A fault probability continuity enhancement method based on multi-directional line kernel and related equipment

Abstract

This invention discloses a method and related equipment for enhancing fault probability continuity based on multi-directional linear kernels. By using multi-directional linear kernel convolution and pixel-wise maximum value extraction, this invention can directionally aggregate discontinuous high-probability points along different dips, effectively solving the problem that single-directional filtering cannot adapt to faults with different dip angles. By estimating layered reflections using horizontal linear kernels and subtracting them from the directional response, the risk of strong reflections along layers being mistakenly linked as faults is effectively suppressed, reducing false anomaly interference. Multiplication gating operations are performed using attribute background maps, ensuring that the enhancement results are mainly retained in areas with structural basis, avoiding cross-regional misconnections without geological evidence. By weighted fusion and smoothing the output of the original fault probability and linear candidate probability, the invention improves fault continuity while preserving the original probability response characteristics, providing more intuitive and reliable auxiliary results for subsequent interpretation. This method can be widely applied in the field of data processing technology.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and related equipment for enhancing the probability continuity of faults based on multi-directional linear nuclei. Background Technology

[0002] In the interpretation of faults in 2D seismic profiles, a common practice is to first calculate fault probabilities or attributes such as coherence, variance, amplitude gradient, and neural network probability, and then have interpreters determine the fault location based on these results. In shallow gas seismic interpretation, shallow small faults, micro-fractures, and gas migration channels are often related to fault continuity; therefore, whether the candidate fault lines are continuous and clear directly affects subsequent tectonic interpretation.

[0003] In actual data, fault response is often not a complete and continuous line, but rather manifests as discontinuous high-probability points, short line segments, or local bright spots. Especially in areas with weak faults, small faults, or strong noise, fault probability maps are prone to problems such as numerous fragmented points, short line segments, and faults not forming a straight line.

[0004] If the original fault probability map is given directly to the interpreters, they still need to manually connect the fault points and delete false anomalies along the layers, which is a lot of work and highly subjective. Summary of the Invention

[0005] The main objective of this invention is to propose a method, apparatus, electronic device, storage medium, and program product for enhancing the probability continuity of tomography based on a multi-directional linear core, aiming to solve at least one problem of the prior art.

[0006] To achieve the above objectives, one aspect of the present invention proposes a method for enhancing the probability continuity of faults based on a multi-directional linear core, the method comprising: Obtain the original fault probability map, and perform normalization and outlier processing on the original fault probability map to obtain the standardized fault probability map. Based on the standardized fault probability map, the fault line seed probability is obtained by using the adaptive threshold method or combination method. The fault line seed probability is convolved with each linear kernel in the preset multi-directional linear kernel set, and then the maximum value of the convolution result is taken pixel by pixel to obtain the directional linear support map. The horizontal linear kernel is used to convolve the fault line seed probability to obtain the horizontal layered response map. The proportion of the horizontal layered response map is subtracted from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. An attribute background map is constructed based on preset seismic tectonic attributes. The target linear support map and the attribute background map are multiplied and gated to obtain a linear candidate probability map. The standardized fault probability map and the linear candidate probability map are weighted and fused, and then smoothed to obtain a continuously enhanced fault probability map.

[0007] In some alternative approaches, the original fault probability map is normalized and outlier processed to obtain a standardized fault probability map, including the following steps: The values ​​of the original fault probability map are normalized to a preset range, and outliers that exceed the preset range are truncated or replaced to obtain a standardized fault probability map.

[0008] In some alternative approaches, fault line seed probabilities are constructed based on standardized fault probability maps using adaptive thresholding or a combination of methods, including the following steps: If the adaptive threshold method is used, the target quantile is obtained by quantization based on the standardized fault probability map, and the mean and standard deviation of the probability map are obtained by statistics. The candidate threshold is obtained by multiplying the standard deviation of the probability plot by a preset coefficient and then adding it to the mean of the probability plot. The larger value between the target quantile and the candidate threshold is used as the initial threshold. The initial threshold is constrained by preset upper and lower limits to obtain the final threshold. The standardized fault probability map is binarized according to the final threshold to obtain a binary seed map, which is then transformed to obtain the fault line seed probability. If a combination method is used, the preset fault line is used as the main seed, and the standardized fault probability map is used as the auxiliary response. The fault line seed probability is obtained through linear weighted fusion and normalization.

[0009] In some alternative approaches, the method may also include the following steps: Obtain the sampling parameters and track spacing of the work area corresponding to the original fault probability map; where the work area sampling parameters include the time sampling interval or the depth sampling interval; Obtain the sampling window along the fault direction, the physical width of the lateral swing, and the maximum expected fault dip angle in the work area; The ratio of the sampling window to the sampling parameters of the work area is rounded to obtain the first value; The first value is incremented to obtain the second value, and the smallest odd number not less than the second value is taken as the length of the linear core. The ratio of the lateral swing physical width to the track spacing is rounded to obtain the third value; The third value is incremented to obtain the fourth value, and the smallest odd number not less than the fourth value is taken as the width of the linear kernel. The fifth value is determined based on the ratio of the sampling parameters of the work area to the track spacing, and the sixth value is determined by performing a tangent calculation on the preset maximum expected fault dip angle of the work area. The maximum slope is obtained by multiplying the fifth and sixth values. Using the positive and negative values ​​of the maximum slope as boundaries, and then combining them with a preset slope step size, a set of directions is constructed to generate a multi-directional linear kernel set. In this set of directions, each slope corresponds to a linear nucleus, and the length and width of each linear nucleus are determined based on the length and width of the linear nucleus.

[0010] In some alternative methods, the target linear support map after suppressing the horizontal response is obtained by subtracting the scale value of the horizontal layered response map from the directional linear support map, including the following steps: The seventh value is obtained by multiplying the horizontal layered response diagram with the preset horizontal response suppression coefficient; The eighth value is obtained based on the difference between the directional linear support diagram and the seventh value; The target linear support diagram after the suppression level response is determined based on the maximum value between the eighth value and the preset lower cutoff limit.

[0011] In some alternative methods, an attribute background map is constructed based on preset seismic tectonic attributes. The target linear support map is then multiplied and gated with the attribute background map to obtain a linear candidate probability map. This includes the following steps: The attribute background map is constructed by weighting and normalizing the various attribute features in the preset seismic tectonic attributes according to fixed prior weights. Among them, the types of preset seismic tectonic attributes include multi-scale coherent discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attributes, and tectonic deformation; Based on the attribute background graph, a linear lower bound gating term is constructed using a preset minimum retention amount; Normalize the product of the target linear support graph and the linear lower bound gating term to obtain the linear candidate probability graph.

[0012] In some alternative approaches, a weighted fusion of the standardized fault probability map and the linear candidate probability map is performed, followed by smoothing to obtain a continuously enhanced fault probability map, including the following steps: Obtain the preset global linear candidate weights, and subtract the global linear candidate weights from 1 to obtain the original probability weights. The ninth value is obtained by multiplying the original probability weights by the standardized fault probability map. The tenth value is obtained by multiplying the global linear candidate weights by the linear candidate probability map. The sum of the ninth and tenth values ​​is normalized to obtain the fusion result; Spatial smoothing is performed on the fusion results to obtain a continuously enhanced fault probability map.

