Multi-well stratigraphic correlation method and device based on spatial perception heterogeneous graph and geology constraint optimization, equipment and storage medium

Abstract

The application discloses a multi-well stratum correlation method and device based on spatial perception heterogeneous graph and geological constraint optimization, equipment and storage medium, relates to the technical field of computers, and comprises the following steps: screening a logging curve sensitive to lithology, performing first and second derivative calculations to obtain a target to-be-processed logging curve, determining three-dimensional coordinates of sampling points in combination with inclinometer and well trajectory data, and constructing an implicit geometric correction three-dimensional space feature vector; extracting logging curve features and fusing the logging curve features with spatial features into node features; constructing a node with the sampling points as the center, and connecting the node with in-well longitudinal edges, inter-well spatial edges and geological prior edges to form a heterogeneous graph structure, so that a marker layer framework is constructed; layer correlation reasoning is performed in combination with a spatial distance penalty term; a stratum correlation profile is generated under a geological hard constraint loss function by using an intelligent model; and a layered data table including a layer group, a small layer name and top and bottom depths is decoded and output, so that the efficiency of isochronous stratum correlation of complex oil and gas reservoir characterization is improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, equipment, and storage medium for multi-well stratigraphic correlation based on spatially perceived heterogeneous maps and geological constraints optimization. Background Technology

[0002] Currently, in oil and gas exploration and development, multi-well stratigraphic correlation is a key technology for identifying the continuity and changes in stratigraphy between wells, providing core basis for sedimentary facies analysis, reservoir prediction, and 3D geological modeling. Conventional correlation mainly relies on well logging curves such as natural gamma, resistivity, density, and sonic transit time, and stratigraphic calibration is completed by manually observing the curve morphology, electrical response, and stratigraphic context (such as cyclic characteristics and marker beds). However, this method is highly dependent on the experience and subjective judgment of geological researchers, resulting in low efficiency. Furthermore, in blocks with dense well networks or complex stratigraphic structures (such as those with faults or facies transitions), it is difficult to guarantee the consistency and repeatability of correlation results.

[0003] To improve automation, algorithms based on curve similarity, such as cross-correlation analysis and window matching, are often used to assist in inter-well stratigraphic matching. These methods can reduce manual workload to some extent, but because they only utilize sequence morphological similarity, they are difficult to handle key geological constraints such as inter-well spatial distribution, lateral stratigraphic variation, and sequence continuity, and their reliability is insufficient in complex sedimentary systems.

[0004] In recent years, deep learning methods such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) have been used for automatic stratification of single wells, classifying layers by learning local or global features of well logging curves. However, these methods are usually designed for single-well processing and fail to express the spatial relationships, lateral collaborative information, and geological patterns of multiple wells in a unified model. Therefore, they cannot fundamentally solve the problems of mismatches, omissions, and inconsistencies in geological patterns in multi-well comparison scenarios.

[0005] Therefore, while existing multi-well stratigraphic correlation methods have evolved from manual experience to automated technologies such as Dynamic Time Warping (DTW) and convolutional neural networks, their core remains limited to measuring the similarity of curve morphology between wells or local pattern matching. These methods cannot effectively characterize the irregular topological structures of multiple wells in three-dimensional space, nor can they incorporate key geological principles such as "sequence non-intersection constraint" as hard constraints into the model's optimization process. Consequently, when facing complex structures, multi-well collaborative correlation, and thin interbedded formations, existing methods result in correlation results with inconsistencies across wells, mismatched and omitted stratigraphic layers, and systematic biases such as violations of basic geological principles (e.g., "cross-layering"), making it difficult to obtain stable and geologically reliable stratigraphic correspondences.

[0006] As can be seen from the above, how to correct the inconsistency between the logging curve representation and the actual spatial distribution of the formation under both deviated and horizontal well conditions during the multi-well formation correlation process based on spatially perceived heterogeneous maps and geological constraints, thereby improving the efficiency of isochronous formation correlation for characterizing complex oil and gas reservoirs, is an urgent problem to be solved. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for multi-well formation correlation based on spatially perceived heterogeneous maps and geological constraints. This method corrects the inconsistency between well logging curves and the actual spatial distribution of formations under both deviated and horizontal well conditions during multi-well formation correlation based on spatially perceived heterogeneous maps and geological constraints, thereby improving the correlation efficiency and geological accuracy of isochronous formations. The specific solution is as follows:

[0008] Firstly, this application provides a multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geologically constrained optimization, including:

[0009] From a number of logging curves, identify the logging curves that meet the preset lithological differentiation sensitivity conditions to be processed;

[0010] The first and second derivatives of the logging curve to be processed are calculated to obtain the target logging curve to be processed. Then, the three-dimensional coordinate information of each sampling point is determined based on the well data. Then, a three-dimensional spatial feature vector with implicit geometric correction is constructed using nonlinear mapping rules and based on the three-dimensional coordinate information and wellbore attitude information. The well data includes inclination data and well trajectory data.

[0011] Well logging curve features are extracted from the target well logging curve to be processed, and the well logging curve features are fused with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes;

[0012] Based on each sampling point and the corresponding node features, nodes are constructed, and vertical edges within the well for representing sedimentary sequences, spatial edges between wells for representing potential stratigraphic connectivity, and geological a priori edges for cross-well connections are constructed between each node to obtain a heterogeneous graph structure; each node is a local micro-stratigraphic unit centered on the sampling point, including curve morphology and contextual information.

[0013] A marker layer framework is constructed based on the heterogeneous graph structure, and layer comparison and reasoning are performed based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information;

[0014] A smart model is generated using stratigraphic correlation profiles. Based on the geological hard constraint loss function, which includes sequence monotonicity constraints and fault tolerance truncation constraints, and the inference results, a stratigraphic correlation profile is generated. The stratigraphic correlation profile is then decoded to obtain a stratigraphic data table of isochronous strata, including stratigraphic group names, sub-layer names, and top and bottom depths.

[0015] Optionally, determining the logging curve to be processed from a plurality of logging curves that meets the preset lithology differentiation sensitivity conditions includes:

[0016] Obtain well logging curves in the target work area, and determine the geological sedimentary pattern and lithological combination corresponding to the target work area. Then, based on the geological sedimentary pattern and the lithological combination, perform a preset lithological differentiation sensitivity judgment on the well logging curves to obtain a main comparison curve that meets the preset lithological differentiation sensitivity condition; the main comparison curve is a gamma curve.

[0017] Auxiliary curves are selected from the remaining logging curves. Then, based on the auxiliary curves and the main comparison curve, the curve to be processed is determined. The curve to be processed is then subjected to wellbore expansion detection, noise identification, and tool failure segment detection using a preset physical response mechanism to obtain the logging curve after detection. Then, the logging curve after detection is subjected to distortion interval removal processing to obtain the logging curve to be processed. The auxiliary curves include shallow and deep resistivity curves and sonic transit time curves. The auxiliary curves are used to describe the reservoirs, tight layers, and thin interbedded layers of multiple wells.

[0018] Optionally, the step of calculating the first and second derivatives of the logging curve to be processed to obtain the target logging curve to be processed, then determining the three-dimensional coordinate information of each sampling point based on each well data, and then constructing an implicit geometrically corrected three-dimensional spatial feature vector based on the three-dimensional coordinate information and wellbore attitude information using nonlinear mapping rules, includes:

[0019] The first derivative of the logging curve to be processed is calculated using a preset central difference algorithm to obtain the corresponding curve change rate feature, and the second derivative of the logging curve to be processed is calculated to obtain the corresponding inflection point feature. The target logging curve to be processed is determined based on the curve change rate feature, the inflection point feature and the logging curve to be processed.

[0020] The multi-well data, including well inclination angle, azimuth angle, and multi-well depth, corresponding to each of the multi-wells are determined. A preset coordinate calculation algorithm is used to determine the three-dimensional coordinate information of each sampling point based on the well inclination angle, the azimuth angle, and the multi-well depth. The multi-well depth includes vertical depth and tilt depth. The three-dimensional coordinate information includes geodetic coordinates and true vertical depth.

