Sedimentary facies constrained sequence domain single well heterogeneous curve high-resolution reconstruction method
By transforming well logging curves into sequence domains and constructing sample data using sedimentary facies and sequence coordinates, a high-resolution reconstruction network model was trained, solving the problem of poor accuracy in well logging curve reconstruction and achieving higher accuracy reconstruction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DAQING OILFIELD CO LTD
- Filing Date
- 2026-02-02
- Publication Date
- 2026-05-01
AI Technical Summary
Existing well logging curve reconstruction methods have poor accuracy under complex geological conditions. Neural networks have difficulty learning the well logging response patterns within sedimentary strata of the same geological age, and the strata differences between different wells cause training samples to be misaligned.
A high-resolution reconstruction method for heterogeneous single-well logging curves constrained by sedimentary facies is adopted. By acquiring logging curves within the target exploration block, the method converts them into logging curves in the sequence domain. Sample data is then constructed using sedimentary facies data and sequence coordinates to train a high-resolution reconstruction network model.
It improves the accuracy of well logging curve reconstruction, solves the problem of training samples not being aligned due to differences in formation thickness, and can capture geological change information more accurately.
Smart Images
Figure CN121956151A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of well logging curve analysis and reconstruction technology, specifically to a high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain. Background Technology
[0002] Well logging curves are curves that measure the changes in formation physical properties with depth using logging instruments. Different types of well logging curves reflect different formation physical properties and can help obtain formation information in oil and gas field exploration and development. However, some well logging curves may be missing or have low resolution, making it difficult to obtain formation information. Therefore, high-resolution reconstruction is necessary.
[0003] Currently, there are two main methods for reconstructing well logging curves: one is to synthesize missing or anomalous curves by establishing a mathematical model between the physical properties of the formation and the well logging response. However, this method relies on the selection of the physical model and the experience of the interpreters, making it difficult to adapt to complex geological conditions. The other method utilizes neural networks to learn the mapping relationship between different well logging curves and reconstructs other well logging curves based on known heterogeneous curves. However, existing neural networks typically process well logging data directly in the depth domain to construct training samples. But the formations of different wells may differ, and the training samples cannot be aligned in the depth direction. This makes it difficult for the neural network to learn the well logging response patterns within sedimentary strata of the same geological age, thus affecting the accuracy of well logging curve reconstruction. Summary of the Invention
[0004] To address the technical problem of poor reconstruction accuracy of well logging curves in existing methods, the present invention aims to provide a high-resolution reconstruction method for heterogeneous curves in single wells within a sedimentary facies-constrained sequence domain. The specific technical solution adopted is as follows: A high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain, the method comprising: Obtain the depth domain logging curves of each type in each well within the target exploration block, and determine the wells to be reconstructed and the types to be reconstructed within the target exploration block; in the wells to be reconstructed, the depth domain logging curves under the types to be reconstructed are taken as the logging curves to be reconstructed, and the depth domain logging curves under the types not to be reconstructed are taken as outlier curves. Acquire the stratigraphic boundary data and sedimentary facies data for each well; based on the stratigraphic boundary data, convert the well logging curves of each depth domain into sequence domain well logging curves, and determine the sedimentary facies corresponding to the sequence coordinates of each sampling point on the sequence domain well logging curve based on the sedimentary facies data; Sequence domain sample data is constructed based on the sequence coordinates and sedimentary facies of the sampling points on the non-reconstruction logging and its corresponding sequence domain logging curves, as well as the sequence domain logging curves of the reconstruction type in the non-reconstruction logging; a high-resolution reconstruction network model is trained using the sequence domain sample data. Based on the sequence coordinates and sedimentary facies of the sampling points on the heterogeneous curves in the well log to be reconstructed and their corresponding sequence domain well logs, the input data is constructed and fed into the trained high-resolution reconstruction network model. Based on the model output, the reconstruction curve of the well log to be reconstructed is determined.
[0005] Furthermore, the stratification boundary data should include at least the geological stratum name, top stratum depth, and bottom stratum depth for each geological stratum in the well logging; the sedimentary facies data should include at least the sedimentary stratum name and sedimentary stratum depth range for each sedimentary stratum in the well logging.
[0006] Furthermore, the method for obtaining the sequence domain logging curve includes: Based on the top and bottom boundary depths of the geological strata, a sequence coordinate normalization transformation is performed on the geological strata to determine the sequence coordinate axes in each geological strata. Based on the depth coordinates of the sampling points in the depth domain logging curve, the sequence coordinates within the geological strata are determined; based on the logging signal values of the sampling points in the depth domain logging curve and the sequence coordinates, the sequence domain logging curve is reconstructed.
