows: the processing object is the contact relationship analogy rule of the target work area (this rule is generated by step 4.1.2 and includes the mapping relationship of "feature matching conditions - stratigraphic contact relationship type", such as "interface tilt angle deviation <15°, continuity ratio ≥75% → conformable contact" and "interface tilt angle deviation >28°, continuity ratio <65% → unconformable contact", each mapping relationship corresponds to a clear parameter threshold);

[0164] First, constraint clause extraction is performed: Each mapping relationship in the contact relationship analogy rules of the target work area is traversed, and the "feature matching conditions" are transformed into targeted constraint clauses. For the "integrated contact" type, the constraint clause is extracted as follows: "The interface tilt angle must fall within the range of [-15°, 15°], the continuity ratio is ≥75%, and the interruption frequency is ≤0.2 times / unit length ('unit length' refers to the lateral basic measurement unit of the seismic exploration interpretation profile of the target work area, such as meters)". For the "unintegrated contact" type, the constraint clause is extracted as follows: "The interface tilt angle must exceed the range of [-28°, 28°], the continuity ratio is ≤65%, and the difference in the attitude of the upper and lower strata of the interface must be ≥15°". All constraint clauses are integrated according to the contact relationship type to form a preliminary constraint clause set.

[0165] Next, construct the constraint parameter matrix: with "contact relationship type" as the row and "constraint parameter item" as the column (constraint parameter items include tilt angle range, continuity ratio threshold, interruption frequency threshold, attitude difference threshold, etc.), fill the parameter thresholds from the initial constraint clause set into the matrix to generate the contact relationship constraint parameter matrix (the rows of this matrix correspond to integrated contact, unconformable contact, etc., the columns correspond to each constraint parameter, and the elements at the intersection are the specific threshold ranges, such as "integrated contact - tilt angle range" corresponding to [-15°, 15°]);

[0166] Then, a spatial consistency check is performed: based on the tectonic patterns in the geological background of the target work area (e.g., the overall structure of the work area is gentle and there are no large-scale faults), check whether there are contradictory clauses in the contact relationship constraint parameter matrix. If a constraint clause for "unconformable contact" requires an attitude difference ≥ 15°, but the regional geological background shows that the tectonic activity in the work area is weak and the attitude difference is generally < 10°, then the threshold of the clause is corrected to "attitude difference ≥ 10°". At the same time, the constraint compatibility of adjacent interfaces is checked. For example, if interface 1 is a conformable contact (tilt angle 5°), and interface 2 below it is marked as an unconformable contact, it is necessary to ensure that the difference between the tilt angle of interface 2 and interface 1 is ≥ the corrected attitude difference threshold to avoid spatial logical contradictions.

[0167] Finally, the contact relationship constraint parameter matrix and constraint clauses after integration and verification are used to generate the stratigraphic contact relationship constraints for the target work area (these constraints specify the quantitative parameter requirements and spatial compatibility rules for each type of contact relationship, which is different from the traditional constraint method that only qualitatively describes the contact relationship, and realizes the quantification and verification of the constraints).

[0168] Preferably, in the specific technical implementation of step 4.2.2, the processing object is the formation contact relationship constraint of the target work area and the lithological vertical transition record in the logging data of the target work area (wherein, the lithological vertical transition record is extracted from the logging data of the target work area generated in step 1.3, and includes "depth location - physical parameter mutation type - corresponding lithological change", such as "depth Z1 - resistivity increases by 50% → mudstone → sandstone" and "depth Z2 - sonic transit time decreases by 30% → sandstone → limestone". The physical parameter mutation point corresponds to the lithological boundary and has a spatial correlation with the formation contact interface).

[0169] First, the vertical transition record extraction of lithology is performed: the logging data of the target area is traversed to identify vertical abrupt change points of physical parameters (resistivity, sonic transit time, density) (the criteria for determining abrupt change points is "the difference in parameter values ​​between adjacent depth points exceeds 20% of the mean of the parameter sequence"). The depth location (Z value) and parameter abrupt change type (sudden increase / sudden decrease) of each abrupt change point are recorded. Combined with the existing lithology sample library of the target area (such as the "resistivity range - lithology" correspondence of the drilling core calibration), the upper and lower lithologies corresponding to the abrupt change points are determined (such as resistivity of 80~150Ω·m corresponds to sandstone, and 2~20Ω·m corresponds to mudstone). A detailed table of vertical transition of lithology is generated (the rows of this table correspond to the abrupt change point number, and the columns correspond to the depth location, parameter abrupt change type, overlying lithology, and underlying lithology).