[0021] Based on the combination of the three-dimensional coordinate information and the wellbore attitude information, a wellbore trajectory geometric feature vector is constructed. The wellbore trajectory geometric feature vector is then input into a multilayer perceptron with learnable parameters. The multilayer perceptron is used to perform linear transformation and nonlinear activation operations on the wellbore trajectory geometric feature vector based on nonlinear mapping rules to obtain an implicitly geometrically corrected three-dimensional spatial feature vector. The three-dimensional spatial feature vector is used to characterize the deviation of the logging curve in representing the formation geometry.

[0022] Optionally, the step of extracting logging curve features from the target logging curve to be processed, and fusing the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial and physical attributes, includes:

[0023] A time-frequency transform is performed on the time-frequency domain corresponding to the target logging curve to be processed using a preset continuous wavelet transform algorithm to obtain a two-dimensional time-frequency domain feature curve to be extracted, including scale and depth dimensions. Then, a preset scale factor is used to extract features from the feature curve to be extracted to obtain logging curve features. The logging curve features include trend features for characterizing sedimentary cycle trends, detail features for characterizing lithological texture details, and contrast features for characterizing local layer boundary alignment.

[0024] The target logging curve is processed using Hilbert transform to obtain the instantaneous phase information of the target logging curve in the depth domain; the instantaneous phase information is used to assist in the location of formation boundaries.

[0025] The logging curve features, the instantaneous phase information, and the three-dimensional spatial feature vector are fused to obtain node features corresponding to each sampling point; the node features include the trend, texture, interface sensitivity, and wellbore trajectory of the target logging curve to be processed.

[0026] Optionally, the step of constructing nodes based on each sampling point and the corresponding node features, and constructing intra-well vertical edges for characterizing sedimentary sequences, inter-well spatial edges for characterizing potential stratigraphic connectivity, and geological a priori edges for cross-well connections between the nodes, to obtain a heterogeneous graph structure, includes:

[0027] The node features corresponding to each sampling point are determined, and nodes are constructed based on the node features and the corresponding sampling points. Then, the nodes in the same well are connected sequentially in order of node depth from shallow to deep, and the connection results are encoded to obtain the vertical edge in the well. The node attribute corresponding to the node is the node feature corresponding to the sampling point.

[0028] Determine the spatial location corresponding to each node, determine the three-dimensional Euclidean distance between the nodes of each multi-well, and connect the nodes whose distances in the three-dimensional Euclidean distances are less than a preset spatial neighborhood threshold to obtain the inter-well spatial edge used to characterize the potential formation connectivity relationship.

[0029] Based on the preset geological stratification conclusions and marker layer information, nodes belonging to the same geological stratum in each of the multiple wells are connected across wells to obtain geological a priori edges for cross-well connections. Then, each node, the vertical edge within the well, the spatial edge between wells, and the geological a priori edges are integrated to obtain a heterogeneous graph structure. The heterogeneous graph structure is used to characterize the vertical sequence relationship and the lateral spatial connectivity relationship of the multiple wells.

[0030] Optionally, the step of constructing a marker layer framework based on the heterogeneous graph structure, and performing layer comparison and inference based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the inference result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information, including:

[0031] Trend features are extracted from the heterogeneous map structure, and stratigraphic correlation reasoning is performed on the heterogeneous map structure based on the trend features to obtain key marker layers, so as to construct a marker layer framework for describing the structural framework of sandstone assemblages using the key marker layers.

[0032] Based on the marker layer framework and the three-dimensional coordinate information, the heterogeneous map structure is divided into several local sub-maps corresponding to different geological units. Detail features and contrast features are extracted within each local sub-map. Then, a spatial distance penalty term is determined based on the three-dimensional Euclidean distance between each node. Layer comparison and reasoning are performed based on the spatial distance penalty term, the detail features, and the contrast features to obtain the reasoning result.

[0033] Optionally, the intelligent model generated using the stratigraphic correlation profile and the stratigraphic correlation profile generated based on the geological hard constraint loss function including sequence monotonicity constraints and fault tolerance truncation constraints, and the inference results, and the stratigraphic correlation profile is decoded to obtain a stratigraphic data table of isochronous strata including stratigraphic group names, sub-layer names, and top and bottom depths, including:

[0034] In the well to be processed, adjacent node pairs with increasing depth are selected, and the adjacent node pairs are predicted using a preset depth value determination model to obtain depth value prediction results. Then, an activation function is used to penalize the prediction results that are out of order in the depth value prediction results to obtain the sequence monotonicity constraint loss.

[0035] The absolute value of the difference in formation depth between nodes between adjacent wells is determined, and the absolute value of the difference in formation depth is truncated based on a preset fault tolerance threshold to obtain the fault tolerance truncation constraint loss. Then, the formation correlation profile generation intelligent model is used to generate a formation correlation profile based on the inference results, the sequence monotonicity constraint loss and the fault tolerance truncation constraint loss to obtain the formation correlation profile.

[0036] The stratigraphic correlation profile is decoded to obtain decoding results that characterize the interface information of continuous stratigraphy. Based on a preset geological naming rule, the decoding results are converted into a structured stratigraphic layer data table. The data table includes the name of the layer group, the name of the sublayer, and the top and bottom boundary depths of each stratum.

[0037] Using a 3D visualization engine and based on the strata data table and the 3D coordinate information of each node, a well-to-well comparison profile and a 3D geological framework are generated in real space. The corresponding profile correction instructions are then determined using the 3D geological framework and based on the well-to-well comparison profile.

[0038] The well-connection comparison profile and the profile correction command are displayed on the interactive interface corresponding to the three-dimensional visualization engine. The profile correction command is converted into geological prior constraint samples, and the geological prior constraint samples are fed back to the stratigraphic comparison profile generation intelligent model for incremental learning to obtain a new stratigraphic comparison profile generation intelligent model.

[0039] Secondly, this application provides a multi-well stratigraphic correlation device based on spatially perceived heterogeneous maps and geological constraint optimization, including:

[0040] The logging curve determination module is used to determine the logging curve to be processed from several logging curves that meet the preset lithology differentiation sensitivity conditions.

[0041] The coordinate information determination module is used to calculate the first and second derivatives of the logging curve to be processed to obtain the target logging curve to be processed. Then, based on the well data, it determines the three-dimensional coordinate information of each sampling point. Then, it uses nonlinear mapping rules and constructs an implicit geometric correction three-dimensional spatial feature vector based on the three-dimensional coordinate information and wellbore attitude information. The well data includes inclination data and well trajectory data.

[0042] The node feature determination module is used to extract logging curve features from the target logging curve to be processed, and fuse the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes.

[0043] The heterogeneous graph structure generation module is used to construct nodes based on each sampling point and the corresponding node features, and to construct intra-well vertical edges for representing sedimentary sequences, inter-well spatial edges for representing potential stratigraphic connectivity, and geological a priori edges for cross-well connections between the nodes, thereby obtaining a heterogeneous graph structure; the nodes are local micro-stratigraphic units centered on the sampling points and including curve morphology and contextual information.

[0044] The reasoning result generation module is used to construct a marker layer framework based on the heterogeneous graph structure, and perform layer comparison and reasoning based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information;

[0045] The stratigraphic correlation profile generation module is used to generate an intelligent model using the stratigraphic correlation profile and generate a stratigraphic correlation profile based on the geological hard constraint loss function including sequence monotonicity constraint and fault tolerance truncation constraint and the inference result. The stratigraphic correlation profile is then decoded to obtain a stratigraphic data table of isochronous strata including stratigraphic group name, sub-layer name and top and bottom depth.

[0046] Thirdly, this application provides an electronic device, comprising:

[0047] Memory, used to store computer programs;

[0048] A processor is used to execute the computer program to implement the aforementioned multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraints optimization.

[0049] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned multi-well stratigraphic correlation method based on spatially aware heterogeneous maps and geological constraint optimization.

