Intelligent prediction method and device for sand body thickness distribution under non-seismic sparse well pattern condition

Abstract

Embodiments of the present application provide a method and device for intelligent prediction of sand body thickness distribution under the condition of a sparse well pattern without seismic data. The method comprises: obtaining measured well point data of a to-be-predicted area; inputting the measured well point data into a sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map. Obtain multiple sets of sample well point data; determine target sample images corresponding to each set of sample well point data from multiple sample images in a sample image library; the sample image library is generated by parameterizing the sedimentary information of multiple configuration units; based on each set of sample well point data and the target sample image corresponding thereto, train an initial prediction model to obtain a sand body thickness distribution prediction map output by the model, and iteratively update the model parameters through a loss function until a training end condition is met; the loss function comprises a well point matching loss term and a structural similarity loss term. The method realizes knowledge-data collaborative prediction of sand body thickness distribution, and improves the prediction accuracy and reliability.

Description

[0001] This application claims priority to Chinese Patent Application No. 202610327998.9, filed on March 17, 2026, entitled "Model Training Method, Sand Body Thickness Distribution Prediction Method and Apparatus", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of oil and gas development technology, and in particular to an intelligent prediction method and device for sand body thickness distribution under seismic-free sparse well network conditions. Background Technology

[0003] Sandbody distribution prediction is a core component of oil and gas reservoir characterization, and its accuracy directly affects the refined evaluation and efficient development of reservoirs. In the development of deep, ultra-deep, and unconventional oil and gas reservoirs, the challenges of seismic-free sparse well networks are often encountered due to complex geological conditions, high drilling costs, and technological limitations.

[0004] In existing technologies, sand body thickness distribution is typically predicted using geostatistical methods and deep learning algorithms. Taking the kriging method as an example, geostatistical methods model spatial correlation using variograms and perform optimal unbiased estimation. However, under seismic-free, sparse well network conditions, the variogram modeling becomes distorted due to insufficient data. Furthermore, the variogram, based on two-point statistics, cannot inherently describe complex geometric configurations such as channels and lobes, leading to poor geological plausibility. Mainstream deep learning methods primarily rely on supervised learning models, requiring training with massive amounts of labeled samples. However, in seismic-free, sparse well network scenarios, the large well spacing makes it difficult to control the scale of sand body distribution, resulting in a training image library lacking universality and poor model conditioning. Moreover, existing model architectures are disconnected from geological mechanisms, leading to poor geological plausibility.

[0005] Therefore, there is an urgent need for a method to predict the thickness distribution of sand bodies under seismic-free sparse well network conditions, which combines geological prior knowledge and sparse well point data to drive model training, so as to achieve the prediction of sand body thickness distribution under seismic-free sparse well network conditions. Summary of the Invention

[0006] This application provides an intelligent prediction method and device for sand body thickness distribution under seismic sparse well network conditions, which realizes knowledge-data collaborative prediction of sand body thickness distribution and improves prediction accuracy and reliability.

[0007] In a first aspect, embodiments of this application provide an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions, including:

[0008] Obtain measured well point data for the area to be predicted;

[0009] The measured well point data of the area to be predicted are input into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted.

[0010] The training process of the sand body thickness distribution prediction model is as follows:

[0011] Acquire multiple sets of sample well point data;

[0012] For each group of sample well point data, a target sample image corresponding to the sample well point data is determined from multiple sample images in a pre-generated sample image library; the sample image library is generated by parameterizing the depositional information of the multiple configuration units.

[0013] Based on the sample well point data and their corresponding target sample images of each group, the initial prediction model is trained to obtain the sand body thickness distribution prediction map output by the model, and the model parameters are iteratively updated through the loss function until the training termination condition is met.

[0014] The loss function includes a wellpoint matching loss term and a structural similarity loss term. The wellpoint matching loss term is related to the deviation between the predicted sand body thickness output by the model and the measured sand body thickness of the sample wellpoint data input to the model. The structural similarity loss term is related to the similarity between the distribution map predicted by the model and the target sample image input to the model. The predicted sand body thickness output by the model is extracted from the predicted sand body thickness distribution map.