[0170] Next, a spatial correlation between the contact interface and the lithological transition is established: based on the interface coordinate sequence (including depth Z value) in the stratigraphic contact relationship constraints of the target work area, the depth range of the contact interface is matched with the depth position in the lithological vertical transition details table. If the depth range of the interface coordinate sequence of a certain contact interface includes the depth position Z1 of a certain abrupt change point, and the contact relationship type of the contact interface is a conformable contact (constraints require high continuity and gentle lithological transition), then the lithological change corresponding to the abrupt change point (such as mudstone → sandstone) is determined to be the upper and lower lithologies of the contact interface. If the contact interface is an unconformable contact (constraints require large differences in attitude and there may be missing lithologies), then lithological transition records with overlapping depth ranges and large parameter abrupt changes are matched first (such as a sudden increase in resistivity of 80%, corresponding to conglomerate → sandstone).

[0171] Then, construct an interface-lithology correspondence table: integrate according to "contact interface number - interface depth range - overlying lithology - underlying lithology - matching basis (parameter mutation type + contact relationship constraint clause)" to generate an interface-lithology correspondence table (such as "interface 1 - depth Z0 - Z2 - mudstone - sandstone - resistivity surge of 50% + integrated contact continuity constraint"), to ensure that the lithology above and below each contact interface is supported by well logging data;

[0172] Finally, lithological sequence constraints are generated: Based on the interface-lithology correspondence table, the lithologies above and below all contact interfaces in the target work area are extracted and sorted from shallow to deep vertical depth to form a vertical sequence chain of "overlying lithology → underlying lithology" (e.g., "mudstone → sandstone → limestone → shale"). Combining the spatial compatibility rules in the stratigraphic contact relationship constraints of the target work area, sequence constraint clauses are supplemented—such as "the lithological sequence corresponding to the conformable contact must maintain a continuous transition without any abrupt lithological changes" and "the lithological sequence corresponding to the unconformable contact is allowed to have 1-2 missing lithologies, but the missing lithologies must conform to the sedimentary discontinuity pattern in the regional geological background." Finally, lithological sequence constraints for the target work area are generated (this constraint clarifies the vertical lithological transition order and the matching rules with the contact relationship, solving the problem of the disconnect between traditional lithological sequence constraints and contact relationships).

[0173] Preferably, step 4.3, constraining the stratigraphic contact relationships and lithological sequence of the target work area, and determining the implicit topological structural characterization of the geological envelope in the target work area based on geological characteristic characterization, specifically includes the following steps:

[0174] Step 4.3.1: Based on the stratigraphic contact relationship constraints and lithological sequence constraints of the target work area, perform constraint adaptation on the geological characteristic characterization of the target work area to generate a constrained geological characteristic characterization.

[0175] Step 4.3.2: Based on the geological characteristic characterization after constraint adaptation, generate an implicit topological structural characterization of the geological envelope in the target work area.

[0176] Preferably, the specific implementation process of step 4.3.1 is as follows: the processing objects are the stratigraphic contact relationship constraints of the target work area, the lithological sequence constraints of the target work area, and the geological characteristic characterization of the target work area (wherein, the geological characteristic characterization of the target work area includes three-dimensional correlation information of "spatial coordinates (XYZ) - geological attributes (lithological type, stratigraphic interface type) - geological significance labeling", which needs to be verified and adapted through two types of constraints to ensure that it conforms to the actual geological laws of the target work area).

[0177] First, the geological characteristic characterization elements are decomposed: the geological characteristic characterization of the target work area is divided into "spatial coordinate units". Each spatial coordinate unit (corresponding to a single XYZ coordinate point) contains "lithological attribute value and interface attribute value (if it is an interface, the type is marked, such as conformable contact / unconformable contact, and if it is not an interface, it is marked 'none')". A geological characteristic element table is generated (the rows of this table correspond to the spatial coordinate unit number, and the columns correspond to the X value, Y value, Z value, lithological attribute value, and interface attribute value. The elements at the intersection are specific parameters, such as "unit 1-X1-Y1-Z1-sandstone-conformable contact").