[0007] Furthermore, the reconstructed sequence domain logging curves include: The sequence coordinate axes in each geological stratum are resampled to determine the sequence coordinates and depth coordinates of all resampled points; the depth interval between adjacent resampled points is no greater than the depth interval between adjacent sampling points in the depth logging curve. Linear interpolation is performed on the depth domain logging curves to determine the logging signal value at the corresponding depth coordinates for each resampling point; the sequence domain logging curves are reconstructed based on the logging signal values of all resampling points.
[0008] Furthermore, the method for determining the sedimentary phase includes: In each sequence domain logging curve, the depth coordinates corresponding to the sequence coordinates of each sampling point are determined, and the sedimentary facies of each sampling point are determined based on the depth range of the sedimentary layer in which the depth coordinates are located.
[0009] Furthermore, the method for obtaining the hierarchical sequence domain sample data includes: Within each non-reconstruction logging segment, a feature input vector is constructed for each sampling point at each sequence coordinate, based on the sequence coordinates, sedimentary facies, and logging signal value of the sampling points on the sequence domain logging curve of each heterogeneous curve. A feature output vector is constructed for each sampling point at each sequence coordinate, based on the logging signal value of the sampling points on the sequence domain logging curve of the reconstruction type. Sequence domain sample data is then constructed based on the feature input vector and feature output vector of the sampling points at each sequence coordinate.
[0010] Furthermore, the method for obtaining the feature input vector includes: The sedimentary facies are ordered and coded based on their physical properties to determine the sedimentary code for each facies. For each sequence coordinate downsampling point on each sequence domain logging curve, a sequence-sedimentary feature coding vector is constructed based on the sequence coordinates, the sedimentary coding, and the feature construction results of the sequence coordinates and the sedimentary coding. The logging signal value corresponding to the sampling point is concatenated with the corresponding sequence-sedimentary feature coding vector to construct a feature input vector.
[0011] Furthermore, the feature construction results include at least the trigonometric function transformation results of the sequence coordinates and the linear combination results of the sequence coordinates and the deposition coding.
[0012] Furthermore, the model architecture of the high-resolution reconstructed network model is a fully connected neural network.
[0013] Furthermore, the types of depth-domain logging curves include at least acoustic curves, density curves, spontaneous potential curves, and microelectrode curves.
[0014] The present invention has the following beneficial effects: This invention first acquires depth-domain logging curves of each type from each well within the target exploration block, determining the wells to be reconstructed and their types, and then identifying the logging curves to be reconstructed and outlier curves. This provides a data foundation for subsequent sequence domain conversion and the construction of sequence domain sample data to train the model. Next, it acquires the stratigraphic boundary data and sedimentary facies data for each well, thereby converting each depth-domain logging curve into a sequence domain logging curve. This avoids the problem of misalignment of formation information due to differences in geological layer thickness, facilitating subsequent model learning of logging response patterns within the formation. Finally, it determines the sequence coordinates corresponding to each sampling point on the sequence domain logging curve. The target sedimentary facies are used to prepare additional physical constraint features for subsequent analysis. Further, based on the sequence coordinates and sedimentary facies of the sampling points on the heterogeneous curves and their corresponding sequence domain well curves within the non-reconstruction wells, and the sequence domain well curves of the type to be reconstructed within the non-reconstruction wells, sequence domain sample data is constructed and a high-resolution reconstruction network model is trained. Further, based on the sequence coordinates and sedimentary facies of the sampling points on the heterogeneous curves and their corresponding sequence domain well curves within the to-be-reconstructed wells, input data is constructed and fed into the trained high-resolution reconstruction network model. Finally, the reconstructed curves of the to-be-reconstructed well curves within the to-be-reconstructed wells are determined based on the model output. This invention solves the problem of misalignment of training samples caused by differences in formation thickness in different wells by transforming from the depth domain to the sequence domain. Simultaneously, it introduces sedimentary facies and sequence coordinates to provide physical constraints and rhythmic features within the geology, thereby constructing sequence domain sample data to train the model. This allows the model to more accurately capture geological change information and improve reconstruction accuracy. Attached Figure Description
[0015] To more clearly illustrate the technical solutions and advantages 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 A flowchart illustrating a high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain, provided in an embodiment of the present invention; Figure 2 A comparison diagram of the reconstruction effect of an AC curve provided in one embodiment of the present invention; Figure 3 This is a comparison chart of the reconstruction effect of a DEN curve provided in one embodiment of the present invention. Detailed Implementation
[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain provided by this invention.
[0020] Please see Figure 1 The document illustrates a flowchart of a high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain, provided by an embodiment of the present invention. The method specifically includes: Step S1: Obtain the depth domain logging curves of each type in each well within the target exploration block, and determine the wells to be reconstructed and the types to be reconstructed within the target exploration block; in the wells to be reconstructed, the depth domain logging curves under the types to be reconstructed are taken as the logging curves to be reconstructed, and the depth domain logging curves under the non-reconstructed types are taken as outlier curves.