[0050] As can be seen from the above, before conducting multi-well formation correlation based on spatially perceived heterogeneous maps and geological constraints, this application needs to determine the logging curves to be processed from several logging curves that meet the preset lithological differentiation sensitivity conditions; calculate the first and second derivatives of the logging curves to be processed to obtain the target logging curves to be processed; then determine the three-dimensional coordinate information of each sampling point based on the data of each well; and then construct a three-dimensional spatial feature vector with implicit geometric correction based on the three-dimensional coordinate information and wellbore attitude information using nonlinear mapping rules; the well data includes inclination data and well trajectory data; extract logging curve features from the target logging curves to be processed; fuse the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial and physical attributes; construct nodes based on each sampling point and the corresponding node features; and construct the relationships between each node. Heterogeneous graph structures are obtained by using vertical edges within wells to characterize sedimentary sequences, spatial edges between wells to characterize potential stratigraphic connectivity, and geological a priori edges for cross-well connections. Nodes are local micro-stratigraphic units centered on sampling points, including curve morphology and contextual information. A marker layer framework is constructed based on the heterogeneous graph structure, and layer comparison and inference are performed based on local subgraphs and spatial distance penalty terms in the marker layer framework to obtain inference results. The spatial distance penalty term is a penalty term determined based on three-dimensional coordinate information. A smart model is generated using stratigraphic correlation profiles, and stratigraphic correlation profiles are generated based on geological hard constraint loss functions including sequence monotonicity constraints and fault tolerance truncation constraints, and inference results. The stratigraphic correlation profiles are decoded to obtain a layered data table of isochronous strata including layer group names, sublayer names, and top and bottom depths.

[0051] Therefore, this application first identifies the logging curves that meet the preset lithological differentiation sensitivity conditions from several logging curves; calculates the first and second derivatives of the logging curves to be processed to obtain the target logging curves to be processed; then, based on the data from each well, determines the three-dimensional coordinate information of each sampling point; and then, using nonlinear mapping rules and based on the three-dimensional coordinate information and wellbore attitude information, constructs a three-dimensional spatial feature vector with implicit geometric correction; secondly, extracts logging curve features from the target logging curves to be processed, and fuses the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial and physical attributes; then, based on each sampling point and the corresponding node features, constructs nodes, and constructs the vertical edges within the well to represent the sedimentary sequence and the potential strata between nodes. The heterogeneous map structure is obtained by defining the spatial edges between wells and the geological prior edges used for cross-well connections. Nodes are local micro-stratigraphic units centered on sampling points, including curve morphology and contextual information. Furthermore, a marker layer framework is constructed based on the heterogeneous map structure, and layer comparison and inference are performed based on local subgraphs and spatial distance penalty terms within the marker layer framework to obtain inference results. The spatial distance penalty term is determined based on three-dimensional coordinate information. Finally, an intelligent model is generated using the stratigraphic correlation profile, and a stratigraphic correlation profile is generated based on a geological hard constraint loss function including sequence monotonicity constraints and fault tolerance truncation constraints, along with the inference results. The stratigraphic correlation profile is then decoded to obtain a layered data table of isochronous stratigraphic layers, including layer group names, sublayer names, and top and bottom depths. In this way, in the process of multi-well stratigraphic correlation based on spatially perceived heterogeneous maps and geologically constrained optimization, the inconsistency between well logging curve representation and the actual spatial distribution of stratigraphy is corrected under both deviated and horizontal well conditions, thereby improving the efficiency of isochronous stratigraphic correlation for characterizing complex oil and gas reservoirs. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0053] Figure 1 This is a flowchart of a multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization disclosed in this application;

[0054] Figure 2 This is a schematic diagram illustrating the construction principle of a specific heterogeneous graph disclosed in this application;

[0055] Figure 3 This is a schematic diagram of a multi-well stratigraphic correlation device based on spatially perceived heterogeneous maps and geological constraint optimization disclosed in this application.

[0056] Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Currently, in oil and gas exploration and development, multi-well stratigraphic correlation is a key technology for identifying the continuity and changes in stratigraphy between wells, providing core basis for sedimentary facies analysis, reservoir prediction, and 3D geological modeling. Conventional correlation mainly relies on logging curves such as natural gamma, resistivity, density, and sonic transit time, and stratigraphic calibration is completed by manually observing the curve morphology, electrical response, and stratigraphic structural context. However, this method is highly dependent on the experience and subjective judgment of geological researchers, resulting in low efficiency, and in blocks with dense well networks or complex stratigraphic structures, it is difficult to guarantee the consistency and repeatability of correlation results. To address this, this application provides a multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization. This method can correct the inconsistency between logging curve representation and the actual spatial distribution of stratigraphy under both deviated and horizontal well conditions during multi-well stratigraphic correlation based on spatially perceived heterogeneous maps and geological constraint optimization, thereby improving the efficiency of isochronous stratigraphic correlation for characterizing complex oil and gas reservoirs.

[0059] See Figure 1 As shown, this embodiment of the invention discloses a multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization, including:

[0060] Step S11: Determine the logging curves to be processed from the logging curves of several wells that meet the preset lithology differentiation sensitivity conditions.

[0061] In this embodiment, stratigraphic correlation is the cornerstone of comprehensive geological research in oilfield development. Addressing the three major pain points of existing automatic correlation technologies in complex fault blocks, facies transitions, and deviated / horizontal well correlations—namely, lack of spatial constraints, lack of sequence consistency, and limited inference scale—this application proposes a novel solution integrating geological sedimentary patterns and cutting-edge map artificial intelligence. The technical solution in this embodiment comprises multiple technical modules, including data preprocessing and multi-logging curve feature construction, well network heterogeneous map modeling, spatially perceptual map inference, and geological hard constraint optimization.

[0062] In this embodiment, the entire process can be divided into seven core steps, as follows: First, sensitive curve selection and data quality control are performed: based on the reservoir sedimentary type and the physical response mechanism of logging curves, the input curves are selected for sensitivity. Natural gamma ray (GR) curves are selected as the main correlation curves reflecting lithological cycles because they are less affected by wellbore conditions and can stably reflect changes in clay content. To improve the identification capability of lithological interfaces, this embodiment can introduce deep and shallow resistivity (RT) or acoustic transit time (AC) as auxiliary features based on the lithological assemblage of the work area to enhance the characterization of reservoirs, tight layers, and thin interbedded layers. Subsequently, this embodiment requires quality screening of all logging curves, including wellbore expansion detection, noise identification, and tool failure segment detection, further eliminating distorted intervals to ensure that the data entering the model has reliable physical meaning. This reduces the propagation of erroneous features from the source, improves curve authenticity, and reduces noise interference, which is the foundation for ensuring the stability of subsequent feature construction.

[0063] Specifically, determining the logging curves to be processed from several logging curves that meet the preset lithological differentiation sensitivity conditions can include: acquiring logging curves corresponding to multiple wells in the target area, determining the geological sedimentary model and lithological combination corresponding to the target area, and then judging the logging curves based on the preset lithological differentiation sensitivity conditions based on the geological sedimentary model and lithological combination to obtain the main comparison curve that meets the preset lithological differentiation sensitivity conditions; the main comparison curve is a gamma curve; selecting auxiliary curves from the remaining logging curves, and then determining the curves to be processed based on the auxiliary curves and the main comparison curve, and using the preset physical response mechanism to perform wellbore expansion detection, noise identification, and tool failure segment detection on the curves to be processed to obtain the logging curves after detection, and then performing distortion interval removal processing on the logging curves after detection to obtain the logging curves to be processed; the auxiliary curves include shallow and deep resistivity curves and sonic transit time curves; the auxiliary curves are used to describe the reservoirs, tight layers, and thin interbedded layers of multiple wells.

[0064] Secondly, the embodiments of this application require data preprocessing, that is, standardization of the optimized logging data. On the one hand, addressing the issue of inconsistent scales across different logging instruments, the embodiments of this application require Z-score standardization of linear scale curves (such as GR):

[0065] ;

[0066] in, These are the original well logging measurements. The target logging curve to be processed is the average value across the entire work area. The standard deviation is denoted as .

[0067] Step S12: Calculate the first and second derivatives of the logging curve to be processed to obtain the target logging curve to be processed. Then, determine the three-dimensional coordinate information of each sampling point based on the multi-well data of each of the multi-wells. Then, construct a three-dimensional spatial feature vector with implicit geometric correction based on the nonlinear mapping rule and the three-dimensional coordinate information and wellbore attitude information. The multi-well data includes inclination data and well trajectory data.

[0068] In this embodiment, the three-dimensional spatial features need to be encoded, that is, the logging curve to be processed is spatially encoded. Furthermore, to highlight the abrupt changes in formation interfaces, this embodiment uses the central difference method to calculate the depth-domain first derivative (rate of change) of the curve. and the second derivative (inflection point) And the corresponding expression is as follows:

[0069] ;

[0070] ;

[0071] in, For the first Curve values ​​at each depth point The sampling interval is denoted as .