[0013] To achieve the above objectives, another aspect of the present invention proposes a fault probability continuity enhancement device based on a multi-directional linear core, the device comprising: The first module is used to obtain the original fault probability map, normalize the original fault probability map and process outliers to obtain a standardized fault probability map. The second module is used to construct fault line seed probabilities based on standardized fault probability maps using adaptive thresholding or combination methods. The third module is used to convolve the fault line seed probability with each linear kernel in the preset multi-directional linear kernel set, and then take the maximum value of the convolution result pixel by pixel to obtain the directional linear support map. The fourth module is used to convolve the fault line seed probability with a horizontal linear kernel to obtain a horizontal layered response map. The proportional value of the horizontal layered response map is subtracted from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. The fifth module is used to construct an attribute background map based on preset seismic tectonic attributes, and to perform multiplication gating operation on the target linear support map and the attribute background map to obtain a linear candidate probability map. The sixth module is used to perform weighted fusion of the standardized fault probability map and the linear candidate probability map, and then obtain the continuously enhanced fault probability map through smoothing.

[0014] To achieve the above objectives, another aspect of the present invention provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned method.

[0015] To achieve the above objectives, another aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0016] To achieve the above objectives, another aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0017] This invention provides at least the following beneficial effects: It offers a method, apparatus, electronic device, storage medium, and program product for enhancing fault probability continuity based on multi-directional linear kernels. The scheme obtains a standardized fault probability map by acquiring an original fault probability map, normalizing and processing outliers, and then constructing a fault line seed probability using an adaptive thresholding or combination method based on the standardized fault probability map. The fault line seed probability is convolved with each linear kernel in a preset multi-directional linear kernel set, and the maximum value is taken pixel-by-pixel from the convolution result to obtain a directional linear support map. A horizontal linear kernel is used to convolve the fault line seed probability to obtain a horizontal layered response map. The proportional value of the horizontal layered response map is subtracted from the directional linear support map to obtain a target linear support map after suppressing the horizontal response. An attribute background map is constructed based on preset seismic tectonic attributes. A multiplication-gated operation is performed between the target linear support map and the attribute background map to obtain a linear candidate probability map. The standardized fault probability map and the linear candidate probability map are weighted and fused, and then smoothed to obtain a continuously enhanced fault probability map. This invention utilizes multi-directional linear kernel convolution and pixel-wise maximum value extraction to directionally aggregate discontinuous high-probability points along different dips. Compared to ordinary smoothing, this method more effectively restores fault continuity and solves the problem that single-directional filtering cannot adapt to faults with different dip angles. By estimating layered reflections using a horizontal linear kernel and subtracting them from the directional response, it effectively suppresses the risk of strong reflections along layers being mistakenly linked as faults, reducing false anomaly interference. Multiplication gating is performed using the attribute background map, ensuring that the enhancement results are mainly retained in areas with structural basis, avoiding cross-regional misconnections without geological basis. By weightedly fusing the original fault probability and linear candidate probability and smoothing the output, it improves fault continuity while preserving the original probability response characteristics, providing more intuitive and reliable auxiliary results for subsequent interpretation. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an implementation environment for the fault probability continuity enhancement method based on multi-directional linear cores provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the method for enhancing the probability continuity of faults based on multi-directional linear nuclei provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the overall process of the fault probability continuity enhancement method based on multi-directional linear core provided in the embodiments of the present invention; Figure 4 This is a schematic diagram of an example of a multi-directional linear core provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating an example of the horizontal response suppression and attribute background constraint process provided in an embodiment of the present invention; Figure 6This is a schematic diagram illustrating an example of a continuous enhancement result output process provided in an embodiment of the present invention; Figure 7 This is a schematic diagram showing a comparison example of the continuity enhancement of section 073 before and after the present invention. Figure 8 This is a schematic diagram showing a comparison example of the continuity enhancement of section 091 before and after according to an embodiment of the present invention; Figure 9 This is a schematic diagram of a three-section comparison example with a partially enlarged cross-section 091 provided in an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0020] It is understood that the terms "first," "second," etc., used in this invention may be used to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of embodiments of this invention, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words "if" or "when" as used herein may be interpreted as "when," "in response to determination," or "in the event of a determination."

[0021] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0022] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this invention is for descriptive purposes only and is not intended to limit the invention.

[0023] To facilitate understanding of the technical solution of this invention, the key terms that may be involved in the technical solution of this invention will first be explained: Fault probability map: the probability that each sampling point belongs to a fault or fracture zone, the value is usually normalized to 0 to 1.

[0024] Linear kernel: A thin, two-dimensional convolutional template used to detect or connect high-probability points arranged along a specific direction.

[0025] Multidirectional linear nuclei: A group of linear nuclei with different inclinations, used to cover fault responses with different dip angles.

[0026] Horizontal linear kernel: A special case of multidirectional linear kernel with slope s=0, i.e. a linear convolution template extending along the time / depth direction, used to estimate near-horizontal layered reflection response.

[0027] Horizontal response suppression: Horizontal linear kernels are used to estimate false anomalies along the layers, and the corresponding parts are subtracted from the directional linear responses to reduce the risk of misinterpreting laminar reflections as faults.

[0028] Attribute background constraints: By utilizing structural attributes such as coherence discontinuity, dip gradient, and Frangi linear enhancement, the linear connection results are constrained to make the connection results more consistent with the seismic tectonic background.

[0029] In related technologies, ordinary smoothing can only blur the probability map and cannot selectively connect fault points along the fault direction; single-direction filtering is difficult to adapt to faults with different dip angles; simple linear enhancement can easily connect horizontal layered reflections into false faults; when there is a lack of structural attribute background constraints, the connection results may cross areas without geological basis.

[0030] In view of this, this invention provides a method and related equipment for enhancing fault probability continuity based on multi-directional linear kernels. This method obtains an original fault probability map, normalizes and processes outliers to obtain a standardized fault probability map; based on the standardized fault probability map, an adaptive thresholding method or a combination method is used to construct fault line seed probabilities; the fault line seed probabilities are convolved with each linear kernel in a preset multi-directional linear kernel set, and the maximum value is taken pixel by pixel to obtain a directional linear support map; a horizontal linear kernel is used to convolve the fault line seed probabilities to obtain a horizontal layered response map; the proportion of the horizontal layered response map is subtracted from the directional linear support map to obtain a target linear support map after suppressing the horizontal response; an attribute background map is constructed based on preset seismic tectonic attributes; the target linear support map and the attribute background map are multiplied and gated to obtain a linear candidate probability map; the standardized fault probability map and the linear candidate probability map are weighted and fused, and then smoothed to obtain a continuously enhanced fault probability map. This invention utilizes multi-directional linear kernel convolution and pixel-wise maximum value extraction to directionally aggregate discontinuous high-probability points along different dips. Compared to ordinary smoothing, this method more effectively restores fault continuity and solves the problem that single-directional filtering cannot adapt to faults with different dip angles. By estimating layered reflections using a horizontal linear kernel and subtracting them from the directional response, it effectively suppresses the risk of strong reflections along layers being mistakenly linked as faults, reducing false anomaly interference. Multiplication gating is performed using the attribute background map, ensuring that the enhancement results are mainly retained in areas with structural basis, avoiding cross-regional misconnections without geological basis. By weightedly fusing the original fault probability and linear candidate probability and smoothing the output, it improves fault continuity while preserving the original probability response characteristics, providing more intuitive and reliable auxiliary results for subsequent interpretation.

[0031] It is understood that the tomographic probability continuity enhancement method based on multi-directional linear nuclei provided by this invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet, laptop, or desktop computer, but it is not limited to these.

[0032] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided by an embodiment of the present invention. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0033] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0034] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0035] Terminal 102 can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the invention does not impose any limitations.

[0036] For example, based on Figure 1 The implementation environment shown in this embodiment of the invention provides a method for enhancing the continuity of tomographic probability based on a multi-directional linear kernel. The following description uses the application of this method in server 101 as an example. It can be understood that this method can also be applied in terminal 102.