[0178] Next, the stratigraphic contact relationship constraint adaptation verification is performed: based on the "contact relationship type - parameter threshold" in the stratigraphic contact relationship constraints of the target work area (e.g., the required tilt angle for conglomerate contact is [-15°, 15°], and the continuity ratio is ≥75%), all spatial coordinate units marked with "interface attribute value" in the geological characteristic element table are traversed: if a unit is marked "conglomerate contact", but the corresponding tilt angle (extracted from the interface parameters characterized by geological characteristics) is 18°, which exceeds the constraint threshold, then the unit is marked as a "constraint conflict feature unit"; if a unit is marked "unconglomerate contact", but the continuity ratio is 80%, which is higher than the constraint threshold of 65%, it is also marked as a conflict unit, and a preliminary constraint conflict list is generated;

[0179] Then, the lithological sequence constraint adaptation verification is performed: based on the "vertical lithological transition sequence" (such as "mudstone → sandstone → limestone") in the lithological sequence constraint of the target work area, the geological characteristic element table is sorted in ascending order of Z value (vertical depth) to form the vertical lithological sequence of each XY plane: if the vertical lithological sequence of a certain XY position is "mudstone → limestone → sandstone", which contradicts the transition sequence in the constraint, then the spatial coordinate unit corresponding to the sequence is marked as a "constraint conflict feature unit", added to the preliminary constraint conflict list, and a complete constraint conflict list is generated;

[0180] Subsequently, conflict-related feature units are processed: conflicts are corrected by combining the regional geological background of the target work area (such as the presence of local small faults or sedimentary discontinuities in the known work area). If a "conformable contact" unit is marked as conflicted due to its tilt angle exceeding the threshold, but the regional geological background shows that there is a small normal fault at that location, its interface attribute value is corrected to "fault contact" and added to the geological characteristic element table. If a vertical lithological sequence conflict is caused by local sedimentary facies change, the lithological attribute value is corrected by referring to the similar facies change rules of neighboring areas, and a corrected geological characteristic element table is generated.

[0181] Finally, the revised geological characteristic element table was integrated, and the "lithological attribute values, interface attribute values, and geological significance labels" were re-linked according to the "XYZ spatial coordinates" to ensure that the geological attributes of each spatial coordinate unit conform to the stratigraphic contact relationship constraints and lithological sequence constraints of the target work area. This generated a constrained geological characteristic representation (which resolved the possible geological law contradictions in the original geological characteristic representation and provided compliant geological basis data for the subsequent generation of implicit topological structural representations).

[0182] Preferably, in the specific technical implementation of step 4.3.2, the processing object is the geological characteristic representation after constraint adaptation (this representation contains compliant "spatial coordinates-geological attribute" association information, which needs to be transformed into a structural representation that can reflect the topological relationship of the geological envelope through implicit expression. The core of implicit topological structural representation is to implicitly reflect the spatial distribution and topological connection relationship of the geological body through a continuous mathematical field (such as a scalar field).

[0183] First, an implicit scalar mapping rule for geological envelopes is constructed: based on the geological attribute type in the geological characteristic representation after constraint adaptation, a unique mapping relationship between "geological attribute - scalar value" is set - such as "sandstone" corresponds to scalar value 1.0, "mudstone" corresponds to scalar value 2.0, and "limestone" corresponds to scalar value 3.0. Stratigraphic interfaces (such as conglomerate contact interfaces and fault contact interfaces) correspond to scalar gradient abrupt change zones (gradient value ≥ preset threshold, such as 0.5 / unit length, where "unit length" is the basic measurement unit of spatial coordinates of the target work area, such as meters). The scalar values ​​of the same geological envelope (such as a set of continuous sandstone strata) are consistent, while the scalar values ​​of different geological envelopes have clear differences. A scalar mapping table for geological envelopes is generated.

[0184] Next, an initial geological envelope scalar field is generated: based on the spatial coordinate system (XYZ) of the target work area, a three-dimensional spatial scalar field matrix is ​​constructed (the rows of the matrix correspond to the X values, the columns correspond to the Y values, the depth dimension corresponds to the Z values, and the elements at the intersection of the matrix are the scalar values ​​to be assigned); each spatial coordinate unit in the geological characteristic characterization after constraint adaptation is traversed, and according to its lithological attribute value or interface attribute, the corresponding scalar value or scalar gradient rule is extracted from the geological envelope scalar mapping table and assigned to the corresponding position in the empty scalar field matrix—for example, if the lithological attribute value of a spatial coordinate unit is "sandstone", then 1.0 is filled in the corresponding position in the scalar field matrix; if a unit is "integrated contact interface", then transition scalar values ​​are assigned between its adjacent coordinate units according to a gradient of 0.5 / unit length, generating the initial geological envelope scalar field;

[0185] Then, the topological continuity of the scalar field is optimized: based on the spatial compatibility rules in the stratigraphic contact relationship constraints of the target work area (such as the scalar gradient direction of the integrated contact interface must be consistent with the stratigraphic dip direction), the scalar value change trend of the initial geological envelope scalar field is verified. If the scalar value of a certain area changes abruptly from 1.0 (sandstone) to 3.0 (limestone) without interface attribute labeling, which does not conform to the transition law of "sandstone → mudstone → limestone", then a transition coordinate unit with a scalar value of 2.0 (mudstone) is inserted to correct the scalar field change trend. For the scalar gradient at the fault contact interface, its direction is ensured to be consistent with the fault dip, and the optimized geological envelope scalar field is generated (the continuous change of this scalar field can implicitly reflect the spatial boundary and internal continuity of the geological envelope, i.e., the topological relationship).