[0021] In one embodiment of the present invention, depth domain logging curves for each type of well logging in the target exploration block are first obtained. This provides a data foundation for the subsequent conversion of the logging curves from the depth domain to the sequence domain and the construction of sequence domain sample data to train the model and reconstruct high-resolution logging curves. The acquisition of depth domain logging curves is a well-known technical means and will not be described in detail here.
[0022] It should be noted that the target exploration block refers to an area with similar sedimentary background and consistent geological stratification system, that is, all well logging in the block has the same geological stratification standard and sedimentary facies classification standard.
[0023] In a preferred embodiment of the present invention, the types of depth domain logging curves include at least acoustic (AC) curves, density (DEN) curves, spontaneous potential (SP) curves, and microelectrode (RMG) curves; implementers may also obtain other types of depth domain logging curves according to actual needs.
[0024] Depth domain logging curves are used to characterize the variation of formation physical properties with depth. Each deep region logging curve is sampled at a fixed depth interval, with depth as the abscissa and the logging signal value at the sampling point under each depth as the ordinate, representing a discrete point sequence. In this embodiment, the sampling depth interval is set to 0.125m, but implementers can adjust it according to actual applications.
[0025] It should be noted that after obtaining the logging curve, certain preprocessing can be performed on the logging curve, such as interpolation to complete individual abnormal data points, removal of duplicate values, and anomaly correction; the above preprocessing methods are all well-known technical means and will not be elaborated further.
[0026] After obtaining the depth domain logging curves of each type in each well, the wells to be reconstructed and the corresponding types of the depth domain logging curves to be reconstructed within the target exploration area can be further determined (hereinafter referred to as the reconstructed types). Among them, the depth domain logging curves to be reconstructed are missing (not measured) or low-quality (low resolution, severely distorted by environmental influences) depth domain logging curves, which are evaluated and determined by geological professionals. The wells to be reconstructed are those containing the depth domain logging curves to be reconstructed.
[0027] In the logging to be reconstructed, the depth domain logging curves under the type to be reconstructed are used as the logging curves to be reconstructed, while the depth domain logging curves under the non-reconstruction type are used as outlier curves. A neural network model is used to learn and analyze the implicit nonlinear geophysical correlation between the outlier curves and the logging curves to be reconstructed in the non-reconstruction type, providing a basis for subsequent logging curve reconstruction.
[0028] In one embodiment of the present invention, any well log to be reconstructed is first selected, and it is assumed that the well log curve to be reconstructed is the acoustic curve (AC). Thus, the heterogeneous curves are determined to be the density curve (DEN), the spontaneous potential curve (SP), and the microelectrode curve (RMG). Taking the reconstruction of the acoustic curve (AC) in the well log to be reconstructed as an example, the analysis and description will not be repeated.
[0029] Step S2: Obtain the stratigraphic boundary data and sedimentary facies data for each well; based on the stratigraphic boundary data, convert the well logging curve of each depth domain into a sequence domain well logging curve, and determine the sedimentary facies corresponding to the sequence coordinates of each sampling point on the sequence domain well logging curve based on the sedimentary facies data.
[0030] Further, stratigraphic boundary data and sedimentary facies data for each well should be obtained. These data need to cover the well depth to provide a basis for subsequent conversion from the depth domain to the sequence domain. A sequence is a unit divided according to the geological age and sedimentary order; stratigraphic depositional patterns are similar within the same sequence.
[0031] It should be noted that the stratigraphic boundary data and sedimentary facies data need to be obtained by professional geologists based on the core data obtained from drilling, the logging data during the drilling process, and the existing high-quality depth domain logging curves (heterogeneous curves), combined with the regional geological characteristics, to manually identify or use software to divide each stratigraphic layer and determine the sedimentary environment type, thereby obtaining the stratigraphic boundary data and sedimentary facies data from the logging. The acquisition process will not be described in detail here.
[0032] In a preferred embodiment of the present invention, the stratification boundary data includes at least the geological stratum name, top stratum depth, and bottom stratum depth of each geological stratum in the well logging; the sedimentary facies data includes at least the sedimentary stratum name and sedimentary stratum depth range of each sedimentary stratum in the well logging.
[0033] Layer boundary data is used to determine the physical boundaries of each geological layer (i.e., the top and bottom boundaries of the layer), including the depth range information of each geological layer. Geological layers are divided according to the sedimentary age and lithological characteristics of the strata, thereby helping to determine the sequence. Strata within the same sequence have similar sedimentary patterns. Although the target exploration block has the same layering system, the thickness of geological layers at different logging locations may vary due to different geological structures. Determining the geological layer boundaries or sequence can help with subsequent conversion from the depth domain to the sequence domain.
[0034] Sedimentary facies data provides information on the lithology and sedimentary environment of strata, including the start and end depth range of each sedimentary facies, thereby helping to determine the sedimentary facies of each geological stratum. Sedimentary facies refers to a stratigraphic unit formed under a specific sedimentary environment. The response characteristics of well logging curves of strata with different sedimentary facies are different. Therefore, sedimentary facies data can be used as additional physical constraint features for model training to help the model distinguish the differences in well logging response under different sedimentary environments.