[0015] In one possible implementation, the model includes: a first sub-model and a second sub-model; the sand body thickness distribution prediction map includes an initial sand body thickness distribution prediction map and a final sand body thickness distribution prediction map;

[0016] The input of the first sub-model is the sample well point data, and the output is the initial sand body thickness distribution prediction map. The loss function of the first sub-model includes: well point matching loss term and structural similarity loss term.

[0017] The input to the second sub-model is the initial sand body thickness distribution prediction map output by the first sub-model, and the output is the optimized final sand body thickness distribution prediction map. The loss function of the second sub-model includes a well point matching loss term.

[0018] In one possible implementation, based on the sample well point data of each group and their corresponding target sample images, an initial prediction model is trained to obtain a sand body thickness distribution prediction map output by the model, including:

[0019] Based on the target sample images corresponding to the sample well point data machines in each group, the first sub-model outputs the initial sand body thickness distribution prediction map, and the first sub-model is trained based on the loss function of the first sub-model.

[0020] The initial sand body thickness distribution prediction map is discretized into a node network. Based on each node in the node network, the connection relationship and weight between the nodes are determined to obtain a weighted heterogeneous map.

[0021] The second sub-model performs feature propagation and topology optimization on the weighted heterogeneous graph to generate the final sand body thickness distribution prediction map, and trains the second sub-model based on the loss function of the second sub-model.

[0022] In one possible implementation, the training method further includes:

[0023] Obtain depositional information for multiple configuration units;

[0024] For each of the aforementioned configuration units, the deposition information of the configuration unit is parameterized, a thickness mapping function is constructed, and a target discrete point set is determined;

[0025] The target discrete point set is converted into a regular grid image using an interpolation algorithm, and the thickness data of each point in the regular grid image is adjusted in batches to generate a sample image library.

[0026] In one possible implementation, the thickness mapping function includes: a planar distribution function and a spatial probability density function. The parameterization of the deposition information of the configuration unit, the construction of the thickness mapping function, and the determination of the target discrete point set include:

[0027] The deposition information of the configuration unit is parameterized to construct the planar distribution function of the configuration unit, which is used to indicate the mapping relationship between the coordinates of each discrete point and the thickness data;

[0028] Normalize the plane distribution function to obtain the spatial probability density function;

[0029] The target discrete point set is determined by probability density screening based on multiple spatial probability density functions.

[0030] In one possible implementation, parameterizing the deposition information of the configuration unit to construct the planar distribution function of the configuration unit includes:

[0031] Based on the deposition information of the configuration unit, a thickness variation function of the configuration unit in different directions is constructed;

[0032] By combining the thickness variation functions in different directions, the planar distribution function of the configuration unit is obtained.

[0033] In one possible implementation, determining the target discrete point set by performing probability density screening based on multiple spatial probability density functions includes:

[0034] For each of the aforementioned configuration units, arbitrarily determine multiple random discrete points;

[0035] Based on the spatial probability density function of the configuration unit, the probability density of the plurality of random discrete points is filtered to determine the target discrete point;

[0036] The target discrete point set is determined based on the target discrete points of multiple configuration units.

[0037] In one possible implementation, based on the spatial probability density function of the configuration unit, probability density filtering is performed on the plurality of random discrete points to determine the target discrete point, including:

[0038] For each of the aforementioned random discrete points, the probability density value of the random discrete point is calculated based on the coordinate information of the random discrete point and the spatial probability density function;

[0039] Determine whether the probability density value is greater than a random threshold;

[0040] If so, the random discrete point is determined as the target discrete point.

[0041] Secondly, embodiments of this application provide an intelligent prediction device for sand body thickness distribution under seismic-free sparse well network conditions, comprising:

[0042] The second acquisition module is used to acquire measured well point data of the area to be predicted.

[0043] The generation module is used to input the measured well point data of the area to be predicted into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted.