[0037] Reference Figure 2 , Figure 2 This is an optional flowchart of a method for enhancing the continuity of tomographic probability based on a multi-directional linear nucleus provided in an embodiment of the present invention. The execution subject of this method can be any of the aforementioned computer devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S100 to S600.

[0038] Step S100: Obtain the original fault probability map, normalize and process outliers on the original fault probability map to obtain the standardized fault probability map. It should be noted that in some embodiments, normalizing the original fault probability map and processing outliers to obtain a standardized fault probability map may include the following steps: normalizing the values ​​of the original fault probability map to a preset range, and truncating or replacing outliers that exceed the preset range to obtain a standardized fault probability map.

[0039] For example, in some specific implementations, an original fault probability map is obtained, which may come from multi-attribute fusion, neural network prediction or other fault identification methods. It is first normalized to 0 to 1, and outliers are robustly processed to obtain a standardized fault probability map.

[0040] Step S200: Based on the standardized fault probability map, the fault line seed probability is constructed using an adaptive threshold method or a combination method. It should be noted that in some embodiments, step S200 may include the following steps: if an adaptive thresholding method is used, the target quantile is obtained based on the standardized fault probability map, and the probability map mean and standard deviation are statistically obtained; wherein, the probability map mean represents the mean obtained based on the standardized fault probability map, and the probability map standard deviation represents the standard deviation obtained based on the standardized fault probability map; the probability map standard deviation is multiplied by a preset coefficient, and then added to the probability map mean to obtain a candidate threshold; the larger value between the target quantile and the candidate threshold is used as the initial threshold, and the initial threshold is constrained by preset upper and lower limits to obtain the final threshold; the standardized fault probability map is binarized according to the final threshold to obtain a binary seed map and transformed to obtain the fault line seed probability; if a combination method is used, the preset fault line is used as the main seed, the standardized fault probability map is used as the auxiliary response, and the fault line seed probability is obtained through linear weighted fusion and normalization processing.

[0041] For example, in some specific implementations, fault line seeds can be constructed by generating binary fault line seeds based on the high percentile threshold of the probability map, or by combining existing predicted fault lines with the original probability map according to weights, and then smoothing them in the time direction and the direction of the track to obtain an initial seed probability with enhanced continuity.

[0042] The high percentile threshold is not a globally fixed constant, but is adaptively determined based on the statistical distribution of the current input probability map. Let the normalized original fault probability map be P0(t,x) (P0 is the standardized fault probability map, and (t,x) represents the index of the sampling point and the seismic trace). The larger value of the probability map p-quantile and the mean-standard deviation constraint can be taken as the initial threshold. Then, upper and lower limit constraints are used to avoid introducing noise due to an excessively low threshold or missing weak faults due to an excessively high threshold.

[0043] (1) (2) in, As the initial threshold, For probability graphs Quantiles and These represent the mean and standard deviation of the probability plot, respectively. p can be taken as 90%-95%. A value of 0.5-1.0 is acceptable. and (That is, the preset upper and lower limits) The upper limit, (As the lower limit) can be limited to the range of approximately 0.55-0.90 based on the data noise level. This is a binary fault line seed map (i.e., a binary seed map). After threshold candidates are generated, isolated small connected regions can be removed while seeds with linear extension features are retained to improve robustness. Among these, It is an indicator function, and its operation rule is: when the condition inside the square brackets is true, I = 1; when the condition inside the square brackets is false, I = 0. Therefore, equation (2) = Its actual function is: for each position Determine the original probability value of this point. Has the final threshold been reached (i.e.) If this is achieved, then the binary seed graph is... The value at this point is 1 (considered a seed point), otherwise it is 0 (considered a non-seed point).

[0044] If there are already predicted fault lines or artificially corrected fault lines If a predefined fault line is used, a combination method is preferred over a threshold method alone. The combination method uses the predefined fault line (which can be a linear mask of the predefined fault line) as the main seed and the original probability map as a weak response supplement. In the current MATLAB implementation, the seed probability weight is set to 0.20 by default, meaning that the original probability is only used as a 20% auxiliary term to avoid the noisy probability map dominating the seed position.

[0045] (3) in, This is a seed probability map for fault lines. For normalization, These are auxiliary weighting coefficients for the original probability map; the engineering judgment criteria are as follows: when only a probability map exists and there are no reliable fault lines, the adaptive threshold method is used; when there are existing predicted fault lines, manually interpreted lines, or results from the previous round and their locations are reliable, the combined method is used; when the data noise is strong but the fault line skeleton is relatively reliable, the weighting is reduced. When there are many weak faults and the probability map response is reliable, the value can be appropriately increased. The range is 0.10-0.30. The current implementation uses manually adjustable parameters and does not automatically estimate weights based on the signal-to-noise ratio.

[0046] Step S300: Convolve the fault line seed probability with each linear kernel in the preset multi-directional linear kernel set, and then take the maximum value of the convolution result pixel by pixel to obtain the directional linear support map. It should be noted that in some embodiments, the method may further include the following steps: obtaining the sampling parameters and trace spacing of the work area corresponding to the original fault probability map; wherein, the sampling parameters of the work area include the time sampling interval or the depth sampling interval; obtaining the sampling window along the fault direction, the physical width of the lateral swing, and the maximum expected fault dip angle of the work area; rounding the ratio of the sampling window to the sampling parameters of the work area to obtain a first value; incrementing the first value to obtain a second value, and taking the smallest odd number not less than the second value as the linear core length; rounding the ratio of the physical width of the lateral swing to the trace spacing to obtain a third value. The numerical values ​​are incremented to obtain the fourth value, and the smallest odd number not less than the fourth value is taken as the width of the linear core. The fifth value is determined based on the ratio of the sampling parameters of the work area to the track spacing. The sixth value is determined by performing a tangent operation on the preset maximum expected fault dip angle of the work area. The maximum slope is obtained by multiplying the fifth and sixth values. The positive and negative values ​​of the maximum slope are used as boundaries, and then a direction set is constructed in combination with the preset slope step size to generate a multi-directional linear core set. Each slope of the direction set corresponds to a linear core, and the length and width of each linear core are determined based on the linear core length and the linear core width.

[0047] For example, in some specific implementations, constructing a multi-directional linear core can be specifically achieved as follows: Multiple sets of slender linear nuclei with different dip directions are configured, such as negative dip, near-vertical dip, and positive dip. Each linear nucleus aggregates a probability response along its corresponding direction, connecting discontinuous high-probability points along possible fault directions.

[0048] The length and width of the linear kernel are expressed in the program as the number of sampling points / channels. Its physical span needs to be calculated based on the time sampling interval, depth sampling interval, or channel spacing of the input profile. Let the time sampling interval be... Lane spacing is If processed in the time domain, then the length Corresponding agreement Time span, width Corresponding agreement The lateral span; if it has been converted to the depth domain, the depth sampling interval can be used. Alternative .

[0049] (4) In the formula, This indicates rounding to the nearest integer. This indicates taking the smallest odd number that is not less than the number inside the parentheses. For the time window or depth window that is desired to cover along the fault direction, The physical width is designed to allow for lateral swaying. Therefore, when the track spacing in work area A is 12.5m and the track spacing in work area B is 25m, the same "13-21 track" width cannot be mechanically applied; the number of tracks should be recalculated according to the target physical width, and an odd number not less than 5 should be used to ensure that the core center is clearly defined.

[0050] The values ​​such as -1.20 and -0.90 in the tendency set are not in radians or physical angles, but rather slope parameters in the two-dimensional pixel / sampling grid. According to the definition of line_kernel in this implementation, the slope s represents the offset of the center of the linear kernel in the channel column number when the row number changes by one time / depth sampling point, i.e., Δcol / Δrow.