[0035] After obtaining the stratification boundary data, the sequence logging can be further analyzed, and then the depth domain can be converted to the sequence domain based on the stratification boundary data, transforming the logging curves of each depth domain into logging curves of the sequence domain.
[0036] Sequence domain logging curves can avoid the problem of misalignment of formation information caused by differences in the thickness of geological strata. For example, the thickness of the same geological stratum may vary in different wells. Well A has a thickness of 10 meters for the same geological stratum, while Well B has a thickness of 5 meters for the same geological stratum. Data from the same geological location cannot be aligned on the depth axis. However, after switching to the sequence domain, the same geological strata from different wells can be mapped to a unified relative space, making the formations from different wells equivalently aligned, such as top boundary to top boundary, middle boundary to middle boundary, and bottom boundary to bottom boundary. This facilitates the subsequent model learning of the logging response patterns within the formation and prepares for reconstruction.
[0037] Preferably, in one embodiment of the present invention, considering that the depth can be converted into a relative positional ratio based on the depth range of each geological stratum, the interference of the geometric factor of formation thickness is eliminated, and the relative structural characteristics of the sedimentation within the geological stratum are preserved. That is, regardless of the variation in formation thickness, the geological stratum can be regarded as a unit 1, and thus the relative sequence coordinates within it can be determined. Furthermore, the sequence domain logging curve can be reconstructed by combining the logging signal values of the sampling points under the sequence coordinates, so as to convert the depth domain logging curve into the sequence domain logging curve. Therefore, the method for obtaining the sequence domain logging curve includes: Based on the top and bottom boundary depths of geological strata, a sequence coordinate normalization transformation is performed on the geological strata to determine the sequence coordinate axes in each geological stratum; based on the depth coordinates of the sampling points in the depth domain logging curves, the sequence coordinates within the geological stratum are determined; based on the logging signal values of the sampling points in the depth domain logging curves and the sequence coordinates, the sequence domain logging curves are reconstructed.
[0038] It should be noted that in one embodiment of the present invention, the ground depth is defined as 0, the underground depth is a positive value, and the depth value is greater the farther away from the ground.
[0039] As an example, for each geological layer k, the depth of the top boundary of the geological layer is first used. With the depth of the layered bottom boundary Determine the thickness of geological strata. Then, the layer thickness is treated as a unit 1 and a sequence coordinate normalization transformation is performed to determine the sequence coordinate axis. Through this transformation, the same formation in different well logs can be mapped to the normalized coordinate space of [0,1], thereby eliminating the influence of formation thickness differences on the alignment of depth data. Then, for any depth coordinate d within this geological stratum, its sequence coordinates can be expressed as: This allows us to determine the sequence coordinates of each sampling point in the depth-domain logging curve within its geological stratum. Sequence coordinate normalization can convert the absolute position (depth coordinates d) of a sampling point within the strata in the depth domain into its relative position (sequence coordinates d). ); for example when When the value is 0.5, it means that regardless of whether the actual thickness of the geological layer is 10 meters or 20 meters, the depth domain sampling point is always located in the middle of the geological layer.
[0040] In one embodiment of the present invention, when the thickness of a geological layer is less than a preset threshold such as 0.2m, the layer can be skipped without performing a sequence transformation, and the adjacent geological layers can be directly used for interpolation filling, which will not be elaborated further.
[0041] By further combining the logging signal values of the sampling points under each sequence coordinate in the depth domain logging curve, the sequence domain logging curve is reconstructed.
[0042] In a preferred embodiment of the present invention, considering resampling on the sequence coordinate axis of each geological stratum and then interpolating the depth domain logging curves to determine the logging signal value at each resampling point, a relatively high-resolution sequence domain logging curve can be constructed, providing a basis for subsequent high-resolution reconstruction; therefore, the reconstructed sequence domain logging curves include: The sequence coordinate axes in each geological stratum are resampled to determine the sequence coordinates and depth coordinates of all resampled points. The depth interval between adjacent resampled points is no greater than the depth interval between adjacent sampling points in the depth logging curve. The depth domain logging curve is linearly interpolated to determine the logging signal value at the corresponding depth coordinates of each resampled point. The sequence domain logging curve is reconstructed based on the logging signal values of all resampled points.
[0043] Specifically, for each geological stratum k, the number of resampling points is set to N, where N is a positive integer. The depth interval between the resampling points is then calculated as follows: The value of N needs to be determined based on the thickness of the geological strata and the sampling interval of the depth domain logging curve. After resampling, the depth interval between adjacent resampling points should not be greater than the sampling depth interval of the depth domain logging curve, so as to ensure that the detailed information of the logging curve is not lost and to improve the resolution.