[0044] The first acquisition module is used to acquire multiple sets of sample well point data;

[0045] The processing module is used to determine, for each group of sample well point data, a target sample image corresponding to the sample well point data from multiple sample images in a pre-generated sample image library; the sample image library includes multiple sample images, which are generated by parameterizing the depositional information of the multiple configuration units;

[0046] The training module is used to train the initial prediction model based on the sample well point data and the corresponding target sample image of each group, to obtain the sand body thickness distribution prediction map output by the model, and to iteratively update the model parameters through the loss function until the training termination condition is met.

[0047] The loss function includes a wellpoint matching loss term and a structural similarity loss term. The wellpoint matching loss term is related to the deviation between the predicted sand body thickness output by the model and the measured sand body thickness of the sample wellpoint data input to the model. The structural similarity loss term is related to the similarity between the distribution map predicted by the model and the target sample image input to the model. The predicted sand body thickness output by the model is extracted from the predicted sand body thickness distribution map.

[0048] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0049] The memory stores computer-executed instructions;

[0050] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0051] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0052] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0053] The intelligent prediction method and device for sand body thickness distribution under seismic-free sparse well network conditions provided in this application acquires measured well point data of the area to be predicted and inputs this data into a pre-trained sand body thickness distribution prediction model to generate a predicted sand body thickness distribution map of the area to be predicted. This overcomes the discontinuity and multiple solutions in the inter-well prediction results of purely data-driven methods in the prior art, improving the prediction robustness and geological reliability under seismic-free sparse well networks. By acquiring multiple sets of sample well point data, for each set of sample well point data, a target sample image corresponding to the sample well point data is determined from multiple sample images in a pre-generated sample image library. Based on each set of sample well point data and its corresponding target sample image, the initial prediction model is trained to obtain the sand body thickness distribution prediction map output by the model. The model parameters are iteratively updated using a loss function until the training termination condition is met. By constructing a sample image library by parameterizing the sedimentary information of multiple morphological units, the problem of insufficient inter-well information is compensated for, providing the model with massive prior mode constraints. By selecting target sample images that match the well point data of each group, the problem of low geological rationality is avoided, and knowledge-data collaborative prediction of sand body thickness distribution is realized, thereby improving prediction accuracy and reliability. Attached Figure Description

[0054] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0055] Figure 1 This is a schematic diagram of an oil and gas development scenario provided in an embodiment of this application;

[0056] Figure 2 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 1 ;

[0057] Figure 3 A schematic diagram of model training provided for an embodiment of this application. Figure 1 ;

[0058] Figure 4 A schematic diagram of model training provided for an embodiment of this application. Figure 2 ;

[0059] Figure 5 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 2 ;

[0060] Figure 6 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 3 ;

[0061] Figure 7 A sand body thickness distribution prediction method provided in this application embodiment Figure 1 ;

[0062] Figure 8 A sand body thickness distribution prediction method provided in this application embodiment Figure 2 ;

[0063] Figure 9 This is a schematic diagram of the structure of an intelligent prediction device for sand body thickness distribution under seismic sparse well network conditions provided in an embodiment of this application;

[0064] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0065] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0066] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0067] "Multiple" refers to two or more, and other quantifiers are similar. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0068] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0069] It should be noted that, in the embodiments of this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0070] Sand body distribution prediction is a core aspect of oil and gas reservoir characterization, and its accuracy directly affects the refined evaluation and efficient development of reservoirs. Figure 1 This is a schematic diagram of an oil and gas development scenario provided in an embodiment of this application, such as... Figure 1 As shown, in the development of deep, ultra-deep, and unconventional oil and gas, the challenges of seismic sparse well networks are often encountered due to complex geological conditions, high drilling costs, and technological limitations. In such scenarios, seismic data suffers from insufficient resolution, large well spacing, and inadequate inter-well information.