[0051] (5) In the formula, Represents the column (track) coordinates of a pixel within the linear kernel. Represents the column coordinates of the center pixel of the kernel. Represents the row (time / depth) coordinates of a pixel within the kernel. Represents the row coordinates of the center pixel of the kernel. This is the equivalent tilt angle relative to the vertical time / depth axis; when the time profile is not converted to depth, it can be used. Or display coordinate scale alternative Perform pixel angle calibration. The direction set is not a fixed, empirically unchangeable parameter; it can be determined based on the maximum expected fault dip angle in the work area. Calculate the maximum slope with respect to the sampling interval, and then adjust the step size accordingly. Construct a set of directions.

[0052] (6) Finally, the directional linear response is calculated: the seed probability is convolved with the linear kernel of each direction, and the maximum value of the response in each direction is taken to obtain the directional linear support map.

[0053] Step S400: Convolve the fault line seed probability with a horizontal linear kernel to obtain a horizontal layered response map. Subtract the proportion of the horizontal layered response map from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. For example, the horizontal linear kernel is a special case of the multi-directional linear kernel set when the slope s=0. Its length is the same as that of the multi-directional linear kernel, and its width uses a fixed number of sampling points (e.g., 3 sampling points).

[0054] It should be noted that in some embodiments, subtracting the proportional value of the horizontal layered response map from the directional linear support map to obtain the target linear support map after suppressing the horizontal response may include the following steps: obtaining a seventh value based on the product of the horizontal layered response map and a preset horizontal response suppression coefficient; obtaining an eighth value based on the difference between the directional linear support map and the seventh value; and determining the target linear support map after suppressing the horizontal response based on the maximum value of the eighth value and a preset truncation lower limit.

[0055] For example, in some specific embodiments, suppressing the horizontal layered response can be achieved as follows: The response along the layer is calculated using a horizontal linear kernel, and the horizontal response is subtracted from the directional linear support plot by a predetermined proportion, which can be expressed as follows: ,in It is a multidirectional linear nuclear response. It is a horizontal linear response. This is the horizontal response suppression coefficient. This can reduce the possibility of strong reflection layers and layered noise being mistakenly identified as faults.

[0056] In the current implementation, the horizontal response suppression coefficient This is a globally adjustable constant that does not dynamically change with time windows or spatial locations. The default value for the continuity enhancement function is a=0.88; in weaker neural network probabilistic post-processing functions, a=0.65 can be used. This coefficient is used to subtract the layer-by-layer response from the normalized directional response, and the specific calculation is as follows.

[0057] (7) When used in different work areas, small-step calibration can be performed within the range of 0.65-0.88: first observe the horizontal response R. hor The distribution of strong layered bands is then selected, and the minimum a value that can suppress most of the high values ​​along the layer while retaining the known low-angle fractures is chosen. If artificial fault labels already exist, fault line continuity, false detection rate along the layer, or F1-score can be used to determine the fault.

[0058] The horizontal linear kernel and the directional linear kernel share the same line length to ensure that they compete on the same layer length scale; however, the width of the horizontal kernel is not 13-21 channels, nor is it equal to the width of the directional linear kernel. The current implementation uses a horizontal kernel of 3×lineLength, that is, only 3 sampling point thicknesses are retained in the time / depth direction, which is used specifically for estimating the near-horizontal layered response.

[0059] Step S500: Construct an attribute background map based on preset seismic tectonic attributes, and perform a multiplication gating operation on the target linear support map and the attribute background map to obtain a linear candidate probability map. In some optional implementations, firstly, two-dimensional seismic profile data corresponding one-to-one with the spatial location of the original fault probability map is acquired; or, various attribute features pre-calculated from the two-dimensional seismic profile data and registered with the original fault probability map are acquired. If the acquired data is two-dimensional seismic profile data, amplitude debiasing and robust normalization preprocessing are performed on the two-dimensional seismic profile data. Then, multi-scale local coherence analysis, Hessian-Frangi linear analysis, dip gradient analysis, multi-directional linear enhancement processing, and dip inconsistency, dip gradient and two-dimensional curvature weighted fusion are respectively used to obtain multi-scale coherence discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attributes and tectonic deformation attributes to determine various attribute features.

[0060] It should be noted that in some embodiments, step S500 may include the following steps: weighting and normalizing the various attribute features in the preset seismic tectonic attributes according to fixed prior weights to construct an attribute background map; wherein, the types of preset seismic tectonic attributes include multi-scale coherent discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attributes, and tectonic deformation; based on the attribute background map, constructing a linear lower bound gating term using a preset minimum retention amount; normalizing the product of the target linear support map and the linear lower bound gating term to obtain a linear candidate probability map.

[0061] For example, in some specific implementations, introducing attribute background constraints to generate a linear candidate probability map can be achieved as follows: An attribute background map is constructed by multi-scale coherent discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attribute, and structural deformation attribute. The attribute background map is formed by weighting each attribute. Then, the target linear support is multiplied by the linear lower bound gating term formed by the attribute background map, so that the connection result is mainly retained in the region with structural basis.

[0062] None of the above five types of attributes require additional on-site data collection; their source is a two-dimensional seismic amplitude profile that is registered point-by-point with the original fault probability map at time / depth sampling points and trace numbers, or pre-calculated from this profile and correlated with the original fault probability. Figure 1 The input attribute map is then used. If the original seismic profile is input, the median amplitude of each trace is subtracted first, and then divided by the 98th percentile of the absolute amplitude of the entire profile. Subsequent attribute maps are then truncated and normalized using the 2nd and 98th percentiles to ensure they fall within the range of 0 to 1. Specifically: Multiscale coherent discontinuity D ms This attribute is derived based on existing local coherence analysis. Within the h×w local window, let N = h×w and A be the preprocessed seismic amplitude, then calculate S. h,w=(ΣA)² / [NΣ(A²)+ε], then take D h,w =Norm(1-S h,w This embodiment uses 5×5, 9×7, and 15×11 windows respectively, and presses D. ms =Norm(0.40D 5×5 +0.35D 9×7 +0.25D 15×11 The fusion aims to simultaneously preserve small-scale fractures and longer-scale discontinuous responses.

[0063] The input source for Hessian-Frangi linear enhancement F: Frangi analysis belongs to the existing Hessian linear structure analysis, but this embodiment constructs a special input seed B for fault attributes. Specifically, it is first derived from D... ms Inconsistent tilt angles D dip Phase discontinuity D phase and transverse amplitude gradient G amp According to B=Norm(0.45D) ms +0.25D dip +0.20D phase +0.10G amp ) Fusion; where D phase G is obtained by multiplying the instantaneous envelope reliability by the phase difference between adjacent channels of each Hilbert analytical signal and then smoothing it locally. amp It is obtained by normalizing the absolute value of the directional gradient of the earthquake amplitude.

[0064] The process of Hessian-Frangi linear enhancement F: Gaussian smoothing is applied to B at σ=1, 2, and 3 respectively, and the two-dimensional Hessian matrix is ​​calculated to obtain the eigenvalues ​​sorted by absolute value. |≤| |;Let R B =| | / (| |+ε),S H = Calculate the response F at each scale σ =exp[-R B ² / (2β²)]·{1-exp[-S H 2 / (2c²)]}. In this embodiment, ε is a very small positive regularization constant (regularization parameter), whose core meaning and purpose is to prevent the denominator from being zero and to ensure the stability of numerical calculation. β=0.5, and c takes the value of S. H The 90th percentile is obtained by taking the maximum value for each pixel at each scale, multiplying it by B, and normalizing it to get F.