[0072] On the other hand, in order to solve the problem of "apparent thickness" distortion when comparing deviated and horizontal wells, this application introduces three-dimensional spatial encoding technology to utilize the inclination data. and azimuth The absolute three-dimensional coordinates (X, Y, Z) of each depth sampling point are calculated based on inclination calculation methods (such as the minimum curvature method). Then, the true vertical depth (TVD) is compared with the geometric feature vector of the wellbore attitude. It is then mapped to a high-dimensional location vector using a multilayer perceptron (MLP). :

[0073] ;

[0074] in, , The weight matrix is ​​a learnable matrix. , is the bias term, and ReLU is the non-linear activation function.

[0075] In this embodiment, the model is able to perceive whether the wellbore is drilling vertically through the formation or along the formation, thereby automatically aligning geometric features during comparison. This allows the model to understand the true geometric relationships between wells, significantly reducing false formation correspondences caused by deviated or horizontal wells. It is worth noting that spatial geometric information can be replaced by well spacing, relative azimuth, depth difference, well trajectory encoder, or spatial statistical models.

[0076] Specifically, the first and second derivatives of the logging curve to be processed are calculated to obtain the target logging curve to be processed. Then, based on the data from each well, the three-dimensional coordinate information of each sampling point is determined. Next, a three-dimensional spatial feature vector with implicit geometric correction is constructed using nonlinear mapping rules and based on the three-dimensional coordinate information and wellbore attitude information. This may include: using a preset central difference algorithm to calculate the first derivative of the logging curve to be processed to obtain the corresponding curve rate of change feature, and calculating the second derivative of the logging curve to be processed to obtain the corresponding inflection point feature. The target logging curve to be processed is determined based on the curve rate of change feature, inflection point feature, and the logging curve to be processed; and determining the well inclination angle, azimuth angle, and other parameters corresponding to each well. The multi-well data at various depths is used to determine the three-dimensional coordinate information of each sampling point based on the well inclination angle, azimuth angle, and multi-well depth using a preset coordinate calculation algorithm. The multi-well depth includes vertical depth and dip depth. The three-dimensional coordinate information includes geodetic coordinates and true vertical depth. Based on the combination of three-dimensional coordinate information and wellbore attitude information, a wellbore trajectory geometric feature vector is constructed. This wellbore trajectory geometric feature vector is then input into a multilayer perceptron with learnable parameters. The multilayer perceptron is used to perform linear transformation and nonlinear activation operations on the wellbore trajectory geometric feature vector based on nonlinear mapping rules to obtain an implicitly geometrically corrected three-dimensional spatial feature vector. The three-dimensional spatial feature vector is used to characterize the deviation of the logging curve from the formation geometry.

[0077] Step S12: Extract logging curve features from the target logging curve to be processed, and fuse the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes.

[0078] In this embodiment, the present application requires the generation of multi-scale time-frequency features and the enhancement of these features. Furthermore, to address the ambiguity arising from relying solely on the original curve shape comparison, this embodiment utilizes signal processing techniques to generate multi-scale (depth-scale domain) features and performs a continuous wavelet transform (CWT) on the preprocessed curve. The one-dimensional depth domain signal is then upscaled to a two-dimensional time-frequency domain using the Morlet wavelet basis. The wavelet coefficients... The calculation expression is as follows:

[0079] ;

[0080] in, Represents the depth variable. The scale factor (large scale corresponds to low-frequency sedimentary cycles, and small scale corresponds to high-frequency lithological abrupt changes). The translation factor is... It is the complex conjugate of the Morlet mother wavelet.

[0081] In this embodiment, the low-frequency coefficients of the target logging curve to be processed are extracted to describe the sedimentary cycle trend, and the high-frequency coefficients are extracted to capture lithological abrupt changes and fine bedding. Simultaneously, the instantaneous phase spectrum of the signal is extracted using the Hilbert Transform. The corresponding expression is as follows:

[0082] ;

[0083] in, Representing the Hilbert transform operator, using phase Occurs at the interface arrive The jump characteristics are used to accurately locate formation boundaries. Furthermore, the embodiments of this application construct a multi-logging curve feature system that simultaneously possesses trend, texture, and interface sensitivity by combining the original curve, derivative features, wavelet components, and instantaneous phase, thereby reducing misjudgments caused by similar curve shapes and improving the accuracy of layer boundary location.

[0084] Furthermore, in the feature construction stage, continuous wavelet transform (CWT) and Hilbert transform can be replaced by time-frequency analysis methods such as empirical mode decomposition (EMD), S-transform, and synchronous compressed time-frequency transform (SST). Alternatively, they can be replaced by one-dimensional convolutional neural networks or recurrent neural networks (RNNs) that can automatically extract features from multiple well logging curves, thereby achieving equivalent feature enhancement.

[0085] Specifically, well logging curve features are extracted from the target well logging curve to be processed. These features are then fused with a three-dimensional spatial feature vector to obtain node features that include both spatial and physical attributes. This process can include: performing a time-frequency transformation on the time-frequency domain corresponding to the target well logging curve using a preset continuous wavelet transform algorithm to obtain a two-dimensional time-frequency domain feature curve to be extracted, including scale and depth dimensions; then using a preset scale factor to extract features from the feature curve to obtain well logging curve features; these features include trend features for characterizing sedimentary cycles, detail features for characterizing lithological texture details, and contrast features for characterizing local layer boundary alignment; processing the target well logging curve using Hilbert transform to obtain instantaneous phase information of the target well logging curve in the depth domain; this instantaneous phase information is used to assist in locating formation boundaries; and fusing the well logging curve features, instantaneous phase information, and three-dimensional spatial feature vector to obtain node features corresponding to each sampling point; these node features include the trend, texture, interface sensitivity, and wellbore trajectory of the target well logging curve.

[0086] Step S14: Construct nodes based on each sampling point and the corresponding node features, and construct vertical edges within the well to represent sedimentary sequences, spatial edges between wells to represent potential stratigraphic connectivity, and geological a priori edges for cross-well connections between the nodes to obtain a heterogeneous graph structure; the node is a local micro-stratigraphic unit centered on the sampling point, including curve morphology and contextual information.

[0087] In this embodiment, the present application requires the construction of a multi-well heterogeneous graph and topology modeling, and the corresponding schematic diagram of the heterogeneous graph construction principle is shown below. Figure 2As shown, this embodiment abstracts the entire well network of the work area into a heterogeneous graph structure. The system discretizes each well into several nodes based on the data scale and employs flexible node partitioning strategies (such as aggregation by sampling point or sliding window). It is worth noting that the sampling points in this embodiment are several discrete data points selected based on the well depth of multiple wells, and each sampling point corresponds only to the logging curve value at a specific depth. Nodes are the basic units for modeling and computation using graph neural networks, belonging to the algorithm model level. To enable the model to efficiently handle formation correlation tasks and fully represent curve morphology information, this embodiment needs to fuse the curve morphology and contextual information of the sampling point within its neighborhood using methods such as sliding windowing and derivatives, and to fuse the curve morphology, contextual information, and the original value of a single sampling point. That is, the nodes in this embodiment are single sampling points physically anchored to the logging curve, but each node represents a local micro-stratum unit centered on the sampling point. Furthermore, the embodiments of this application can extract first derivative, second derivative, sliding window statistical features and trend features, so that the feature vector of the node can encode the curve shape and context information of the sampling point within the neighborhood window (e.g., within a predetermined range above and below the sampling point), thereby enabling each node to have the ability to perceive changes in local geological structure.

[0088] Furthermore, to express geological relationships, this application's embodiments design three types of edges to describe geological relationships: constructing "intra-well longitudinal edges" to connect nodes at adjacent depths within the same well to maintain the continuity of the sedimentary sequence; constructing "inter-well spatial edges" to connect nodes between different wells whose physical distance is within a certain threshold, serving as potential comparison paths; and constructing "geological prior edges" to directly connect artificially confirmed marker layers, serving as strong constraints to guide model convergence.

[0089] It is worth mentioning that the above-mentioned heterogeneous graph structure can simultaneously encode the vertical sequence information of a single well and the lateral topological relationship of multiple wells, enabling the model to comprehensively utilize cross-well and cross-scale integrated information to solve the problem of missing lateral constraints in multi-well comparisons.