[0071] Based on the above scenarios, it is evident that in existing technologies, the Kriging method, which uses variograms to model spatial correlations and perform optimal unbiased estimation, suffers from data inaccuracies under seismic-free, sparse-well-network conditions. Furthermore, the variograms, based on two-point statistics, are inherently unable to describe complex geometric configurations such as channels and leaf-like structures, leading to poor geological plausibility. Deep learning algorithms, on the other hand, rely on massive amounts of labeled samples for training. However, in seismic-free, sparse-well-network scenarios, the limited wellpoint data makes it difficult to construct a universally applicable training image library, easily leading to model overfitting. Moreover, existing model architectures are disconnected from geological mechanisms, resulting in poor geological plausibility.

[0072] The intelligent prediction method for sand body thickness distribution under seismic sparse well network conditions provided in this application, by parameterizing sedimentary information to generate a sample image library for model training, and jointly driving model training with sparse well point data, solves the technical problems of low prediction accuracy and poor geological rationality of prediction results due to insufficient sample data in the prior art.

[0073] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0074] Figure 2 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 1 ,like Figure 2 As shown, the method includes:

[0075] S201. Obtain the measured well point data of the area to be predicted;

[0076] Well point data includes, but is not limited to, coordinate information and measured sand body thickness. The area to be predicted refers to the region where sand body thickness distribution needs to be predicted under seismic sparse well network conditions.

[0077] S202. Input the measured well point data of the area to be predicted into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted.

[0078] The training process for the sand body thickness distribution prediction model is as follows:

[0079] S203. Obtain multiple sets of sample well point data.

[0080] Each set of sample well point data includes well point data for at least one well point.

[0081] In this step, multiple sets of sample well point data are acquired for model training.

[0082] Optionally, well point data of multiple well points within the sample area can be obtained. At least one well point can be arbitrarily selected, and the multiple well points can be divided into multiple groups of well points. Based on the coordinate information of each group of well points and the measured value of sand body thickness, multiple groups of sample well point data can be obtained.

[0083] Optionally, well point data of multiple well points within the sample area can be obtained, and the well point data of multiple well points within each well area can be used as a group of sample well point data according to the well area division within the sample area.

[0084] Optionally, well point data from multiple sample areas can be acquired, and the coordinate information of the well points in each sample area and the measured value of the sand body thickness can be used as a set of sample well point data. Based on the well point data from multiple sample areas, multiple sets of sample well point data can be determined.

[0085] S204. For each group of sample well point data, determine the target sample image corresponding to the sample well point data from multiple sample images in the pre-generated sample image library.

[0086] The sample image library is generated by parameterizing the deposition information of multiple configuration units. The sample image library includes multiple sample images.

[0087] A morphological unit refers to a three-dimensional geological body that constitutes a sedimentary system. A morphological unit has a specific internal structure, external geometry, lithological assemblage, and boundary characteristics. Examples of morphological units include channels and lobes.

[0088] Sedimentary information is used to indicate the sedimentary characteristics of morphological units, such as sand body boundaries, morphological unit types, etc.

[0089] Parameterizing the depositional information of multiple configuration units refers to converting the depositional information of configuration units into mathematical functions or probability models, which are then used to guide the generation of sample images.

[0090] The target sample image refers to the sample image determined from the sample image library that matches the well point data of each group of samples. There may be one or more target sample images, and this application does not limit this.

[0091] Under seismic-free sparse well network conditions, existing technologies struggle to control the scale of sand body distribution due to large well spacing, and insufficient integration of geological knowledge leads to poor spatial continuity and low geological rationality in sand body distribution prediction results.

[0092] In this step, for each group of sample well point data, well point features are extracted from each group of sample well point data. Based on the well point features, target sample images that match each group of sample well point data are selected from multiple sample images in a pre-generated sample image library.

[0093] Specifically, for each group of sample well point data, well point features of each group of sample well point data are extracted, and the similarity between the well point features of each group and the features of the sample images is calculated. For example, the similarity between well point features and features of sample images can be measured by cosine similarity, Euclidean distance, etc. Sample images with similarity greater than a preset threshold are used as target sample images to match the group of sample well point data.