[0044] In this embodiment, since the depth interval of the depth domain logging curve is 0.125 meters, if the formation thickness is 5 meters, the number of resampling points N within this geological stratum should be no less than 40; setting N to 64, where 64 is a power of 2, facilitates computer processing; the implementer can also adjust it themselves; furthermore, the sequence coordinates of each resampling point i can be determined in each geological stratum. Among them, the first resampling point Corresponding top boundary of the layer The last sampling point correspond .
[0045] Furthermore, the depth coordinates of resampling point i can be determined. with sequence coordinates The conversion relationship between them: An inverse transformation of the stratigraphic coordinates is performed to determine the depth coordinates of each resampling point i. Then, linear interpolation is performed on the depth domain logging curves to determine the depth coordinates corresponding to each resampling point i. The well logging signal values of the resampled points are used as the horizontal axis parameter, and the well logging signal values of the resampled points are used as the vertical axis parameter to reconstruct and fit the sequence domain well logging curve; among them, linear interpolation and reconstruction fitting are well-known techniques and will not be described in detail.
[0046] Further, based on sedimentary facies data, the sedimentary facies corresponding to the sequence coordinates of each sampling point on the sequence domain logging curve are determined. Sedimentary facies can serve as an additional physical constraint feature to help subsequent training models distinguish the differences in logging responses under different sedimentary environments.
[0047] Preferably, in one embodiment of the present invention, the method for determining the depositional phase includes: In each sequence domain logging curve, the depth coordinates corresponding to the sequence coordinates of each sampling point are determined, and the sedimentary facies of each sampling point are determined based on the depth range of the sedimentary layer where the depth coordinates are located.
[0048] Specifically, for each sequence domain logging curve, the depth coordinates corresponding to the sequence coordinates of each sampling point in the sequence domain logging curve are first determined. Then, the sedimentary facies data are traversed, and the depth range of the sedimentary layer to which the depth coordinates belong is found in the sedimentary facies data, thereby determining the name of the sedimentary layer to which it belongs.
[0049] Step S3: Construct sequence domain sample data based on the sequence coordinates and sedimentary facies of the sampling points on the non-reconstruction logging curves and their corresponding sequence domain logging curves, as well as the sequence domain logging curves of the reconstruction type in the non-reconstruction logging curves; and train a high-resolution reconstruction network model using the sequence domain sample data.
[0050] Sequence domain logging curves of high-quality heterogeneous curves (such as DEN curves, SP curves and RMG curves) in non-reconstruction logging, and sequence domain logging curves of high-quality depth domain logging curves (such as AC curves) under the reconstruction type are used to generate sequence domain sample data, which provides a data foundation for the training of subsequent high-resolution reconstruction network models. Furthermore, in the logging to be reconstructed, the sequence domain logging curves of missing (unmeasured) or low-quality (low resolution, severely distorted by environmental influences) depth domain logging curves can be inferred from the known high-quality heterogeneous logging curves in the sequence domain logging curves.
[0051] It should be noted that the reconstructable state of the depth domain logging curves may differ within each well. It is necessary to further screen out the wells where all depth domain logging curves are intact, i.e., the depth domain logging curves within the selected non-reconstructable wells are all intact, thereby providing a good data foundation for subsequent model training.
[0052] Considering that completely different geological environments may produce similar heterogeneous curve responses—for example, a high resistivity (logging signal value) in an RMG curve may correspond to tight sandstone or carbonate rock—logging signal values can only reflect physical characteristics to a certain extent and cannot provide accurate stratigraphic information. Introducing sedimentary facies can introduce geological background constraints, thereby improving the distinguishability of stratigraphic lithology information behind similar logging signal values. Simultaneously, geological sedimentation exhibits rhythmicity, characterized by vertical gradation. Introducing sequence coordinates and sedimentary facies can avoid ignoring the spatial gradation characteristics within geological strata, thus preventing the reconstruction curve from lacking detail. Therefore, in a preferred embodiment of the present invention, the method for obtaining sequence domain sample data includes: Within each non-reconstruction logging segment, a feature input vector is constructed for each sampling point at each sequence coordinate, based on the sequence coordinates, sedimentary facies, and logging signal value of the sampling points on the sequence domain logging curve of each heterogeneous curve. A feature output vector is constructed for each sampling point at each sequence coordinate, based on the logging signal value of the sampling points on the sequence domain logging curve of the reconstruction type. Sequence domain sample data is then constructed based on the feature input vector and feature output vector of the sampling points at each sequence coordinate.