[0186] Finally, the topological meaning of the scalar field is marked: the correspondence between "scalar value - geological envelope attribution" and "scalar gradient - stratigraphic boundary" in the optimized geological envelope scalar field is clarified. For example, the connected region with a scalar value of 1.0 is the "lower sandstone geological envelope of the target area", and the region with a scalar gradient ≥ 0.5 / unit length is the "stratigraphic boundary of the geological envelope". These topological meanings are associated and integrated with the optimized geological envelope scalar field to generate an implicit topological structural representation of the geological envelope in the target area (this representation implicitly expresses the spatial distribution, stratigraphic boundaries and mutual topological connections of the geological envelope through a continuous scalar field, which is different from the traditional explicit geometric modeling method that relies on discrete surfaces and is more suitable for the continuous feature processing of the subsequent neural network model).

[0187] Optionally, step 5, processing the implicit topological structural representation based on the trained third neural network model to predict the stratigraphic boundaries of the geological envelope in the target work area, specifically includes:

[0188] Step 5.1: Based on the implicit construction feature enhancement layer in the trained third neural network model, perform boundary-sensitive feature enhancement on the implicit topological construction representation to generate enhanced implicit construction features.

[0189] Step 5.2: Based on the multi-scale boundary feature capture layer in the trained third neural network model, perform multi-scale boundary feature extraction on the enhanced implicit construction features to generate a multi-scale boundary feature set.

[0190] Step 5.3: Based on the preliminary prediction layer of the layer boundary in the trained third neural network model, perform preliminary prediction of the layer boundary on the multi-scale boundary feature set to generate initial layer boundary data.

[0191] Step 5.4: Based on the boundary topology verification and optimization layer in the trained third neural network model, perform topology verification and optimization on the initial stratigraphic boundary data to generate the final stratigraphic boundary data of the geological envelope in the target work area.

[0192] Preferably, the specific implementation process of step 5.1 is as follows: the object of processing is the implicit topological structural representation (this representation is a three-dimensional scalar field matrix, the rows of the matrix correspond to the X coordinate of the transverse survey line of the target work area, the columns correspond to the Y coordinate of the longitudinal survey line, the depth dimension corresponds to the Z coordinate, and the elements at the intersection of the matrix are scalar values—areas with the same or similar scalar values ​​correspond to the same geological envelope, and areas with significant changes in scalar value gradients correspond to the stratigraphic boundaries of the geological envelope, such as sandstone geological envelopes corresponding to scalar values ​​of 1.0 and mudstone corresponding to 2.0, and the scalar gradient at the boundary between the two is ≥0.5 / unit length, where "unit length" is the basic measurement unit of the spatial coordinates of the target work area, such as meters);

[0193] Based on the implicit feature enhancement layer in the trained third neural network model (this layer has a built-in boundary-sensitive enhancement algorithm, the core of which is to focus on the scalar gradient significant region and avoid boundary confusion caused by the over-enhancement of features in non-boundary regions), the gradient of the scalar field of the implicit topological construction representation is first calculated: for each element (X,Y,Z) in the three-dimensional scalar field matrix, the difference between its scalar value and the scalar values ​​of the adjacent 6 directions (up, down, left, right, front, back) is calculated, according to the formula "gradient value = √[(ΔX scalar difference))". 2 +(ΔY scalar difference)) 2 +(ΔZ scalar difference)) 2 ]”Calculate the scalar field gradient of the element and generate the scalar field gradient matrix (the dimension is consistent with the implicit topological construction representation, and the elements are the gradient values ​​at the corresponding positions);

[0194] Next, the boundary candidate regions are defined: Based on the scalar field gradient matrix, a gradient threshold is set (the threshold is the mean of the scalar field gradient matrix + 1 standard deviation, such as 0.4 / unit length if the mean is 0.3 / unit length and the standard deviation is 0.1). Regions with gradient values ​​≥ the threshold are determined as boundary candidate regions, and regions with gradient values ​​< the threshold are determined as non-boundary regions. A boundary candidate region mask is generated (the mask matrix dimension is the same as the scalar field, with 1 for boundary candidate regions and 0 for non-boundary regions).