[0094] By constructing a sample image library using sedimentary information from multiple configuration units parameterized by seismically detached well networks, the problem of missing inter-well information due to large well spacing was solved, providing a massive amount of prior pattern constraints for the model. By selecting target sample images that match the well point data of each group of samples, the problem of low geological plausibility was avoided.

[0095] S205. Based on the sample well point data of each group and their corresponding target sample images, the initial prediction model is trained to obtain the sand body thickness distribution prediction map output by the model, and the model parameters are iteratively updated through the loss function until the training termination condition is met.

[0096] The loss function includes a well point matching loss term and a structural similarity loss term.

[0097] The wellpoint matching loss term is related to the deviation between the predicted sand body thickness output by the model and the measured sand body thickness from the sample wellpoint data input to the model. It is used to indicate the error between the predicted and measured sand body thickness values ​​at each wellpoint. Examples of wellpoint matching loss terms include mean squared error loss and mean absolute error loss.

[0098] The structural similarity loss term is correlated with the similarity between the model-predicted distribution map and the target sample image input to the model. It is used to indicate the similarity between the model-output sand body thickness distribution prediction map and the target sample image in three dimensions: brightness, contrast, and structure.

[0099] The predicted sand body thickness output by the model is extracted from the predicted sand body thickness distribution map.

[0100] The initial prediction model takes sample well data as input and outputs a predicted sand body thickness distribution map. The initial prediction model can be a single network or a combination of multiple networks.

[0101] Initial prediction models can be, for example, U-Net, Spatially Constrained Bayesian Network (SCB-Net), graph neural networks, etc.

[0102] The initial prediction model can also be a combination of U-Net and graph neural network, or a combination of SCB-Net and graph neural network. U-Net and SCB-Net are used to make predictions based on sample well point data to generate an initial sand body thickness distribution prediction map, and graph neural network is used to optimize the initial sand body thickness distribution prediction map to generate an optimized final sand body thickness distribution prediction map.

[0103] In this step, the initial prediction model is trained based on the sample well point data and their corresponding target sample images for each group, resulting in a predicted sand body thickness distribution map output by the model. A well point matching loss term is determined based on the deviation between the predicted sand body thickness value output by the model and the measured sand body thickness value from the sample well point data input to the model. A structural similarity loss term is determined based on the similarity between the distribution map predicted by the model and the target sample image input to the model. The well point matching loss term and the structural similarity loss term are weighted to obtain the loss function value, and the model parameters are updated until the training termination condition is met, resulting in a model for predicting sand body thickness distribution. The training termination condition could be, for example, the convergence of the loss function or reaching the maximum number of training iterations.

[0104] In one possible implementation, the model includes: a first sub-model and a second sub-model; the sand body thickness distribution prediction map includes an initial sand body thickness distribution prediction map and a final sand body thickness distribution prediction map;

[0105] The input to the first sub-model is sample well point data, and the output is an initial sand body thickness distribution prediction map. The loss function of the first sub-model includes: well point matching loss term and structural similarity loss term.

[0106] The input to the second sub-model is the initial sand body thickness distribution prediction map output by the first sub-model, and the output is the optimized final sand body thickness distribution prediction map. The loss function of the second sub-model includes: well point matching loss term.

[0107] By predicting and optimizing the sand body thickness distribution prediction map using the first and second sub-models respectively, a final sand body thickness distribution prediction map with strict matching well points and consistent geology was generated.

[0108] In one possible implementation, the initial prediction model is trained based on the sample well point data and their corresponding target sample images to obtain the sand body thickness distribution prediction map output by the model, which will be described in detail below:

[0109] Based on the target sample images corresponding to the well point data of each group, the first sub-model outputs an initial sand body thickness distribution prediction map, and trains the first sub-model based on its loss function. The initial sand body thickness distribution prediction map is discretized into a node network, and the connection relationship and weight between nodes are determined based on each node in the node network to obtain a weighted heterogeneous map. The second sub-model performs feature propagation and topology optimization on the weighted heterogeneous map to generate the final sand body thickness distribution prediction map, and trains the second sub-model based on its loss function.