[0053] In a preferred embodiment of the present invention, considering that the top and bottom physical properties of the same sedimentary facies are drastically different, simply encoding the sedimentary facies can easily overlook its internal variation rhythm. However, combining sequence coordinates not only introduces geological constraints to distinguish different sedimentary environments but also identifies the vertical gradation characteristics of the same sedimentary facies within strata, thereby enabling the neural network to capture sedimentary facies relationships and improve the reconstruction accuracy of different sedimentary environments. Based on this, the method for obtaining the feature input vector includes: Based on the physical properties of sedimentary facies, an ordered encoding of sedimentary facies is performed to determine the sedimentary code for each sedimentary facies. For each sampling point under each sequence coordinate on the sequence domain logging curve, a sequence-sedimentary feature encoding vector is constructed based on the sequence coordinates, sedimentary codes, and the feature construction results of sequence coordinates and sedimentary codes. The logging signal value corresponding to the sampling point is concatenated with the corresponding sequence-sedimentary feature encoding vector to construct a feature input vector.
[0054] As an example, sedimentary facies are first ordered based on their physical properties, and the numerical values of the sedimentary codes must strictly vary monotonically according to the physical properties. For example, based on grain size (from coarse to fine, the code values decrease sequentially): mudstone sedimentary facies (such as floodplain mud) are coded as 1, siltstone sedimentary facies (such as crevasse spurs) are coded as 2, fine sandstone sedimentary facies (such as floodplains) are coded as 3, and coarse sandstone sedimentary facies (such as channel sands) are coded as 4, thus determining the sedimentary code for each sedimentary facies; Furthermore, within each non-reconstruction logging, the sequence coordinates and sedimentary codes of each sampling point on the sequence domain logging curve of each heterogeneous curve are used to construct features; wherein, in a preferred embodiment of the present invention, the feature construction results include at least the trigonometric function transformation results of the sequence coordinates and the linear combination results of the sequence coordinates and sedimentary codes; Then, combining sequence coordinates and sedimentary coding, a sequence-sedimentary feature coding vector is constructed for each sequence coordinate downsampling point i; the sequence-sedimentary feature coding vector is represented as follows: ; in, and The result of the trigonometric transformation of the stratigraphic coordinates. and This is a linear combination of sequence coordinates and sedimentary codes; This indicates a sedimentary code used to distinguish different sedimentary environments; Sine coding representing sequence coordinates; in the middle of geological strata ( ), sinusoidal coding is 1; at the top or bottom boundary of geological stratification ( or The sinusoidal encoding is set to 0; it is used to distinguish the middle and boundary of geological strata. Cosine coding representing sequence coordinates at the top boundary of geological strata ( ), cosine coding is 1; at the bottom boundary of geological stratification ( The cosine encoding is set to -1; it is used to reflect the top and bottom gradation characteristics of geological stratification. The interaction term representing sedimentary facies and sequence coordinates allows different sedimentary facies to have differentiated coded responses at different sequence coordinates; it is used to reflect the vertical rhythmic characteristics within sedimentary facies. The interaction term representing sedimentary facies and complementary sequence coordinates, with Complementary, enhancing the encoding and expression capabilities.
[0055] Furthermore, the logging signal value corresponding to sampling point i is concatenated with the corresponding sequence-sedimentary feature encoding vector to construct a feature input vector with stronger expressive power; the feature input vector is represented as follows: ;in, It is the transpose symbol; The logging signal value of sampling point i in the SP curve; The logging signal value of sampling point i in the RMG curve; Let i be the logging signal value of sampling point i in the DEN curve.
[0056] Then, within each non-reconstruction logging session, the logging signal values of the sampling points on the sequence domain logging curve under the reconstruction type are... As the feature output vector of each sampling point under each hierarchical coordinate, the feature output vector is represented as follows: .
[0057] The feature input vector has a dimension of 8, which means it uses sequence domain logging curves of three heterogeneous curves, such as DEN curves, SP curves, and RMG curves, and 5-dimensional sequence-sedimentary feature encoding; the output feature vector has a dimension of 1, which means it uses sequence domain logging curves of the heterogeneous curve AC curve under the type to be reconstructed; implementers can also adjust the dimension according to the actual application. If more sequence domain logging curves of heterogeneous curves need to be used, the dimension of the feature input vector should be increased accordingly; if more sequence domain logging curves under the type to be reconstructed need to be reconstructed, the dimension of the feature output vector should be increased accordingly.
[0058] It should be noted that the vector elements in the input feature vector and the output feature vector also need to be normalized in the corresponding dimensions. Specifically, the minimum-maximum normalization method can be used to scale the vector elements in the corresponding dimensions to the range of [0, 1]. Since the resistivity values usually span multiple orders of magnitude, a logarithmic transformation (Log10) needs to be performed before normalization to avoid high values squeezing low-value features. Then, the input feature vector and output feature vector of each sampling point under each sequence coordinate in each non-reconstruction well are used as a sequence domain sample data to construct a training sample set; and a high-resolution reconstruction network model is further constructed and trained.
[0059] In a preferred embodiment of the present invention, the model architecture of the high-resolution reconstruction network model is a fully connected neural network, but implementers may also use other model architectures.