[0195] Then, boundary-sensitive feature enhancement is performed: the implicit topological construction representation is multiplied element-wise with the boundary candidate region mask, retaining the scalar values ​​of the boundary candidate regions and weakening the scalar values ​​of non-boundary regions (the scalar values ​​of non-boundary regions are multiplied by a weakening coefficient of 0.5 to avoid interference with subsequent boundary extraction); at the same time, gradient-weighted enhancement is performed on the scalar values ​​in the boundary candidate regions—the higher the gradient value of the scalar field, the greater the enhancement weight (e.g., a gradient value of 0.6 / unit length corresponds to a weight of 1.2, and 0.4 / unit length corresponds to a weight of 1.0), and the enhanced scalar field matrix is ​​generated according to the formula "enhanced scalar value = original scalar value × enhancement weight".

[0196] Finally, the enhanced scalar field matrix is ​​kept consistent with the spatial coordinate system of the original implicit topological construction representation to generate enhanced implicit construction features (in which the features of the layer boundary regions with significant scalar gradients are highlighted, and the redundant information of the non-boundary regions is suppressed, which is different from the limitation of traditional uniform enhancement of the whole region, which easily leads to confusion between boundary and non-boundary features).

[0197] Preferably, in the specific technical implementation of step 5.2, the processing object is the enhanced implicit structural features (the feature is the enhanced three-dimensional scalar field matrix, the scalar gradient features of the stratigraphic boundary region have been highlighted, but the boundary details at different scales - such as the microscopic lithological thin interlayer boundary and the macroscopic stratigraphic interface boundary - still need to be fully captured through multi-scale extraction).

[0198] Based on the multi-scale boundary feature capture layer in the trained third neural network model (this layer has three different sizes of three-dimensional convolution kernels built-in: small kernels are suitable for micro-scale boundary extraction, medium kernels are suitable for medium-scale boundary extraction, and large kernels are suitable for macro-scale boundary extraction, avoiding the omission of key boundary information in single-scale extraction), the small-scale boundary feature extraction is first performed: a small-sized three-dimensional convolution kernel (such as 3×3×3 pixels, corresponding to the actual spatial range of the target work area must match the scale of the micro-boundary, such as 0.5m×0.5m×0.5m) is used to perform convolution operation on the enhanced implicit structural features. The sliding step size is consistent with the sampling interval of the scalar field matrix to extract the boundary features of micro-thin interlayers (such as mudstone interlayers with a thickness of <2m) and generate a small-scale boundary feature map (the dimension is consistent with the enhanced implicit structural features, and the elements are micro-boundary feature values).

[0199] Next, perform mesoscale boundary feature extraction: use a medium-sized 3D convolution kernel (e.g., 5×5×5 pixels, corresponding to 1m×1m×1m in actual space) to repeat the above convolution operation, extract the boundary features of the mesoscale stratigraphic segment (e.g., sandstone segment with a thickness of 2-10m), and generate a mesoscale boundary feature map.

[0200] Then, large-scale boundary feature extraction is performed: a large-size three-dimensional convolution kernel (e.g., 7×7×7 pixels, corresponding to 2m×2m×2m in actual space) is used to extract the boundary features of the stratigraphic interface in the macroscopic region (e.g., the interface between sandstone and mudstone with a thickness >10m) and generate a large-scale boundary feature map.

[0201] Finally, the small-scale, medium-scale, and large-scale boundary feature maps are integrated according to "feature point number - scale type - feature value" to generate a multi-scale boundary feature set (this set is a three-dimensional matrix, where rows correspond to the spatial feature point numbers of the target work area, columns correspond to the scale type (small / medium / large), and the depth dimension corresponds to the boundary feature value, such as "feature point 1 - small scale - 0.8" and "feature point 1 - medium scale - 0.6", to ensure that boundary features of different scales at the same spatial location can participate in subsequent predictions in a coordinated manner).

[0202] Preferably, in a scenario, when step 5.3 is specifically implemented, the processing object is a multi-scale boundary feature set (this set contains boundary features of different scales, and the features need to be mapped to the probability of existence of layer boundaries through classification operations to achieve preliminary boundary localization);

[0203] Based on the preliminary prediction layer of the layer boundary in the trained third neural network model (this layer is a fully connected classifier, the input is the feature vector of the multi-scale boundary feature set, and the output is the "probability of existence of layer boundary" of the corresponding spatial location), the multi-scale boundary feature set is first reshaped into vectors: the "small-scale feature value-medium-scale feature value-large-scale feature value" of each spatial feature point is concatenated into a three-dimensional feature vector (e.g. the vector of feature point 1 is [0.8,0.6,0.5]), generating a feature vector matrix (rows correspond to feature point numbers, columns correspond to 3 scale feature values);

[0204] Next, the feature vector matrix is ​​input into the classifier of the preliminary prediction layer of the layer boundary. The activation function is used to output the probability of the existence of the layer boundary corresponding to each feature vector (the probability value ranges from 0 to 1. The closer the value is to 1, the higher the probability that the feature point belongs to the layer boundary). A boundary probability vector is generated (the length is the same as the number of feature points, and the elements are the probability values ​​of the corresponding feature points).