[0110] For example, Figure 3 A schematic diagram of model training provided for an embodiment of this application. Figure 1 ,like Figure 3 As shown, each set of sample well point data is input into the first sub-model. Based on the coordinate information of each well point in each set of sample well point data and the measured value of sand body thickness, the features of each set of sample well point data are gradually extracted through a series of convolution and downsampling operations. At the same time, the size of the feature map is reduced, and then the initial sand body thickness distribution prediction map is generated through upsampling operations.

[0111] From the initial sand body thickness distribution prediction map, the predicted sand body thickness values ​​for each well point are extracted. The deviation between the measured and predicted sand body thickness values ​​is calculated to obtain the well point matching loss term. The structural similarity between the initial sand body thickness distribution prediction map and the target sample images corresponding to the well point data of this set is calculated, and the statistical value of the structural similarity of each target sample image is calculated to obtain the structural similarity loss term.

[0112] The loss function value of the first sub-model is obtained by weighting the well point matching loss term and the structural similarity loss term. The model parameters of the first sub-model are then updated based on the loss function value of the first sub-model until the training conditions are met. The training conditions may be, for example, the convergence of the loss function or reaching a preset number of training steps.

[0113] For example, Figure 4 A schematic diagram of model training provided for an embodiment of this application. Figure 2 ,like Figure 4 As shown, the initial sand body thickness distribution prediction map is input into the second sub-model, and the initial sand body thickness distribution prediction map is discretized into a node network. The node features of each node include coordinate information and the sand body thickness prediction value output by the first sub-model.

[0114] Since the configuration units have similar properties in the vicinity, for each node in the node network, the adjacent nodes within a preset distance of the node are connected to determine the spatial proximity edge, wherein the edge weight between the node and the adjacent node is negatively correlated with the distance.

[0115] Since units with similar sedimentary microfacies should have similar properties, for each node in the node network, similar nodes with similar predicted sand body thickness values ​​output by the first sub-model are connected to determine similar attribute edges to enhance the clustering optimization of units with the same configuration.

[0116] Based on the connection relationships and weights between nodes, a weighted heterogeneous graph is obtained. Through the second sub-model, graph convolution operation is performed on the weighted heterogeneous graph. The intensity of message passing is controlled by the normalized adjacency matrix to ensure that the feature update conforms to the strength of geological correlation represented by the edge weights.

[0117] Information propagates along spatially adjacent edges, forcing the optimization results to maintain spatial smoothness and continuity, suppressing isolated non-geological anomalies. Information propagates along edges with similar attributes, promoting the convergence of node characteristics belonging to the same structural unit, enhancing the internal homogeneity of the structural unit. After information aggregation through graph convolution and weight correction, the final sand body thickness distribution prediction map is generated.

[0118] From the final sand body thickness distribution prediction map, the predicted sand body thickness values ​​for each well point are extracted. The deviation between the measured sand body thickness value and the predicted sand body thickness value output by the second sub-model is calculated to obtain the well point matching loss term, i.e., the loss function value of the second sub-model. The model parameters of the second sub-model are updated by minimizing the loss function value of the second sub-model to obtain the second sub-model.

[0119] The prediction is made by using the first sub-model, and the output of the first sub-model is optimized by the second sub-model. This makes the final prediction result minimize the error at the well point while conforming to geological rationality and spatial continuity, thus improving the accuracy and reliability of sand body thickness distribution prediction.