[0090] Furthermore, in the graph model construction and inference stages, heterogeneous graph structures can be replaced by various graph neural network structures such as cross-well Transformer, Graph Convolutional Network (GCN), GraphSAGE (Graph Sample and AggregateE, a graph neural network algorithm framework), and Heterogeneous Graph Transformer (HGT), or message passing neural networks (MPNN). The hierarchical inference strategy, moving from coarse to fine, can also be implemented using strategies such as dynamic programming global alignment. Furthermore, the hierarchical inference in the embodiments of this application can also be implemented using a single-stage end-to-end model, dynamic programming alignment, or reinforcement learning strategies. Moreover, geological hard constraints can be equivalently implemented using energy functions, conditional random fields, soft constraints, Bayesian prior fields, and other methods.

[0091] Specifically, nodes are constructed based on each sampling point and its corresponding node features. Furthermore, vertical edges within the well are constructed to represent sedimentary sequences, spatial edges between wells are constructed to represent potential formation connectivity, and geological prior edges are constructed for cross-well connections, resulting in a heterogeneous graph structure. This process may include: determining the node features corresponding to each sampling point; constructing nodes based on these features and the corresponding sampling points; then sequentially connecting nodes within the same well in ascending order of depth; and encoding the connection results to obtain vertical edges within the well. The node attributes corresponding to each node are the node features corresponding to the sampling points. The spatial location corresponding to each node is determined, and the three-dimensional Euclidean distance between nodes in each multi-well is determined. Nodes with a distance less than a preset spatial neighborhood threshold in each three-dimensional Euclidean distance are connected to obtain the inter-well spatial edge used to characterize the potential stratigraphic connectivity relationship. Based on the preset geological stratification conclusion and marker layer information, nodes belonging to the same geological stratum in each multi-well are connected across wells to obtain the geological a priori edge used for cross-well connection. Then, each node, the vertical edge within the well, the inter-well spatial edge, and the geological a priori edge are integrated to obtain the heterogeneous graph structure. The heterogeneous graph structure is used to characterize the vertical sequence relationship and the lateral spatial connectivity relationship of the multi-well.

[0092] Step S15: Construct a marker layer framework based on the heterogeneous graph structure, and perform layer comparison and reasoning based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information.

[0093] In this embodiment, the present application can perform graph reasoning based on spatial awareness and a hierarchical strategy. That is, the present application adopts a "coarse-to-fine" hierarchical reasoning strategy to perform the comparison task. First, at the macroscopic stage, the model can use low-frequency trend features to perform coarse-grained reasoning across the entire work area to identify key marker beds with region-wide consistency and establish a first-level structural framework at the sandstone group level. Here, the low-frequency trend features are the trend components in the node feature vectors. Subsequently, at the microscopic stage, the present application divides the large map into multiple local sub-maps using the first-level framework as boundaries, and uses high-frequency detail features to perform fine-grained alignment at the small-layer level within the sub-maps. Here, the high-frequency detail features are the detail components (such as lithological abrupt change features) in the node feature vectors, used for fine-grained alignment of small layers or sedimentary units.

[0094] It is worth mentioning that, during the inference process, this application's embodiment introduces a "spatial distance gating attention mechanism" at the computation node. with neighboring nodes Similarity (attention coefficient) When this occurs, add a spatial distance penalty:

[0095] ;

[0096] in, , For node feature vectors, , For attention projection matrix, For feature dimension, The three-dimensional Euclidean distance between nodes. This is a learnable or preset spatial penalty coefficient. In this way, the above mechanism physically blocks unreasonable spurious comparisons due to excessive distance, guiding the model to prioritize finding stratigraphic correspondences between spatially adjacent well points, which conforms to the geological "principle of local continuity". This significantly improves the robustness and accuracy of comparisons under complex well networks, and enhances the robustness of comparisons, thereby ensuring that the results have spatial continuity and geological rationality.

[0097] Specifically, a marker layer framework is constructed based on the heterogeneous map structure, and layer comparison and inference are performed based on the local submaps in the marker layer framework and the spatial distance penalty term to obtain the inference result. The spatial distance penalty term is a penalty term determined based on three-dimensional coordinate information, which may include: extracting trend features from the heterogeneous map structure, and performing stratigraphic comparison and inference on the heterogeneous map structure based on the trend features to obtain key marker layers, so as to construct a marker layer framework for describing the structural framework of sandstone assemblages using the key marker layers; dividing the heterogeneous map structure into several local submaps corresponding to different geological units based on the marker layer framework and three-dimensional coordinate information, extracting detailed features and contrast features inside each local submap, and then determining the spatial distance penalty term based on the three-dimensional Euclidean distance between each node, so as to perform layer comparison and inference based on the spatial distance penalty term, detailed features and contrast features to obtain the inference result.

[0098] Step S16: Generate an intelligent model using the stratigraphic correlation profile and generate a stratigraphic correlation profile based on the geological hard constraint loss function and the inference result. Decode the stratigraphic correlation profile to obtain a stratigraphic data table of isochronous strata including layer group name, sublayer name and top and bottom depth.

[0099] In this embodiment, the model used to drive geological hard constraints needs to be optimized. To prevent "cross-layer" or "sequence inversion" errors that may occur with a purely data-driven model, this embodiment introduces two types of geological hard constraint loss functions during model training to construct a joint loss function. .

[0100] Furthermore, embodiments of this application define a "sequence non-intersection constraint". This applies to two adjacent nodes with increasing depth in the reference well. , (Right now ), requiring its matching depth in the target well, , The relationship must be monotonically non-decreasing. Once the model predicts a deep inversion ( (i.e., when layer penetration occurs), the ReLU term in the loss function is activated:

[0101] ;

[0102] In this embodiment, a positive penalty gradient is generated, and then the model parameters are forcibly corrected through backpropagation to ensure that the output comparison results have no crossover in the topology.

[0103] Furthermore, this application's embodiments introduce a "Robust Smoothness Constraint" with fault tolerance capability; that is, a truncated loss function is used to address the difference in formation depth between adjacent wells. Apply constraints:

[0104] ;

[0105] in, This is the preset fault tolerance threshold.

[0106] It is worth mentioning that the above thresholds represent the maximum allowable depth variation between adjacent wells while maintaining formation continuity. The specific values ​​can be estimated based on the average well spacing and maximum formation dip angle of the work area, or determined based on the statistical distribution of depth differences in known formation comparison samples.

[0107] In one specific implementation, when In non-fault zones, losses increase quadratically, with strongly constrained strata exhibiting gentle, gradual changes; when In the fault zone, the loss is truncated to a constant. The coefficient no longer increases with increasing depth difference, which prevents the model from forcibly smoothing drastic depth changes, thus preserving geological abrupt changes while smoothing the structure. Therefore, this embodiment of the application, by jointly optimizing the basic matching loss and these hard constraint terms, ultimately outputs a fine stratigraphic correlation profile that conforms to both data characteristics and strictly follows sedimentary geological laws, thereby ensuring that the output does not violate basic geological laws.

[0108] Finally, this embodiment of the application requires the output and interactive visualization of stratigraphic correlation results. That is, this embodiment decodes the results inferred by the model to generate a geologically significant layered data table (including layer groups, sublayer names, and top and bottom depths), and supports exporting it to a standard format. Simultaneously, a 3D visualization engine projects the correlation results onto real space, rendering well-connected profiles and a 3D stratigraphic framework in real time to intuitively display thickness variations and structural features. In this way, this embodiment supports geologically guided interactive fine-tuning. Subsequently, this embodiment can modify the visualization interface to generate new constraint samples to feed back to the system, achieving closed-loop incremental learning, continuously improving the model's practicality, and thus realizing a closed-loop fusion of machine results and expert knowledge.