[0205] Then, a probability threshold is set (the threshold is determined based on the statistical analysis of the labeled layer boundary samples in the target work area; for example, if the minimum probability of a boundary point in the sample is 0.6, then the threshold is set to 0.6). The boundary probability vector is traversed: feature points with a probability value ≥ the threshold are marked as “initial boundary points” and their spatial coordinates (X,Y,Z) are recorded; feature points with a probability value < the threshold are marked as “non-boundary points” and excluded from the boundary extraction range.

[0206] Finally, based on the proximity of spatial coordinates, the "initial boundary points" are connected into continuous boundary segments in the horizontal, vertical, and depth directions. Then, adjacent boundary segments are integrated into complete boundary curves or surfaces to generate initial stratigraphic boundary data (this data includes "boundary number - boundary coordinate sequence - corresponding geological envelope pair", such as "boundary 1 - [(X1,Y1,Z1)...(Xn,Yn,Zn)] - sandstone envelope - mudstone envelope", clearly defining the adjacent geological envelopes corresponding to each initial boundary).

[0207] Preferably, the specific implementation process of step 5.4 is as follows: the object of processing is the initial stratigraphic boundary data (this data contains the initially extracted stratigraphic boundaries, but there may be two types of problems: one is that the boundary segment contradicts the contact relationship constraint of the strata, such as the dip angle of a certain boundary segment is 20°, which exceeds the range of [-15°, 15°] of the integrated contact constraint; the other is that there is a local interruption in the boundary, such as the boundary line segment appearing blank due to insufficient sampling density of the original data).

[0208] Based on the boundary topology verification and optimization layer in the trained third neural network model (this layer integrates topology rule verification and interpolation completion algorithms to ensure that the optimized boundary conforms to geological natural laws), topology verification is first performed: key parameters (tilt angle range, continuity ratio threshold) in the stratigraphic contact relationship constraints of the target work area are called, and each boundary segment in the initial stratigraphic boundary data is traversed: if a boundary segment is marked as "integrated contact boundary", but its tilt angle (calculated through the boundary coordinate sequence, such as the angle between two adjacent points (Xi,Zi) and (Xi+1,Zi+1)) is 18°, which exceeds the constraint range, then the segment is marked as a "topological contradiction segment"; if the continuity ratio (length of continuous segment / total length of segment) of a boundary segment is 70%, which is lower than the constraint threshold of 75%, it is also marked as a contradiction segment, and a list of topological contradictions is generated.

[0209] Next, the topological conflict segments are processed: For the marked conflict segments, the geological background of the target work area is considered for correction. If the conflict originates from local small fault interference, the contact relationship type of the segment is changed from "integrated contact boundary" to "fault contact boundary", and its tilt angle constraint is updated (the fault contact boundary constraint angle range is [-30°, 30°]). If the conflict originates from data error, the conflict segment is deleted, and the continuous valid boundary line segments at both ends are retained.

[0210] Then, boundary continuity optimization is performed: for boundaries that are still interrupted after correction (such as blank areas between two effective boundary segments), linear interpolation or trend interpolation (based on the trend of the boundary segments at both ends) is used to calculate the boundary coordinate points of the blank areas and add them to the boundary coordinate sequence to ensure the spatial continuity of the boundary.

[0211] Finally, all the corrected boundary fragments are integrated and sorted out again according to "boundary number - complete boundary coordinate sequence - contact relationship type - corresponding geological envelope pair" to generate the final stratigraphic boundary data of the geological envelope in the target work area (the boundary spatial morphology of this data conforms to the stratigraphic contact relationship constraints and has no obvious interruption, and can be directly used for subsequent boundary positioning and analysis of geological envelopes, which is different from the traditional boundary data that only relies on neural network prediction without verification, thus improving the reliability of boundary prediction).