[0120] This application provides an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions. It acquires measured well point data of the area to be predicted and inputs this data into a pre-trained sand body thickness distribution prediction model to generate a predicted sand body thickness distribution map for the area. This overcomes the discontinuity and multiple solutions in the inter-well prediction results of purely data-driven methods in existing technologies, improving the prediction robustness and geological reliability under seismic-free sparse well networks. By acquiring multiple sets of sample well point data, for each set of sample well point data, a target sample image corresponding to the sample well point data is determined from multiple sample images in a pre-generated sample image library. Based on each set of sample well point data and its corresponding target sample image, the initial prediction model is trained to obtain the sand body thickness distribution prediction map output by the model. The model parameters are iteratively updated using a loss function until the training termination condition is met. By constructing a sample image library using parameterized sedimentary information of multiple morphological units, the shortcomings are compensated for, providing the model with massive prior pattern constraints. By selecting target sample images that match the well point data of each group, the problem of low geological rationality is avoided, and knowledge-data collaborative prediction of sand body thickness distribution is realized, thereby improving prediction accuracy and reliability.

[0121] Figure 5 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 2 In this embodiment Figure 2 Based on the examples, supplementary explanations are provided for the intelligent prediction method of sand body thickness distribution under seismic sparse well network conditions, such as... Figure 3 As shown, the method includes:

[0122] S501. Obtain deposition information for multiple configuration units.

[0123] By combining modern sedimentary data, field outcrops, and literature, sedimentary information for multiple morphological units was obtained.

[0124] S502. For each configuration unit, parameterize the deposition information of the configuration unit, construct a thickness mapping function, and determine the target discrete point set.

[0125] The thickness mapping function is used to indicate the sand body thickness or occurrence probability at each point within the configuration unit. For example, the thickness mapping function can be constructed using spatial interpolation, trend surface analysis, or other methods.

[0126] In this step, for each configuration element, the configuration element is transformed into a thickness mapping function. Through the thickness mapping function, the function values ​​of multiple discrete points within the configuration element are calculated, and the target discrete points of the configuration element are selected based on the function values. The target discrete point set is obtained according to the set of target discrete points of each configuration element.

[0127] For example, the thickness of the sand body at each discrete point is determined according to the thickness mapping function, the thickness range of the sand body of the configuration unit is determined based on the sedimentary information, and it is determined whether the thickness of the sand body at each discrete point is within the thickness range of the sand body of the configuration unit. If so, the discrete point is taken as the target discrete point; otherwise, the discrete point is determined not to be the target discrete point.

[0128] For example, based on the thickness mapping function, the probability of occurrence of each discrete point is determined. Based on a preset threshold or a random threshold, it is determined whether the probability of occurrence of each discrete point is greater than the preset threshold or the random threshold. If so, the discrete point is determined to be the target discrete point; otherwise, the discrete point is determined not to be the target discrete point.

[0129] In one possible implementation, the thickness mapping function includes a planar distribution function and a spatial probability density function. The thickness mapping function is constructed based on the deposition information of the aforementioned parameterized configuration units, and the target discrete point set is determined in detail below:

[0130] The depositional information of the parameterized configuration units is used to construct the planar distribution function of the configuration units; the planar distribution function is normalized to obtain the spatial probability density function; based on multiple spatial probability density functions, probability density screening is performed to determine the target discrete point set.

[0131] The planar distribution function is used to indicate the mapping relationship between the coordinates of each discrete point and the thickness data. The planar distribution function can be a thickness variation function in a single direction, such as tangential to the source direction (x-direction), along the source direction (y-direction), or oblique, or it can be obtained by combining thickness variation functions in multiple directions.

[0132] The thickness variation function is used to mathematically characterize the spatial variation law of thickness. It is constructed based on the core function that controls the shape in this direction. The core function can be, for example, a Gaussian function, a polynomial function, etc.

[0133] The spatial probability density function is used to indicate the mapping relationship between the coordinates of each discrete point and the probability density, and the probability density is used to indicate the probability of the discrete point appearing.

[0134] The deposition information of parameterized configuration units is obtained. Based on core functions such as Gaussian functions and polynomial functions, the planar distribution function of the configuration unit is constructed. The planar distribution function is integrated and normalized over its domain Ω to obtain the spatial probability density function. Based on the spatial probability density function of each configuration unit, the probability density is screened to determine the target discrete point set.