[0065] Inclination gradient G dipThis attribute is derived based on existing seismic reflection dip and gradient analysis. The time / depth gradient g is calculated from the preprocessed seismic data. t and gradient g x Perform local smoothing and calculate the local tilt angle θ = atan2(-g) x ,g t +ε); then smooth sinθ and cosθ respectively, and use the circumferential mean to obtain the average tilt angle. Finally, press G dip =Norm{ The intensity of the tilt angle abrupt change is obtained.

[0066] Linear enhancement attribute L: This attribute is the result of the combined processing designed for fault probability in this invention. Based on the base probability P... b For input; P b A standardized fault probability map or a multi-attribute fusion probability map from the preceding stage can be used. Let P... b Convolve each kernel with a linear kernel of length 25, width 15, and slope set {-0.70, -0.35, 0, 0.35, 0.70}, and take the maximum value pixel by pixel to obtain R. dir Simultaneously, R is obtained by 3×25 level kernel convolution. hor Then calculate L = Norm{Smooth5×3[max(R dir -0.75R hor ,0)·P b This is to highlight the slender, tilted or near-vertical response and suppress the response along the layer. The kernel size and coefficients mentioned above are parameters for this embodiment and can be adjusted according to the sampling interval and data quality.

[0067] Structural deformation property S def This attribute is a structural deformation synthesis attribute proposed by this invention. First, D is obtained from the circumferential dispersion of the local tilt angle. dip =Norm[1- ]; then, based on the smoothed local slope k=-g x / (g t +ε) Calculate the two-dimensional curvature K=Norm{| p / x| / [(1+k²)^(3 / 2)+ε]};Finally press S def =Norm(0.40D dip +0.30G dip +0.30K) weighted fusion. This property is used to simultaneously characterize tilt discrepancies, tilt abrupt changes, and in-phase axis bending.

[0068] The above D ms F and G dipThese are specific implementations of commonly used local coherence, Hessian-Frangi, and tilt gradient analyses, respectively; L and S def This invention involves a combination of these basic attributes. All five attribute maps maintain the same row and column dimensions and coordinate correspondence as the original fault probability map, and then attribute background maps are constructed according to the fixed prior weights in equation (8). If these registration attribute maps have been provided in advance by the system, the continuity enhancement module can read them directly without recalculation.

[0069] The preferred implementation uses a weighted summation with fixed prior weights to construct the attribute background map, and then applies multiplicative gating to the directional linear supports; it does not arbitrarily choose between product fusion and weighted summation, nor does it automatically determine the weights through PCA or machine learning. The currently implemented weight vector is as follows.

[0070] (8) in, For attribute background image, For multi-scale coherent discontinuities, F represents Hessian-Frangi linear enhancement. For the slope gradient, To enhance the properties of fault lines, These are structural deformation properties. The weights mentioned above are derived from prior experience regarding fault boundaries, slender fractures, and structural deformation sensitivity, and do not represent PCA principal component loads.

[0071] (9) in, This is a linear candidate probability plot. The minimum retention is 0.35; this minimum retention of the attribute background is not a cumulative energy percentage, nor does it directly truncate regions where C is below 0.35; it is a linear lower bound gating term. Even if C=0, the region still retains 35% of the linear support, preventing the true faults in deep, weakly attributed or low signal-to-noise ratio regions from being completely suppressed; as C increases, the retention factor smoothly increases from 0.35 to 1.

[0072] In some alternative implementations, the attribute background constraints can be used in different combinations, such as coherent attributes, curvature attributes, tilt gradients, structural tensor anisotropy, or neural network intermediate responses.

[0073] Step S600: The standardized fault probability map and the linear candidate probability map are weighted and fused, and then the continuously enhanced fault probability map is obtained through smoothing. It should be noted that in some embodiments, step S600 may include the following steps: obtaining a preset global linear candidate weight, subtracting the global linear candidate weight from 1 to obtain the original probability weight; obtaining a ninth value based on the product of the original probability weight and the standardized fault probability map; obtaining a tenth value based on the product of the global linear candidate weight and the linear candidate probability map; normalizing the sum of the ninth and tenth values ​​to obtain the fusion result; and performing spatial smoothing on the fusion result to obtain the continuously enhanced fault probability map.

[0074] For example, in some specific implementations, the generation of the probability of continuous enhancement faults can be achieved as follows: The original fault probability and the linear candidate probability are fused according to weights and then smoothed appropriately to output a continuously enhanced fault probability map; further, fault lines can be extracted according to thresholds to output a raster map or vector map.

[0075] The current preferred implementation uses a fixed-ratio fusion instead of dynamically changing weights based on local attribute background values. The local construction confidence has already been entered into equation (9). Therefore, the final fusion stage maintains a global linear candidate weight. (Used to control the proportion of linear candidate probability maps in the final fusion result) to reduce parameter coupling.

[0076] (10) In the formula, For the fusion result, Input the original fault probability (using the standardized fault probability map after standardization processing). The linear candidate probability (i.e., the linear candidate probability map) is obtained after directional linear support, horizontal suppression, and attribute background constraints. The default value is 0.80, which means that the goal of this stage is to improve the continuity of fault candidate lines, and linear candidates that have passed the attribute background constraints are given priority. The remaining 0.20 retains the original probability to avoid over-connection that leads to deviation from the original fault response.

[0077] To explain in detail the principle of the technical solution of the present invention, the overall process of the present invention will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and should not be regarded as a limitation of the present invention.

[0078] To address the shortcomings of existing technologies, an automatic fault probability continuity enhancement method is needed to organize the existing fault probability response into a more continuous and easily interpretable fault line response.

[0079] Existing technical solutions primarily target the automatic interpretation of 3D stratigraphy. Their continuity enhancement involves processing fault attribute data along a single time path using a class u'(t)^2-u'(t-1)u'(t+1), then calculating the mean, variance, and probability values ​​to assist in 3D stratigraphic tracking. This invention differs in its object and process: the input is a 2D fault probability map or fault line seed. Its core principles are multi-directional linear kernel convolution, maximum response in each direction, horizontal stratigraphic response suppression, attribute background constraints, and fusion of original probability / linear candidate probability. This results in a more continuous fault probability or fault line outcome, rather than stratigraphic tracking.

[0080] This invention provides a method for enhancing the continuity of fault probability based on multi-directional linear kernels, which can be used as a post-processing module for multi-attribute fault probability, neural network fault probability, or manually annotated fault lines. Figure 3 The diagram shows an overall flowchart example of the fault probability continuity enhancement method based on a multi-directional linear core according to the present invention. In some specific embodiments, the fault probability continuity enhancement method based on a multi-directional linear core can be implemented through the following steps: Step S1: Input the original fault probability map. This probability map can come from multi-attribute fusion, neural network prediction, or other fault identification methods. First, normalize it to 0 to 1, and robustly handle outliers.

[0081] Step S2: Construct fault line seeds. Binary fault line seeds can be generated based on the high percentile threshold of the probability map, or existing predicted fault lines can be combined with the original probability map according to weights, followed by time and trajectories smoothing to obtain initial seed probabilities with enhanced continuity.

[0082] In practice, the high percentile threshold is not a globally fixed constant, but is adaptively determined based on the statistical distribution of the current input probability map. Let the normalized original fault probability map be P0(t,x). The larger value of the probability map p-quantile and the mean-standard deviation constraint can be taken as the initial threshold. Then, upper and lower limit constraints are used to avoid introducing noise due to an excessively low threshold or missing weak faults due to an excessively high threshold.

[0083] (1) (2) in, For probability graphs Quantiles and These represent the mean and standard deviation of the probability plot, respectively. p can be taken as 90%-95%. A value of 0.5-1.0 is acceptable. , The threshold can be limited to approximately 0.55-0.90 based on the data noise level. After threshold candidates are generated, isolated small connected components can be removed while seeds with linear extension features are retained to improve robustness.