[0109] Specifically, an intelligent model is generated using stratigraphic correlation profiles. Based on a geological hard constraint loss function including sequence monotonicity constraints and fault tolerance truncation constraints, a stratigraphic correlation profile is generated using the inference results. The stratigraphic correlation profile is then decoded to obtain a layered data table of isochronous strata, including stratigraphic group names, sub-layer names, and top and bottom depths. This process may include: selecting adjacent node pairs with increasing depths from the multiple wells to be processed, and using a preset depth value determination model to predict the depth values ​​of adjacent node pairs. Then, an activation function is used to penalize predictions with reverse order in the depth value prediction results to obtain the sequence monotonicity constraint loss; determining the absolute value of the stratigraphic burial depth difference between nodes of adjacent wells, and truncating the absolute value of the stratigraphic burial depth difference based on a preset fault tolerance threshold to obtain the fault tolerance truncation constraint loss; and then using the intelligent model generated from the stratigraphic correlation profiles to process the inference results, sequence monotonicity constraint loss, and fault tolerance truncation constraint loss. The process involves generating stratigraphic correlation profiles, decoding these profiles to obtain decoding results that characterize the interfaces between continuous stratigraphic layers, and converting these results into a structured multi-well stratigraphic layered data table based on preset geological naming rules. The data table includes layer group names, sub-layer names, and the corresponding top and bottom boundary depths for each stratum. A 3D visualization engine is used to generate well-connected correlation profiles and a 3D geological framework at their actual spatial locations, based on the multi-well stratigraphic layered data table and the 3D coordinates of each node. The 3D geological framework is then used to determine corresponding profile correction instructions based on the well-connected correlation profiles. The well-connected correlation profiles and correction instructions are displayed on the interactive interface of the 3D visualization engine, and the correction instructions are converted into geological prior constraint samples. These geological prior constraint samples are then fed back to the intelligent stratigraphic correlation profile generation model for incremental learning, resulting in a new intelligent stratigraphic correlation profile generation model.

[0110] As can be seen from the above, the embodiments of this application first determine the logging curves to be processed from several logging curves that meet the preset lithological differentiation sensitivity conditions; calculate the first and second derivatives of the logging curves to be processed to obtain the target logging curves to be processed; then determine the three-dimensional coordinate information of each sampling point based on the data of each well; and then construct a three-dimensional spatial feature vector with implicit geometric correction based on the three-dimensional coordinate information and the wellbore attitude information using nonlinear mapping rules; secondly, extract logging curve features from the target logging curves to be processed, fuse the logging curve features and the three-dimensional spatial feature vector to obtain node features including spatial and physical attributes; then, construct nodes based on each sampling point and the corresponding node features, and construct the vertical edges between each node to characterize the sedimentary sequence and the edges to characterize the potential... The heterogeneous map structure is obtained by defining the spatial edges between wells and the geological prior edges used for cross-well connections in the stratigraphic connectivity relationship. Nodes are local micro-stratigraphic units centered on sampling points, including curve morphology and contextual information. Furthermore, a marker layer framework is constructed based on the heterogeneous map structure, and layer comparison and inference are performed based on the local subgraphs and spatial distance penalty terms within the marker layer framework to obtain the inference results. The spatial distance penalty term is determined based on three-dimensional coordinate information. Finally, an intelligent model is generated using the stratigraphic correlation profile, and a stratigraphic correlation profile is generated based on the geological hard constraint loss function, including sequence monotonicity constraints and fault tolerance truncation constraints, and the inference results. The stratigraphic correlation profile is then decoded to obtain a layered data table of isochronous stratigraphic layers, including layer group names, sublayer names, and top and bottom depths. In this way, in the process of multi-well stratigraphic correlation based on spatially perceived heterogeneous maps and geologically constrained optimization, the inconsistency between well logging curve representation and the actual spatial distribution of stratigraphy is corrected under both deviated and horizontal well conditions, thereby improving the efficiency of isochronous stratigraphic correlation in characterizing complex oil and gas reservoirs.

[0111] Accordingly, see Figure 3 As shown, this application also provides a multi-well stratigraphic correlation device based on spatially perceived heterogeneous maps and geological constraints optimization, comprising:

[0112] The logging curve determination module 11 is used to determine the logging curve to be processed from a number of logging curves that meet the preset lithology differentiation sensitive conditions.

[0113] The coordinate information determination module 12 is used to calculate the first and second derivatives of the logging curve to be processed to obtain the target logging curve to be processed. Then, based on the well data, it determines the three-dimensional coordinate information of each sampling point. Then, it uses nonlinear mapping rules and constructs an implicit geometric correction three-dimensional spatial feature vector based on the three-dimensional coordinate information and wellbore attitude information. The well data includes inclination data and well trajectory data.

[0114] The node feature determination module 13 is used to extract logging curve features from the target logging curve to be processed, and fuse the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes.

[0115] The heterogeneous graph structure generation module 14 is used to construct nodes based on each sampling point and the corresponding node features, and to construct intra-well vertical edges for representing sedimentary sequences, inter-well spatial edges for representing potential stratigraphic connectivity, and geological a priori edges for cross-well connections between each node, thereby obtaining a heterogeneous graph structure; the node is a local micro-stratigraphic unit centered on the sampling point, including curve morphology and contextual information.

[0116] The reasoning result generation module 15 is used to construct a marker layer framework based on the heterogeneous graph structure, and perform layer comparison and reasoning based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information;

[0117] The stratigraphic correlation profile generation module 16 is used to generate an intelligent model using the stratigraphic correlation profile and generate a stratigraphic correlation profile based on the geological hard constraint loss function including sequence monotonicity constraint and fault tolerance truncation constraint and the inference result. The stratigraphic correlation profile is then decoded to obtain a stratigraphic data table of isochronous strata including stratigraphic group name, sub-layer name and top and bottom depth.

[0118] In some specific embodiments, the well logging curve determination module 11 may specifically include:

[0119] The well logging curve acquisition unit is used to acquire well logging curves corresponding to multiple wells in the target work area, determine the geological sedimentary pattern and lithological combination corresponding to the target work area, and then perform a preset lithological differentiation sensitivity condition judgment on the well logging curves based on the geological sedimentary pattern and the lithological combination to obtain a main comparison curve that meets the preset lithological differentiation sensitivity condition; the main comparison curve is a gamma curve.

[0120] An auxiliary curve determination unit is used to select an auxiliary curve from the remaining logging curves, then determine the curve to be processed based on the auxiliary curve and the main comparison curve, and use a preset physical response mechanism to perform wellbore expansion detection, noise identification, and tool fault segment detection on the curve to be processed to obtain a logging curve after detection. Then, the logging curve after detection is processed to remove distortion intervals to obtain the logging curve to be processed. The auxiliary curve includes shallow and deep resistivity curves and sonic transit time curves. The auxiliary curve is used to describe the reservoir, tight layer, and thin interbedded layers of multiple wells.

[0121] In some specific embodiments, the coordinate information determination module 12 may specifically include:

[0122] The curve change rate feature determination unit is used to calculate the first derivative of the logging curve to be processed using a preset central difference algorithm to obtain the corresponding curve change rate feature, and to calculate the second derivative of the logging curve to be processed to obtain the corresponding inflection point feature, so as to determine the target logging curve to be processed based on the curve change rate feature, the inflection point feature and the logging curve to be processed.

[0123] A multi-well data determination unit is used to determine multi-well data corresponding to each of the multi-wells, including well inclination angle, azimuth angle, and multi-well depth, so as to determine the three-dimensional coordinate information of each sampling point using a preset coordinate calculation algorithm based on the well inclination angle, the azimuth angle, and the multi-well depth; the multi-well depth includes vertical depth and dip depth; the three-dimensional coordinate information includes geodetic coordinates and true vertical depth;

[0124] The spatial feature vector generation unit is used to construct a wellbore trajectory geometric feature vector based on the combination of the three-dimensional coordinate information and the wellbore attitude information, and input the wellbore trajectory geometric feature vector into a multilayer perceptron with learnable parameters. The multilayer perceptron is then used to perform linear transformation and nonlinear activation operations on the wellbore trajectory geometric feature vector based on nonlinear mapping rules to obtain an implicitly geometrically corrected three-dimensional spatial feature vector. The three-dimensional spatial feature vector is used to characterize the deviation of the logging curve in representing the formation geometry.

[0125] In some specific embodiments, the node feature determination module 13 may specifically include:

[0126] The feature curve determination unit is used to perform time-frequency transformation on the time-frequency domain corresponding to the target logging curve to be processed using a preset continuous wavelet transform algorithm to obtain a two-dimensional time-frequency domain feature curve to be extracted, including scale and depth dimensions. Then, the feature curve to be extracted is extracted using a preset scale factor to obtain logging curve features. The logging curve features include trend features for characterizing sedimentary cycle trends, detailed features for characterizing lithological texture details, and contrast features for characterizing local layer boundary alignment.