[0060] Training high-resolution reconstruction network models is a well-known technique. Here, we briefly describe its structure and general training process: Input layer: Receives feature input vectors The number of nodes in the input layer is equal to the dimension of the feature input vector. In this embodiment, the number of nodes in the input layer is 8. Hidden layer: Located between the input layer and the output layer, it is used to extract a non-linear representation of the input features; in this embodiment, there are 3 hidden layers, with 64, 32 and 16 neurons in each layer respectively; the hidden layer neurons use ReLU (corrected linear unit) as the activation function; Output layer: Outputs the predicted logging signal value in the curve to be reconstructed (normalized preprocessed value, which needs to be denormalized to restore the predicted value to the original value range); the number of nodes in the output layer is equal to the dimension of the output feature vector; in this embodiment, the number of nodes in the output layer is 1; the output layer does not use an activation function (or uses a linear activation function) because the logging signal value is a continuous real value and does not require nonlinear transformation.
[0061] The sequence domain sample data was divided into training, validation, and test sets in a 7:2:1 ratio. The training process employed backpropagation and gradient descent optimization, with mean squared error (MSE) as the loss function. Model training was implemented using the Keras framework based on TensorFlow. Key hyperparameters during training were set as follows: Adam optimizer, learning rate of 0.001, batch size of 64, and maximum number of training epochs of 500. To prevent overfitting, training was automatically terminated and the model weights were rolled back to the minimum value when the validation set loss stopped decreasing after 20 consecutive epochs. He Normal initialization was used to obtain the trained high-resolution reconstructed network model.
[0062] Step S4: Based on the heterogeneous curves in the well log to be reconstructed and the sequence coordinates and sedimentary facies of the sampling points on the corresponding sequence domain well logs, construct input data and input it into the trained high-resolution reconstruction network model. Based on the model output, determine the reconstruction curve of the well log to be reconstructed in the well log to be reconstructed.
[0063] After obtaining the trained high-resolution reconstruction network model, based on the method of constructing the input feature vector in step S3, the sequence coordinates and sedimentary facies of the sampling points on the heterogeneous curves in the well to be reconstructed and their corresponding sequence domain well curves are used to construct the input feature vector to be predicted, i.e., the input data. This will not be elaborated further.
[0064] The input data is then fed into the trained high-resolution reconstruction network model, and the model automatically outputs the prediction result of each input data, namely the predicted well logging signal value (normalized preprocessed value). The prediction result is then denormalized to restore it to the original value range. This is a well-known technique and will not be elaborated further.
[0065] Since the high-resolution reconstruction network model is trained in the sequence domain, the resulting predictions are also based on sequence domain data; further, based on the depth coordinates in step S2... with sequence coordinates The transformation relationship between the two is used to perform inverse sequence transformation on the sequence domain data to determine the depth coordinates; each predicted logging signal value is then stitched together in the order of the depth coordinates, and then interpolated to obtain the reconstructed curve of the logging curve to be reconstructed within the logging to be reconstructed; the specific stitching and interpolation fitting process will not be described in detail.
[0066] This method was applied in the interpretation of water-flooded reservoirs in old wells in the Lamadian oilfield (target exploration block). Sequence domain sample data was constructed and a high-resolution reconstruction network model was trained to reconstruct the missing acoustic curves AC and density curves DE.
[0067] In Example 1, 25 well logs with AC curves were selected for reconstruction verification. The input data were SP and RMG to reconstruct the AC curves; please refer to [link to example]. Figure 2 The diagram shows a comparison of the reconstruction effect of an AC curve provided by an embodiment of the present invention, with the average correlation coefficient between the reconstructed curve and the measured curve reaching 0.92. In Example 2, 25 well logs with DEN curves were selected for reconstruction verification. The input data were AC and RMG to reconstruct the DEN curves; please refer to [link to example]. Figure 3 The diagram shows a comparison of the reconstruction effect of a DEN curve provided by an embodiment of the present invention. The average correlation coefficient between the reconstructed curve and the measured curve reaches 0.90.
[0068] In summary, this invention first transforms depth-domain logging curves into sequence-domain logging curves based on the stratigraphic boundary and sedimentary facies data of each well, and determines the sedimentary facies corresponding to the sequence coordinates of each sampling point. Further, based on heterogeneous curves within the non-reconstruction wells and the sequence coordinates and sedimentary facies of the sampling points on their corresponding sequence-domain logging curves, as well as the sequence-domain logging curves of the type to be reconstructed within the non-reconstruction wells, sequence-domain sample data is constructed, and a high-resolution reconstruction network model is trained using this sample data. Based on heterogeneous curves within the to-be-reconstructed wells and the sequence coordinates and sedimentary facies of the sampling points on their corresponding sequence-domain logging curves, input data is constructed and fed into the trained high-resolution reconstruction network model. The reconstruction curve of the to-be-reconstructed logging curve within the to-be-reconstructed well is determined based on the model output. This invention solves the problem of misalignment of training samples caused by differences in formation thickness in different wells by transforming from the depth domain to the sequence domain. Simultaneously, by introducing sedimentary facies and sequence coordinates to construct sequence-domain sample data for model training, the model can more accurately capture geological change information and improve reconstruction accuracy.