[0212] The training of the third neural network model aims to enable the model to accurately predict the stratigraphic boundaries of geological envelopes from implicit topological structural representations. The training data consists of implicit topological structural representation samples and corresponding labeled stratigraphic boundaries (the labeled samples include the spatial coordinates of known stratigraphic boundaries and their corresponding implicit representation feature labels). During training, the implicit topological structural representation samples are first input into the implicit structural feature enhancement layer. The model learns the scalar gradient determination rules and feature enhancement weights for stratigraphic boundary regions, generating enhanced implicit structural feature samples. The multi-scale boundary feature capture layer learns the parameters of convolutional kernels of different sizes to master the extraction strategies for micro, meso, and macro-scale boundary features, generating multi-scale boundary feature set samples. The preliminary stratigraphic boundary prediction layer learns the mapping relationship between multi-scale features and boundary existence probabilities, outputting initial stratigraphic boundary data samples. The boundary topology verification and optimization layer learns the adaptation rules of stratigraphic contact relationship constraints and lithological sequence constraints, correcting topological contradictions in the initial boundary samples. Training employs the Dice loss function (adapting to the class imbalance problem in boundary prediction). The final layer boundary data samples output by the model are compared with the boundary coordinate labels in the labeled samples. The feature enhancement weights, convolution kernel parameters, prediction layer mapping coefficients, and validation rule parameters are iteratively adjusted through backpropagation. Training is completed when the loss function value converges and the spatial overlap between the model-predicted boundary and the labeled boundary meets the standard, enabling the model to predict reliable layer boundaries from implicit representations.

[0213] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting the stratigraphic boundaries of geological envelopes based on artificial intelligence, characterized in that, include: Step 1: Determine the seismic exploration data and well logging data for the target work area; Step 2: Extract seismic reflection features from seismic exploration data based on the trained first neural network model, and simultaneously extract well logging sequence features from well logging data using the trained second neural network model; Step 3: Generate a geological characteristic profile of the target work area based on seismic reflection characteristics and well logging sequence characteristics; Step 4: Based on the stratigraphic contact relationship constraints and lithological sequence constraints for the target work area, and according to the geological characteristic characterization, determine the implicit topological structural characterization of the geological envelope in the target work area; Step 5: Process the implicit topological structural representation based on the trained third neural network model to predict the stratigraphic boundaries of the geological envelope in the target work area. Step 2 involves extracting seismic reflection features from seismic exploration data based on the trained first neural network model, and simultaneously extracting well logging sequence features from well logging data using the trained second neural network model. Specifically, this includes: Step 2.1: Based on the first neural network model that has been trained, perform multi-scale feature extraction on the denoised seismic exploration data to obtain the initial seismic reflection features; Step 2.2: Based on the trained second neural network model, extract vertical sequence features from the well logging data to generate initial well logging sequence features; Step 2.3: Remove redundant features that are not related to the formation boundary from the initial seismic reflection features and initial well logging sequence features to generate the selected seismic reflection features and selected well logging sequence features; Step 3, based on seismic reflection characteristics and well logging sequence characteristics, generates a geological characteristic profile of the target work area, specifically including: Step 3.1: Based on the spatial coordinate system of the target work area, spatial coordinate anchoring is performed on the seismic reflection features and well logging sequence features. Then, the correlation weights are calculated based on the degree of correlation between the spatially anchored seismic reflection features and well logging sequence features and the formation boundary, so as to generate a feature correlation mapping table. Step 3.2: Based on the feature association mapping table, the seismic reflection features and well logging sequence features are weighted and fused, and contradictory features after fusion are eliminated by stratigraphic topology verification to generate a fused geological feature set; Step 3.3: Based on the geological background of the target work area, the fused geological feature set is spatially continuous and the geological significance of each feature is marked to generate a geological characteristic representation of the target work area. Step 4, based on the stratigraphic contact relationship constraints and lithological sequence constraints for the target work area, and according to the geological characteristic characterization, determines the implicit topological structural characterization of the geological envelope in the target work area. Specifically, based on the seismic exploration interpretation profile data of the target work area, stratigraphic contact relationship constraints and lithological sequence constraints for the target work area are constructed, and according to the geological characteristic characterization, the implicit topological structural characterization of the geological envelope in the target work area is determined. Step 5, which involves processing the implicit topological structural representation based on the trained third neural network model to predict the stratigraphic boundaries of the geological envelope in the target work area, specifically includes: Step 5.1: Based on the implicit construction feature enhancement layer in the trained third neural network model, perform boundary-sensitive feature enhancement on the implicit topological construction representation to generate enhanced implicit construction features. Step 5.2: Based on the multi-scale boundary feature capture layer in the trained third neural network model, perform multi-scale boundary feature extraction on the enhanced implicit construction features to generate a multi-scale boundary feature set. Step 5.3: Based on the preliminary prediction layer of the layer boundary in the trained third neural network model, perform preliminary prediction of the layer boundary on the multi-scale boundary feature set to generate initial layer boundary data. Step 5.4: Based on the boundary topology verification and optimization layer in the trained third neural network model, perform topology verification and optimization on the initial stratigraphic boundary data to generate the final stratigraphic boundary data of the geological envelope in the target work area.