[0084] If there are already predicted fault lines or artificially corrected fault lines Therefore, this embodiment prioritizes the combination method rather than relying solely on the threshold method. The combination method uses a linear mask as the main seed and the original probability map as a supplementary weak response. In the current MATLAB implementation, the seed probability weight is set to 0.20 by default, meaning that the original probability is only used as a 20% auxiliary term to avoid the noisy probability map dominating the seed position.

[0085] (3) The engineering evaluation criteria are as follows: When only a probability map is available and no reliable fault lines are found, an adaptive threshold method is used; when predicted fault lines, manually interpreted lines, or previous results are available and their locations are reliable, a combined method is used; when data noise is high but the fault line skeleton is relatively reliable, the threshold is lowered. When there are many weak faults and the probability map response is reliable, the value can be appropriately increased. The range is 0.10-0.30. The current implementation uses manually adjustable parameters and does not automatically estimate weights based on the signal-to-noise ratio.

[0086] Step S3: Construct a multi-directional linear nucleus. For example... Figure 4 The image shows an example of a multi-directional linear nucleus. Specifically, multiple sets of slender linear nuclei with different dip directions can be set, such as negative dip, near-vertical dip, and positive dip. Each linear nucleus aggregates a probability response along its corresponding direction, connecting discontinuous high-probability points along possible fault directions.

[0087] The length and width of the linear kernel are expressed in the program as the number of sampling points / channels. Its physical span needs to be calculated based on the time sampling interval, depth sampling interval, or channel spacing of the input profile. Let the time sampling interval be... Lane spacing is If processed in the time domain, then the length Corresponding agreement Time span, width Corresponding agreement The lateral span; if it has been converted to the depth domain, the depth sampling interval can be used. Alternative .

[0088] (4) In the formula, This indicates rounding to the nearest integer. This indicates taking the smallest odd number that is not less than the number inside the parentheses. For the time window or depth window that is desired to cover along the fault direction, The physical width is designed to allow for lateral swaying. Therefore, when the track spacing in work area A is 12.5m and the track spacing in work area B is 25m, the same "13-21 track" width cannot be mechanically applied; the number of tracks should be recalculated according to the target physical width, and an odd number not less than 5 should be used to ensure that the core center is clearly defined.

[0089] The values ​​such as -1.20 and -0.90 in the tendency set are not in radians or physical angles, but rather slope parameters in the two-dimensional pixel / sampling grid. According to the definition of line_kernel in this implementation, the slope s represents the offset of the center of the linear kernel in the channel column number when the row number changes by one time / depth sampling point, i.e., Δcol / Δrow.

[0090] (5) In the formula, Represents the column (track) coordinates of a pixel within the linear kernel. Represents the column coordinates of the center pixel of the kernel. Represents the row (time / depth) coordinates of a pixel within the kernel. Represents the row coordinates of the center pixel of the kernel. This is the equivalent tilt angle relative to the vertical time / depth axis; when the time profile is not converted to depth, it can be used. Or display coordinate scale alternative Perform pixel angle calibration. The direction set is not a fixed, empirically unchangeable parameter; it can be determined based on the maximum expected fault dip angle in the work area. Calculate the maximum slope with respect to the sampling interval, and then adjust the step size accordingly. Construct a set of directions.

[0091] (6) In this embodiment, the 91×21 linear core and the [-1.20, 1.20] dip set correspond to the gridded sampling and apparent dip angle range of the current 073 / 091 test profile. If the actual dip angle exceeds the preset set, the algorithm will not fail to the fixed mode, but will adaptively supplement the direction by expanding lineSlopes or reducing the direction step size.

[0092] It should be noted that the horizontal linear nucleus is a special case of the multi-directional linear nucleus set when the slope s=0, i.e., a horizontal linear nucleus extending along the time / depth direction with no channel offset. The length of the horizontal linear nucleus is related to the length of the multi-directional linear nucleus. Same (calculated according to formula (4)), width Three sampling points are fixed (i.e., three channels are sampled along the channel direction) to maintain sensitivity to near-horizontal laminar reflections while avoiding over-smoothing. The mathematical expression for the horizontal linear kernel is: ,in, , The row and column coordinates of the kernel center pixel are, in order. The linear nucleus length is calculated according to equation (4). The fault line seed probability is... With horizontal linear nucleus By performing convolution, the horizontal layered response map can be obtained. .

[0093] Step S4: Calculate the directional linear response. Convolve the seed probability with the directional linear kernel respectively, and take the maximum value of the response in each direction to obtain the directional linear support map.

[0094] Step S5, as follows Figure 5 The image shows an example of the horizontal response suppression and attribute background constraint process. Suppressing the horizontal layered response involves calculating the response along the layer using a horizontal linear kernel and subtracting a preset proportion of the horizontal response from the directional linear support diagram. This can be represented as follows: ,in It is a multidirectional linear nuclear response. It is a horizontal linear response. This is the horizontal response suppression coefficient. This can reduce the possibility of strong reflection layers and layered noise being mistakenly identified as faults.

[0095] In the current implementation, the horizontal response suppression coefficient This is a globally adjustable constant that does not dynamically change with time windows or spatial locations. The default value for the continuity enhancement function is a=0.88; in weaker neural network probabilistic post-processing functions, a=0.65 can be used. This coefficient is used to subtract the layer-by-layer response from the normalized directional response, and the specific calculation is as follows.

[0096] (7) When used in different work areas, small-step calibration can be performed within the range of 0.65-0.88: first observe the horizontal response R. hor The distribution of strong layered bands is then selected, and the minimum a value that can suppress most of the high values ​​along the layer while retaining the known low-angle fractures is chosen. If artificial fault labels already exist, fault line continuity, false detection rate along the layer, or F1-score can be used to determine the fault.

[0097] The horizontal linear kernel and the directional linear kernel share the same line length to ensure that they compete on the same layer length scale; however, the width of the horizontal kernel is not 13-21 channels, nor is it equal to the width of the directional linear kernel. The current implementation uses a horizontal kernel of 3×lineLength, that is, only 3 sampling point thicknesses are retained in the time / depth direction, which is used specifically for estimating the near-horizontal layered response.

[0098] Step S6, as follows Figure 5As shown, attribute background constraints are introduced: an attribute background map is constructed by multi-scale coherent discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attribute and structural deformation attribute. The attribute background map is formed by weighting each attribute. Then, the target linear support is multiplied by the linear lower bound gate term formed by the attribute background map, so that the connection result is mainly retained in the region with structural basis.

[0099] The preferred implementation uses a weighted summation with fixed prior weights to construct the attribute background map, and then applies multiplicative gating to the directional linear supports; it does not arbitrarily choose between product fusion and weighted summation, nor does it automatically determine the weights through PCA or machine learning. The currently implemented weight vector is as follows.

[0100] (8) in, For multi-scale coherent discontinuities, F represents Hessian-Frangi linear enhancement. For the slope gradient, To enhance the properties of fault lines, These are structural deformation properties. The weights mentioned above are derived from prior experience regarding fault boundaries, slender fractures, and structural deformation sensitivity, and were fixed and adjusted on the 073 / 091 section. They do not represent PCA principal component loads.

[0101] (9) The minimum retention of 0.35 for the attribute background is not a cumulative energy percentage, nor does it directly truncate regions where C is below 0.35; it is a linear lower bound gating term. Even if C=0, the region still retains 35% of the linear support, preventing the true faults in deep, weak attribute or low signal-to-noise ratio regions from being completely suppressed; as C increases, the retention factor smoothly increases from 0.35 to 1.