[0069] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0070] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. A high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain, characterized in that: The method includes: Obtain the depth domain logging curves of each type in each well within the target exploration block, and determine the wells to be reconstructed and the types to be reconstructed within the target exploration block; in the wells to be reconstructed, the depth domain logging curves under the types to be reconstructed are taken as the logging curves to be reconstructed, and the depth domain logging curves under the types not to be reconstructed are taken as outlier curves. Acquire the stratigraphic boundary data and sedimentary facies data for each well; based on the stratigraphic boundary data, convert the well logging curves of each depth domain into sequence domain well logging curves, and determine the sedimentary facies corresponding to the sequence coordinates of each sampling point on the sequence domain well logging curve based on the sedimentary facies data; Sequence domain sample data is constructed based on the sequence coordinates and sedimentary facies of the sampling points on the non-reconstruction logging and its corresponding sequence domain logging curves, as well as the sequence domain logging curves of the reconstruction type in the non-reconstruction logging; a high-resolution reconstruction network model is trained using the sequence domain sample data. Based on the sequence coordinates and sedimentary facies of the sampling points on the heterogeneous curves in the well log to be reconstructed and their corresponding sequence domain well logs, the input data is constructed and fed into the trained high-resolution reconstruction network model. Based on the model output, the reconstruction curve of the well log to be reconstructed is determined.
2. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 1, characterized in that, The stratification boundary data should include at least the geological stratum name, top boundary depth, and bottom boundary depth of each geological stratum in the well logging; the sedimentary facies data should include at least the sedimentary stratum name and sedimentary facies depth range of each sedimentary stratum in the well logging.
3. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 2, characterized in that, The method for obtaining the sequence domain logging curves includes: Based on the top and bottom boundary depths of the geological strata, a sequence coordinate normalization transformation is performed on the geological strata to determine the sequence coordinate axes in each geological strata. Based on the depth coordinates of the sampling points in the depth domain logging curve, the sequence coordinates within the geological strata are determined; based on the logging signal values of the sampling points in the depth domain logging curve and the sequence coordinates, the sequence domain logging curve is reconstructed.
4. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 3, characterized in that, Reconstructed sequence domain logging curves include: The sequence coordinate axes in each geological stratum are resampled to determine the sequence coordinates and depth coordinates of all resampled points; the depth interval between adjacent resampled points is no greater than the depth interval between adjacent sampling points in the depth logging curve. Linear interpolation is performed on the depth domain logging curves to determine the logging signal value at the corresponding depth coordinates for each resampling point; the sequence domain logging curves are reconstructed based on the logging signal values of all resampling points.
5. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 2, characterized in that, The method for determining the sedimentary phase includes: In each sequence domain logging curve, the depth coordinates corresponding to the sequence coordinates of each sampling point are determined, and the sedimentary facies of each sampling point are determined based on the depth range of the sedimentary layer in which the depth coordinates are located.
6. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 1, characterized in that, The method for obtaining the hierarchical domain sample data includes: Within each non-reconstruction logging segment, a feature input vector is constructed for each sampling point at each sequence coordinate, based on the sequence coordinates, sedimentary facies, and logging signal value of the sampling points on the sequence domain logging curve of each heterogeneous curve. A feature output vector is constructed for each sampling point at each sequence coordinate, based on the logging signal value of the sampling points on the sequence domain logging curve of the reconstruction type. Sequence domain sample data is then constructed based on the feature input vector and feature output vector of the sampling points at each sequence coordinate.
7. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 6, characterized in that, The method for obtaining the feature input vector includes: The sedimentary facies are ordered and coded based on their physical properties to determine the sedimentary code for each facies. For each sequence coordinate downsampling point on each sequence domain logging curve, a sequence-sedimentary feature coding vector is constructed based on the sequence coordinates, the sedimentary coding, and the feature construction results of the sequence coordinates and the sedimentary coding. The logging signal value corresponding to the sampling point is concatenated with the corresponding sequence-sedimentary feature coding vector to construct a feature input vector.
8. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 7, characterized in that, The feature construction results include at least the trigonometric function transformation results of the sequence coordinates and the linear combination results of the sequence coordinates and the deposition code.
9. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 1, characterized in that, The high-resolution reconstructed network model has a fully connected neural network architecture.
10. The high-resolution reconstruction method for heterogeneous curves in a single well within a sedimentary facies-constrained sequence domain according to claim 1, characterized in that, The types of depth-domain logging curves include at least sonic curves, density curves, spontaneous potential curves, and microelectrode curves.