2. The method according to claim 1, characterized in that, Step 1: Determine the seismic exploration data and well logging data for the target work area, specifically including: Step 1.1: Perform adaptive noise suppression processing on the original seismic exploration data of the target work area to generate denoised seismic exploration data; Step 1.2: Perform depth correction and outlier removal on the original logging data of the target work area to generate logging data correction results; Step 1.3: Perform spatial coordinate matching on the denoised seismic exploration data and the corrected well logging data to generate seismic exploration data and well logging data for the target work area.

3. The method according to claim 2, characterized in that, Step 1.1: Perform adaptive noise suppression processing on the original seismic exploration data of the target work area to generate denoised seismic exploration data. This specifically includes: Step 1.1.1: Divide the original seismic exploration data of the target work area into spatially continuous windows, calculate the amplitude standard deviation, frequency distribution characteristics and amplitude variation rate between adjacent windows for each window, and determine the statistical characteristics of seismic data windowing accordingly. Step 1.1.2: Based on the windowed statistical characteristics of seismic data, and combined with pre-labeled valid reflection signal samples and noise samples, the original seismic exploration data of the target work area is judged window by window for signal type determination. Dynamic weighted filtering is applied to windows marked as noise, and the original features are retained for windows marked as valid signals, so as to generate denoised seismic exploration data.

4. The method according to claim 1, characterized in that, Step 2.1: Based on the trained first neural network model, perform multi-scale feature extraction on the denoised seismic exploration data to obtain initial seismic reflection features, specifically including: Step 2.1.1: Based on the multi-scale convolutional feature extraction layer in the first neural network model that has been trained, perform feature extraction at different spatial scales on the denoised seismic exploration data to generate a preliminary multi-scale seismic feature map. Step 2.1.2: Based on the cross-scale feature attention fusion layer pair in the first neural network model that has been trained, calculate the attention weights and perform feature fusion on the preliminary multi-scale seismic feature map to generate the fused seismic feature map. Step 2.1.3: Based on the residual noise suppression layer in the first neural network model that has been trained, perform residual calculation on the fused seismic feature map to generate a denoised seismic feature map. Step 2.1.4: Based on the feature dimension compression and filtering layer in the first neural network model that has been trained, the denoised seismic feature map is subjected to dimension compression to remove redundant feature dimensions and effective feature filtering to retain features that are strongly correlated with the layer boundary, so as to generate initial seismic reflection features.

5. The method according to claim 1, characterized in that, Step 2.2: Based on the trained second neural network model, extract vertical sequence features from the well logging data to generate initial well logging sequence features, specifically including: Step 2.2.1: Based on the vertical window serialization layer in the trained second neural network model, perform vertical window partitioning and sequence transformation on the logging data to generate a set of vertical logging sequence fragments; Step 2.2.2: Based on the sequence residual denoising layer in the trained second neural network model, perform residual operations on the vertical logging sequence fragment set to generate a denoised vertical logging sequence fragment set. Step 2.2.3: Based on the multi-feature attention enhancement layer pairs in the trained second neural network model, calculate the attention weights for the denoised vertical logging sequence fragment set to generate the attention-enhanced logging sequence feature set; Step 2.2.4: Based on the sequence feature filtering layer in the trained second neural network model, perform feature correlation analysis and filtering on the attention-enhanced logging sequence feature set to generate initial logging sequence features.

6. The method according to claim 1, characterized in that, Step 3: Based on seismic reflection characteristics and well logging sequence characteristics, generate a geological characteristic profile of the target work area, which also includes: Step 3.0: Generate the regional geological background of the target work area based on the existing geological outcrop data and borehole core description data of the target work area and its surroundings.

7. The method according to claim 6, characterized in that, Step 3.0: Based on existing geological outcrop data and borehole core description data of the target work area and its surroundings, generate the regional geological background of the target work area, specifically including: Step 3.0.1: Classify and label the existing geological outcrop data and borehole core description data of the target work area and its surroundings according to lithology, sedimentary structure and fossil assemblage, so as to generate a classified geological basic dataset. Step 3.0.2: Perform vertical sequence correlation analysis on the lithological assemblages and sedimentary structures in the classified geological basic dataset to extract key features reflecting the sedimentary environment in order to generate a sedimentary environment feature set; Step 3.0.3: Based on the sedimentary environment feature set and combined with the existing tectonic movement records around the target work area, establish the temporal correlation between sedimentary environment features and tectonic movement stages to generate a tectonic evolution time series table. Step 3.0.4: Based on the sedimentary environment feature set and the tectonic evolution time series table, comprehensively verify the sedimentary environment features and tectonic evolution stages, and supplement the description of stratigraphic development patterns to generate the regional geological background of the target work area.

Images