[0102] Step S7: Generate the probability of continuously enhancing faults. For example... Figure 6 As shown, this is an example of the continuous enhancement result output process. In some specific implementations, the original fault probability and the linear candidate probability are fused according to weights and then appropriately smoothed to output a continuously enhanced fault probability map; further, fault lines can be extracted according to a threshold to output a raster map or vector map.

[0103] The current preferred implementation uses a fixed-ratio fusion instead of dynamically changing weights based on local attribute background values. The local construction confidence has already been entered into equation (9). Therefore, the final fusion stage maintains a global linear candidate weight. This reduces parameter coupling.

[0104] (10) In the formula, To input the original fault probability, The linear candidate probability is the result of directional linear support, horizontal suppression, and attribute background constraints. The default value is 0.80, which means that the goal of this stage is to improve the continuity of fault candidate lines, and linear candidates that have passed the attribute background constraints are given priority. The remaining 0.20 retains the original probability to avoid over-connection that leads to deviation from the original fault response.

[0105] This default value comes from the parameter comparison and interpretation effect check on section 073 / 091: Even at very low levels, the probability of discontinuity remains significant. Excessively high values ​​may lead to overconnections in low-attribute background areas. If subsequent work areas have artificial tomographic labels, these can be assessed using linear continuity, layer-by-layer false detection rate, artificial tomographic matching degree, or F1-score. Recalibrate.

[0106] In some specific embodiments, it is assumed that the input is the original fault probability map pre-calculated from a two-dimensional seismic profile. When continuity is enhanced, the linear core length can be 25 to 91 sampling points, the linear core width can be 13 to 21 channels, and the linear core dip set can include multiple directions such as -1.20, -0.90, -0.60, -0.30, 0, 0.30, 0.60, 0.90, and 1.20.

[0107] The horizontal response suppression coefficient can be set between 0.65 and 0.88 to control the intensity of deduction for false anomalies along the bedding planes. The principle behind this range is that both the directional linear response and the horizontal response have been normalized to 0 to 1, and the coefficient 'a' corresponds to the proportion of horizontal layered components deducted from the directional response. When the seismic profile shows strong layered reflections, if 'a' is below approximately 0.65, the deduction of false anomalies along the bedding planes is insufficient, easily leading to the misidentification of continuous reflection layers as faults. If 'a' is above approximately 0.88, although the bedding plane response is suppressed more cleanly, low-dipping faults or near-horizontal small faults may be excessively weakened. Therefore, 0.65 to 0.88 is the preferred adjustment range, which can suppress most false bedding plane responses while retaining necessary low-dipping fault clues.

[0108] The attribute background constraint can be obtained by weighting multi-scale coherent discontinuities, Frangi line enhancement, dip gradient, fault line enhancement attributes, and tectonic deformation comprehensive attributes. To avoid the complete suppression of deep weak responses, a minimum retention amount of attribute background can be set, for example, 0.35.

[0109] Figures 7 to 9 The processing effects are shown on actual cross-sections (sections 073 and 091). Figure 7 This is an example comparing the continuity of section 073 before and after enhancement. Figure 8 This is a comparison example of the continuity enhancement before and after the 091 section. Figure 9Here is a three-part comparison example after local magnification of section 091 (the left side is the original probability candidate band before enhancement, the middle is the short linear nucleus intermediate response without final constraint and fusion, and the right side is the continuous probability candidate band after enhancement; all three examples are displayed in the same bright green, the only difference being the processing stage and the degree of continuity of the candidate band, used to illustrate that this method can improve the continuity of fault candidate lines compared to the original probability results); the original probability map before enhancement can show some fault responses, but there are fragmented points, discontinuities, and values ​​along the layer height; Figure 9 The linear kernel located in the middle supports a response that is an unconstrained intermediate processing quantity obtained from short linear kernels. It can show local short-segment responses, but has not yet undergone horizontal response suppression, attribute background constraint, and final fusion. After multi-directional linear kernel enhancement, horizontal response suppression, and attribute background constraint, Figure 9 The final continuous candidate band on the right side is more complete, making it easier for interpreters to track and map later.

[0110] In summary, the problem this invention aims to solve is to enhance discontinuous and fragmented fault probability responses into more continuous fault probability maps or fault line results by utilizing multi-directional linear cores, horizontal response suppression, and attribute background constraints, without altering the original source of fault probability. This provides more intuitive auxiliary data for the analysis of shallow gas-related fractures, plugging conditions, and migration channels.

[0111] Compared with the prior art, the present invention has at least the following beneficial effects: First, it can connect discontinuous high-probability points along the possible extension direction of the fault, making it more targeted than ordinary smoothing.

[0112] Second, by covering faults with different dip angles through multi-directional linear cores, it can adapt to various fault morphologies such as near-vertical, positive dip, and negative dip.

[0113] Third, horizontal response suppression reduces the risk of strong reflections along the layer and layered noise being mistakenly connected into faults.

[0114] Fourth, by using attribute background constraints, we avoid over-connecting in areas without structural basis, making the enhancement results more in line with seismic interpretation habits.

[0115] Fifth, the output results can be used as fault probability maps, fault lines, or editable vector results, which facilitate subsequent interpretation, mapping, and analysis of shallow gas-related faults, trap boundaries, and migration channels.

[0116] This invention also provides a fault probability continuity enhancement device based on a multi-directional linear core, which can implement the above-described method. This device may include: The first module is used to obtain the original fault probability map, normalize the original fault probability map and process outliers to obtain a standardized fault probability map. The second module is used to construct fault line seed probabilities based on standardized fault probability maps using adaptive thresholding or combination methods. The third module is used to convolve the fault line seed probability with each linear kernel in the preset multi-directional linear kernel set, and then take the maximum value of the convolution result pixel by pixel to obtain the directional linear support map. The fourth module is used to convolve the fault line seed probability with a horizontal linear kernel to obtain a horizontal layered response map. The proportional value of the horizontal layered response map is subtracted from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. The fifth module is used to construct an attribute background map based on preset seismic tectonic attributes, and to perform multiplication gating operation on the target linear support map and the attribute background map to obtain a linear candidate probability map. The sixth module is used to perform weighted fusion of the standardized fault probability map and the linear candidate probability map, and then obtain the continuously enhanced fault probability map through smoothing.

[0117] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0118] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0119] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0120] like Figure 10 As shown, Figure 10 The hardware structure of an electronic device 1000 according to another embodiment is illustrated. The electronic device 1000 includes: The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (aSIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention. The memory 1002 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RaM). The memory 1002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001. Input / output interface 1003 is used to implement information input and output; The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004); The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0121] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0122] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0123] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0124] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0125] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0126] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0127] The present invention provides a method, apparatus, electronic device, storage medium, and program product for enhancing fault probability continuity based on multi-directional linear kernels. The method involves: acquiring an original fault probability map; normalizing and processing outliers in the original fault probability map to obtain a standardized fault probability map; constructing fault line seed probabilities using an adaptive thresholding method or a combination method based on the standardized fault probability map; convolving the fault line seed probabilities with each linear kernel in a preset multi-directional linear kernel set, and then taking the maximum value pixel by pixel of the convolution result to obtain a directional linear support map; convolving the fault line seed probabilities with a horizontal linear kernel to obtain a horizontal layered response map; subtracting the proportion of the horizontal layered response map from the directional linear support map to obtain a target linear support map after suppressing the horizontal response; constructing an attribute background map based on preset seismic tectonic attributes; performing a multiplication-gated operation between the target linear support map and the attribute background map to obtain a linear candidate probability map; and weighted fusing the standardized fault probability map and the linear candidate probability map, followed by smoothing to obtain a continuously enhanced fault probability map. This invention utilizes multi-directional linear kernel convolution and pixel-wise maximum value extraction to directionally aggregate discontinuous high-probability points along different dips. Compared to ordinary smoothing, this method more effectively restores fault continuity and solves the problem that single-directional filtering cannot adapt to faults with different dip angles. By estimating layered reflections using a horizontal linear kernel and subtracting them from the directional response, it effectively suppresses the risk of strong reflections along layers being mistakenly linked as faults, reducing false anomaly interference. Multiplication gating is performed using the attribute background map, ensuring that the enhancement results are mainly retained in areas with structural basis, avoiding cross-regional misconnections without geological basis. By weightedly fusing the original fault probability and linear candidate probability and smoothing the output, it improves fault continuity while preserving the original probability response characteristics, providing more intuitive and reliable auxiliary results for subsequent interpretation.