[0127] The instantaneous phase information generation unit is used to process the target logging curve to be processed using Hilbert transform to obtain the instantaneous phase information of the target logging curve in the depth domain; the instantaneous phase information is used to assist in the location of formation boundaries;

[0128] The feature vector fusion unit is used to fuse the logging curve features, the instantaneous phase information and the three-dimensional spatial feature vector to obtain the node features corresponding to each sampling point; the node features include the trend, texture, interface sensitivity and wellbore trajectory of the target logging curve to be processed.

[0129] In some specific embodiments, the heterogeneous graph structure generation module 14 may specifically include:

[0130] A node construction unit is used to determine the node features corresponding to each sampling point, construct nodes based on the node features and the corresponding sampling points, and then connect the nodes in the same multi-well according to the node depth from shallow to deep, and encode the connection results to obtain the vertical edge in the well; the node attribute corresponding to the node is the node feature corresponding to the sampling point;

[0131] The Euclidean distance determination unit is used to determine the spatial position corresponding to each of the nodes, and to determine the three-dimensional Euclidean distance between the nodes of each of the multi-wells, and to connect the nodes whose distances in the three-dimensional Euclidean distances are less than a preset spatial neighborhood threshold, so as to obtain the inter-well spatial edge used to characterize the potential formation connectivity relationship.

[0132] A geological prior edge generation unit is used to connect nodes belonging to the same geological stratum in each of the multiple wells across wells based on preset geological stratification conclusions and marker layer information, thereby obtaining geological prior edges for cross-well connections. Then, the nodes, the vertical edges within the wells, the spatial edges between wells, and the geological prior edges are integrated to obtain a heterogeneous graph structure. The heterogeneous graph structure is used to characterize the vertical sequence relationship and the lateral spatial connectivity relationship of the multiple wells.

[0133] In some specific embodiments, the reasoning result generation module 15 may specifically include:

[0134] A marker layer framework construction unit is used to extract trend features from the heterogeneous map structure and perform stratigraphic correlation reasoning on the heterogeneous map structure based on the trend features to obtain key marker layers, so as to construct a marker layer framework for describing the structural framework of sandstone groups using the key marker layers.

[0135] The reasoning result generation subunit is used to divide the heterogeneous map structure into several local sub-maps corresponding to different geological units based on the marker layer framework and the three-dimensional coordinate information, and to extract detailed features and contrast features within each local sub-map. Then, a spatial distance penalty term is determined based on the three-dimensional Euclidean distance between each node, and layer comparison and reasoning are performed based on the spatial distance penalty term, the detailed features and the contrast features to obtain the reasoning result.

[0136] In some specific embodiments, the stratigraphic correlation profile generation module 16 may specifically include:

[0137] The depth prediction result determination unit is used to select adjacent node pairs with increasing depths in the multi-well to be processed, and use a preset depth value determination model to predict the adjacent node pairs to obtain depth prediction results. Then, an activation function is used to penalize the prediction results that are out of order in the depth prediction results to obtain the sequence monotonicity constraint loss.

[0138] The absolute value determination unit for the difference in burial depth of strata is used to determine the absolute value of the difference in burial depth of strata at each node between adjacent wells, and to truncate the absolute value of the difference in burial depth of strata based on a preset fault tolerance threshold to obtain the fault tolerance truncation constraint loss. Then, the intelligent model for generating stratigraphic correlation profiles is used to generate stratigraphic correlation profiles based on the inference results, the sequence monotonicity constraint loss and the fault tolerance truncation constraint loss to obtain stratigraphic correlation profiles.

[0139] The decoding result determination unit is used to decode the stratigraphic correlation profile to obtain decoding results that characterize the interface information of continuous stratigraphy, and convert the decoding results into a structured multi-well stratigraphic layer data table based on a preset geological naming rule; the data table includes the layer group name, the sub-layer name, and the top and bottom boundary depths of each stratum;

[0140] The profile correction instruction determination unit is used to generate a well-to-well comparison profile and a three-dimensional geological grid in real space using a three-dimensional visualization engine and based on the multi-well formation layer data table and the three-dimensional coordinate information of each node, so as to determine the corresponding profile correction instruction using the three-dimensional geological grid and based on the well-to-well comparison profile.

[0141] The instruction conversion unit is used to display the well-connected comparison profile and the profile correction instruction on the interactive interface corresponding to the three-dimensional visualization engine, and to convert the profile correction instruction into geological prior constraint samples, so as to feed the geological prior constraint samples back to the stratigraphic comparison profile generation intelligent model for incremental learning, and obtain a new stratigraphic comparison profile generation intelligent model.

[0142] Furthermore, embodiments of this application also disclose an electronic device, Figure 4This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraints optimization disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0143] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0144] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0145] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the multi-well stratigraphic correlation method based on spatially aware heterogeneous maps and geological constraints optimization, which is executed by the electronic device 20 according to any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0146] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed multi-well stratigraphic correlation method based on spatially aware heterogeneous maps and geological constraint optimization. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0147] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0148] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0149] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0150] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0151] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization, characterized in that, include: From a number of logging curves, identify the logging curves that meet the preset lithological differentiation sensitivity conditions to be processed; The first and second derivatives of the logging curve to be processed are calculated to obtain the target logging curve to be processed. Then, the three-dimensional coordinate information of each sampling point is determined based on the data of each well. Then, a three-dimensional spatial feature vector with implicit geometric correction is constructed using nonlinear mapping rules and based on the three-dimensional coordinate information and wellbore attitude information. The well data includes inclination data and well trajectory data. Well logging curve features are extracted from the target well logging curve to be processed, and the well logging curve features are fused with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes; Based on each sampling point and the corresponding node features, nodes are constructed, and vertical edges within the well for representing sedimentary sequences, spatial edges between wells for representing potential stratigraphic connectivity, and geological a priori edges for cross-well connections are constructed between each node to obtain a heterogeneous graph structure; each node is a local micro-stratigraphic unit centered on the sampling point, including curve morphology and contextual information. A marker layer framework is constructed based on the heterogeneous graph structure, and layer comparison and reasoning are performed based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information; A smart model is generated using stratigraphic correlation profiles. Based on the geological hard constraint loss function, which includes sequence monotonicity constraints and fault tolerance truncation constraints, and the inference results, a stratigraphic correlation profile is generated. The stratigraphic correlation profile is then decoded to obtain a stratigraphic data table of isochronous strata, including stratigraphic group names, sub-layer names, and top and bottom depths.

2. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization as described in claim 1, characterized in that, The process of determining the logging curves to be processed from a plurality of logging curves that meet the preset lithology differentiation sensitivity conditions includes: Obtain well logging curves in the target work area, and determine the geological sedimentary pattern and lithological combination corresponding to the target work area. Then, based on the geological sedimentary pattern and the lithological combination, perform a preset lithological differentiation sensitivity judgment on the well logging curves to obtain a main comparison curve that meets the preset lithological differentiation sensitivity condition; the main comparison curve is a gamma curve. Auxiliary curves are selected from the remaining logging curves. Then, based on the auxiliary curves and the main comparison curve, the curve to be processed is determined. The curve to be processed is then subjected to wellbore expansion detection, noise identification, and tool failure segment detection using a preset physical response mechanism to obtain the logging curve after detection. Then, the logging curve after detection is subjected to distortion interval removal processing to obtain the logging curve to be processed. The auxiliary curves include shallow and deep resistivity curves and sonic transit time curves. The auxiliary curves are used to describe the reservoirs, tight layers, and thin interbedded layers of multiple wells.

3. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization according to claim 1, characterized in that, The process involves calculating the first and second derivatives of the logging curve to be processed to obtain the target logging curve. Then, based on the well data, the three-dimensional coordinate information of each sampling point is determined. Finally, using a nonlinear mapping rule and based on the three-dimensional coordinate information and wellbore attitude information, an implicitly geometrically corrected three-dimensional spatial feature vector is constructed, including: The first derivative of the logging curve to be processed is calculated using a preset central difference algorithm to obtain the corresponding curve change rate feature, and the second derivative of the logging curve to be processed is calculated to obtain the corresponding inflection point feature. The target logging curve to be processed is determined based on the curve change rate feature, the inflection point feature and the logging curve to be processed. The multi-well data, including well inclination angle, azimuth angle, and multi-well depth, corresponding to each of the multi-wells are determined. A preset coordinate calculation algorithm is used to determine the three-dimensional coordinate information of each sampling point based on the well inclination angle, the azimuth angle, and the multi-well depth. The multi-well depth includes vertical depth and tilt depth. The three-dimensional coordinate information includes geodetic coordinates and true vertical depth. Based on the combination of the three-dimensional coordinate information and the wellbore attitude information, a wellbore trajectory geometric feature vector is constructed. The wellbore trajectory geometric feature vector is then input into a multilayer perceptron with learnable parameters. The multilayer perceptron is used to perform linear transformation and nonlinear activation operations on the wellbore trajectory geometric feature vector based on nonlinear mapping rules to obtain an implicitly geometrically corrected three-dimensional spatial feature vector. The three-dimensional spatial feature vector is used to characterize the deviation of the logging curve in representing the formation geometry.

4. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization according to claim 1, characterized in that, The step involves extracting logging curve features from the target logging curve to be processed, fusing the logging curve features with the three-dimensional spatial feature vector, and obtaining node features including spatial and physical attributes, including: A time-frequency transform is performed on the time-frequency domain corresponding to the target logging curve to be processed using a preset continuous wavelet transform algorithm to obtain a two-dimensional time-frequency domain feature curve to be extracted, including scale and depth dimensions. Then, a preset scale factor is used to extract features from the feature curve to be extracted to obtain logging curve features. The logging curve features include trend features for characterizing sedimentary cycle trends, detail features for characterizing lithological texture details, and contrast features for characterizing local layer boundary alignment. The target logging curve is processed using Hilbert transform to obtain the instantaneous phase information of the target logging curve in the depth domain; the instantaneous phase information is used to assist in the location of formation boundaries. The logging curve features, the instantaneous phase information, and the three-dimensional spatial feature vector are fused to obtain node features corresponding to each sampling point; the node features include the trend, texture, interface sensitivity, and wellbore trajectory of the target logging curve to be processed.

5. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization according to claim 4, characterized in that, The process involves constructing nodes based on each sampling point and its corresponding node features, and then constructing intra-well vertical edges to characterize sedimentary sequences, inter-well spatial edges to characterize potential formation connectivity, and geological prior edges for cross-well connections between these nodes, resulting in a heterogeneous graph structure, including: The node features corresponding to each sampling point are determined, and nodes are constructed based on the node features and the corresponding sampling points. Then, the nodes in the same well are connected sequentially in order of node depth from shallow to deep, and the connection results are encoded to obtain the vertical edge in the well. The node attribute corresponding to the node is the node feature corresponding to the sampling point. Determine the spatial location corresponding to each node, determine the three-dimensional Euclidean distance between the nodes of each multi-well, and connect the nodes whose distances in the three-dimensional Euclidean distances are less than a preset spatial neighborhood threshold to obtain the inter-well spatial edge used to characterize the potential formation connectivity relationship. Based on the preset geological stratification conclusions and marker layer information, nodes belonging to the same geological stratum in each of the multiple wells are connected across wells to obtain geological a priori edges for cross-well connections. Then, each node, the vertical edge within the well, the spatial edge between wells, and the geological a priori edges are integrated to obtain a heterogeneous graph structure. The heterogeneous graph structure is used to characterize the vertical sequence relationship and the lateral spatial connectivity relationship of the multiple wells.

6. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization according to claim 5, characterized in that, The process involves constructing a marker layer framework based on the heterogeneous graph structure, and performing layer comparison and reasoning based on the local subgraphs and spatial distance penalty terms within the marker layer framework to obtain the reasoning result. The spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information, including: Trend features are extracted from the heterogeneous map structure, and stratigraphic correlation reasoning is performed on the heterogeneous map structure based on the trend features to obtain key marker layers, so as to construct a marker layer framework for describing the structural framework of sandstone assemblages using the key marker layers. Based on the marker layer framework and the three-dimensional coordinate information, the heterogeneous map structure is divided into several local sub-maps corresponding to different geological units. Detail features and contrast features are extracted within each local sub-map. Then, a spatial distance penalty term is determined based on the three-dimensional Euclidean distance between each node. Layer comparison and reasoning are performed based on the spatial distance penalty term, the detail features, and the contrast features to obtain the reasoning result.

7. The multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraint optimization according to any one of claims 1 to 6, characterized in that, The method utilizes stratigraphic correlation profiles to generate intelligent models and, based on geological hard constraint loss functions including sequence monotonicity constraints and fault tolerance truncation constraints, generates stratigraphic correlation profiles using the inference results. These stratigraphic correlation profiles are then decoded to obtain a stratigraphic data table of isochronous strata, including stratigraphic group names, sub-layer names, and top and bottom depths. In the well to be processed, adjacent node pairs with increasing depth are selected, and the adjacent node pairs are predicted using a preset depth value determination model to obtain depth value prediction results. Then, an activation function is used to penalize the prediction results that are out of order in the depth value prediction results to obtain the sequence monotonicity constraint loss. The absolute value of the difference in formation depth between nodes between adjacent wells is determined, and the absolute value of the difference in formation depth is truncated based on a preset fault tolerance threshold to obtain the fault tolerance truncation constraint loss. Then, the formation correlation profile generation intelligent model is used to generate a formation correlation profile based on the inference results, the sequence monotonicity constraint loss and the fault tolerance truncation constraint loss to obtain the formation correlation profile. The stratigraphic correlation profile is decoded to obtain decoding results that characterize the interface information of continuous stratigraphy. Based on a preset geological naming rule, the decoding results are converted into a structured stratigraphic layer data table. The data table includes the name of the layer group, the name of the sublayer, and the top and bottom boundary depths of each stratum. Using a 3D visualization engine and based on the strata data table and the 3D coordinate information of each node, a well-to-well comparison profile and a 3D geological framework are generated in real space. The corresponding profile correction instructions are then determined using the 3D geological framework and based on the well-to-well comparison profile. The well-connection comparison profile and the profile correction command are displayed on the interactive interface corresponding to the three-dimensional visualization engine. The profile correction command is converted into geological prior constraint samples. The geological prior constraint samples are fed back to the stratigraphic comparison profile generation intelligent model for incremental learning to obtain a new stratigraphic comparison profile generation intelligent model.

8. A multi-well stratigraphic correlation device based on spatially perceived heterogeneous maps and geological constraints optimization, characterized in that, include: The logging curve determination module is used to determine the logging curve to be processed from several logging curves that meet the preset lithology differentiation sensitivity conditions. The coordinate information determination module is used to calculate the first and second derivatives of the logging curve to be processed to obtain the target logging curve to be processed. Then, based on the data of each well, it determines the three-dimensional coordinate information of each sampling point. Then, it uses nonlinear mapping rules and constructs an implicit geometric correction three-dimensional spatial feature vector based on the three-dimensional coordinate information and the wellbore attitude information. The well data includes inclination data and well trajectory data. The node feature determination module is used to extract logging curve features from the target logging curve to be processed, and fuse the logging curve features with the three-dimensional spatial feature vector to obtain node features including spatial attributes and physical attributes. The heterogeneous graph structure generation module is used to construct nodes based on each sampling point and the corresponding node features, and to construct intra-well vertical edges for representing sedimentary sequences, inter-well spatial edges for representing potential stratigraphic connectivity, and geological a priori edges for cross-well connections between the nodes, thereby obtaining a heterogeneous graph structure; the nodes are local micro-stratigraphic units centered on the sampling points and including curve morphology and contextual information. The reasoning result generation module is used to construct a marker layer framework based on the heterogeneous graph structure, and perform layer comparison and reasoning based on the local subgraphs in the marker layer framework and the spatial distance penalty term to obtain the reasoning result; the spatial distance penalty term is a penalty term determined based on the three-dimensional coordinate information; The stratigraphic correlation profile generation module is used to generate an intelligent model using the stratigraphic correlation profile and generate a stratigraphic correlation profile based on the geological hard constraint loss function including sequence monotonicity constraint and fault tolerance truncation constraint and the inference result. The stratigraphic correlation profile is then decoded to obtain a stratigraphic data table of isochronous strata including stratigraphic group name, sub-layer name and top and bottom depth.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the multi-well stratigraphic correlation method based on spatially perceived heterogeneous maps and geological constraints optimization as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the multi-well stratigraphic correlation method based on spatially aware heterogeneous maps and geological constraints optimization as described in any one of claims 1 to 7.

Images