[0128] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0129] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present invention, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0130] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0132] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the claims of the present invention.

Claims

1. A method for enhancing the probability continuity of faults based on multi-directional linear nuclei, characterized in that, The method includes the following steps: Obtain the original fault probability map, and perform normalization and outlier processing on the original fault probability map to obtain a standardized fault probability map. Based on the standardized fault probability map, the fault line seed probability is constructed using an adaptive threshold method or a combination method. The fault line seed probability is convolved with each linear kernel in the preset multi-directional linear kernel set, and then the maximum value of the convolution result is taken pixel by pixel to obtain the directional linear support map. A horizontal linear kernel is used to convolve the fault line seed probability to obtain a horizontal layered response map. The proportional value of the horizontal layered response map is subtracted from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. The horizontal linear kernel is a special case of the multi-directional linear kernel with a slope of 0. The length of the horizontal linear kernel is the same as the length of the multi-directional linear kernel, and the width of the horizontal linear kernel uses a fixed number of sampling points. An attribute background map is constructed based on preset seismic tectonic attributes. The target linear support map is multiplied and gated with the attribute background map to obtain a linear candidate probability map. The types of preset seismic tectonic attributes include multi-scale coherent discontinuity, Hessian-Frangi linear enhancement, dip gradient, linear enhancement attributes, and tectonic deformation. The standardized fault probability map and the linear candidate probability map are weighted and fused, and then smoothed to obtain a continuously enhanced fault probability map. The step of constructing the fault line seed probability based on the standardized fault probability map using an adaptive threshold method or a combination method includes the following steps: If an adaptive threshold method is used, the target quantile is obtained based on the standardized fault probability map, and the probability map mean and standard deviation are statistically obtained; wherein, the probability map mean represents the mean obtained based on the standardized fault probability map, and the probability map standard deviation represents the standard deviation obtained based on the standardized fault probability map. The candidate threshold is obtained by multiplying the standard deviation of the probability plot by a preset coefficient and then adding it to the mean of the probability plot. The larger value between the target quantile and the candidate threshold is used as the initial threshold, and the initial threshold is constrained by preset upper and lower limits to obtain the final threshold; The standardized fault probability map is binarized according to the final threshold to obtain a binary seed map, which is then transformed to obtain the fault line seed probability. If a combination method is used, the preset fault line is used as the main seed, and the standardized fault probability map is used as the auxiliary response. The fault line seed probability is obtained through linear weighted fusion and normalization.

2. The method according to claim 1, characterized in that, The process of normalizing and processing outliers in the original fault probability map to obtain a standardized fault probability map includes the following steps: The values ​​of the original fault probability map are normalized to a preset range, and outliers that exceed the preset range are truncated or replaced to obtain the standardized fault probability map.

3. The method according to claim 1, characterized in that, The method further includes the following steps: Obtain the sampling parameters and track spacing of the work area corresponding to the original fault probability map; wherein, the sampling parameters of the work area include the time sampling interval or the depth sampling interval; Obtain the sampling window along the fault direction, the physical width of the lateral swing, and the maximum expected fault dip angle in the work area; The ratio of the sampling window to the sampling parameters of the work area is rounded to obtain a first value; The first value is incremented to obtain the second value, and the smallest odd number not less than the second value is taken as the length of the linear core. The ratio of the lateral swing physical width to the track spacing is rounded to obtain a third value; The third value is incremented to obtain the fourth value, and the smallest odd number not less than the fourth value is taken as the width of the linear kernel. The fifth value is determined based on the ratio of the sampling parameters of the work area to the track spacing, and the sixth value is determined by performing a tangent calculation on the preset maximum expected fault dip angle of the work area. The maximum slope is obtained by multiplying the fifth and sixth values. Using the positive and negative values ​​of the maximum slope as boundaries, and then combining them with a preset slope step size, a set of directions is constructed to generate the multi-directional linear kernel set. Each slope of the direction set corresponds to a linear kernel, and the length and width of each linear kernel are determined based on the length and width of the linear kernel.

4. The method according to claim 1, characterized in that, The process of subtracting the scale value of the horizontal layered response map from the directional linear support map to obtain the target linear support map after suppressing the horizontal response includes the following steps: The seventh value is obtained by multiplying the horizontal layered response diagram with the preset horizontal response suppression coefficient; The eighth value is obtained based on the difference between the directional linear support diagram and the seventh value; The target linear support diagram after the suppression level response is determined based on the maximum value between the eighth value and the preset lower cutoff limit.

5. The method according to claim 1, characterized in that, The step of constructing an attribute background map based on preset seismic tectonic attributes, and then performing a multiplication-gated operation between the target linear support map and the attribute background map to obtain a linear candidate probability map includes the following steps: The attribute background map is constructed by weighting and summing various attribute features in the preset seismic tectonic attributes according to fixed prior weights and normalizing them. Based on the attribute background graph, a linear lower limit gating term is constructed using a preset minimum retention amount; The product of the target linear support graph and the linear lower bound gating term is normalized to obtain the linear candidate probability graph.

6. The method according to claim 1, characterized in that, The step of weightedly fusing the standardized fault probability map and the linear candidate probability map, and then smoothing the map to obtain a continuously enhanced fault probability map, includes the following steps: Obtain the preset global linear candidate weights, and subtract the global linear candidate weights from 1 to obtain the original probability weights. The ninth value is obtained by multiplying the original probability weights and the standardized fault probability map. The tenth value is obtained by multiplying the global linear candidate weights by the linear candidate probability map. The sum of the ninth and tenth values ​​is normalized to obtain the fusion result; wherein, the global linear candidate weight is used to control the proportion of the linear candidate probability map in the fusion result, and the global linear candidate weight is 0.8, indicating that the linear candidate probability map is given priority. Spatial smoothing is performed on the fusion result to obtain the continuously enhanced fault probability map.

7. A fault probability continuity enhancement device based on a multi-directional linear core, characterized in that, The apparatus for implementing the method of claim 1 includes: The first module is used to obtain the original fault probability map, and to normalize and process outliers in the original fault probability map to obtain a standardized fault probability map. The second module is used to construct the fault line seed probability based on the standardized fault probability map using an adaptive threshold method or a combination method. The third module is used to convolve the fault line seed probability with each linear kernel in the preset multi-directional linear kernel set, and then take the maximum value of the convolution result pixel by pixel to obtain the directional linear support map. The fourth module is used to convolve the fault line seed probability with a horizontal linear kernel to obtain a horizontal layered response map, and to subtract the proportion value of the horizontal layered response map from the directional linear support map to obtain the target linear support map after suppressing the horizontal response. The fifth module is used to construct an attribute background map based on preset seismic tectonic attributes, and to perform a multiplication gating operation on the target linear support map and the attribute background map to obtain a linear candidate probability map. The sixth module is used to perform weighted fusion of the standardized fault probability map and the linear candidate probability map, and then obtain a continuously enhanced fault probability map through smoothing.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images