[0135] For example, the planar distribution function H(x,y) is obtained by combining the thickness variation functions along the tangential and parallel directions of the material source. The spatial probability density function is obtained by integrating and normalizing the planar distribution function H(x,y) over its domain Ω using the following formula:

[0136]

[0137] Where f(x,y) is the spatial probability density function and H(x,y) is the planar distribution function.

[0138] By parameterizing the planar distribution function and spatial probability density function, probability density screening is performed to determine the target discrete point set, transforming sedimentary information into computable constraints and solving the problem of insufficient inter-well information.

[0139] In one possible implementation, the deposition information of the above-mentioned parameterized configuration units is used to construct the planar distribution function of the configuration units in detail:

[0140] Based on the depositional information of the configuration units, a thickness variation function of the configuration unit thickness in different directions is constructed; by combining the thickness variation functions in different directions, the planar distribution function of the configuration unit is obtained.

[0141] The thickness variation function in direction u is as follows:

[0142]

[0143] in: This is the thickness magnitude coefficient in the u-direction. Core functions for controlling the horizontal shape, such as Gaussian functions and polynomial functions. Control parameters for controlling the center position of thickness distribution. Characteristic parameters for controlling the spread range. This is the error term characterizing the random perturbation in the u direction. The domain defined in the u direction is constrained by the boundaries of the configuration elements. The u direction includes the tangential direction to the source (x direction) and the parallel direction to the source (y direction).

[0144] By combining the thickness variation functions in different directions, the planar distribution function of the configuration unit is obtained.

[0145]

[0146] in: It is a coupling function (such as multiplication, addition, or more complex nonlinear combinations) used to synthesize the thickness contributions in the x and y directions. It is a boundary indicator function, which takes a value of approximately 1 within the sand body boundary of the configuration unit and decays sharply to 0 outside the boundary, thereby achieving geological boundary constraints. Define the planar spatial domain for a preset region composed of multiple configuration units.

[0147] By constructing a planar distribution function using sedimentation information, the sedimentation information is transformed into computable constraints, providing a data foundation for determining the target discrete point set.

[0148] In one possible implementation, a detailed explanation is provided on determining the target discrete point set based on probability density filtering using multiple spatial probability density functions:

[0149] For each configuration unit, arbitrarily determine multiple random discrete points; based on the spatial probability density function of the configuration unit, perform probability density screening on the multiple random discrete points to determine the target discrete points; based on the target discrete points of multiple configuration units, determine the target discrete point set.

[0150] For each configuration unit, multiple random discrete points are arbitrarily determined by random point projection. Based on the spatial probability density function of the configuration unit, the probability density value of each random discrete point is determined. Based on the probability density value, random discrete points that conform to geological laws are retained as target discrete points. Based on the target discrete points of multiple configuration units, the target discrete point set is determined.

[0151] For example, the Monte Carlo rejection sampling algorithm is used to randomly select points in space, arbitrarily determining multiple random discrete points.

[0152] For example, the probability density value is compared with a preset threshold or a random threshold to determine whether the probability of occurrence of each discrete point is greater than the preset threshold or the random threshold. If so, the random discrete point is determined to be the target discrete point; otherwise, the random discrete point does not conform to geological laws and is not the target discrete point.

[0153] By randomly selecting points and using probability density filtering, target discrete points were determined, and random discrete points that conform to geological laws were retained as target discrete points.

[0154] In one possible implementation, the spatial probability density function based on the configuration unit is described in detail, and the probability density screening of multiple random discrete points is performed to determine the target discrete point:

[0155] For each random discrete point, the probability density value of the discrete point is calculated based on the coordinate information and spatial probability density function of the random discrete point; it is then determined whether the probability density value is greater than a random threshold; if so, the random discrete point is determined as the target discrete point.

[0156] The random threshold ranges from 0 to 1.

[0157] For each random discrete point, the probability density value of the random discrete point is calculated based on its coordinate information and spatial probability density function; it is then determined whether the probability density value is greater than a random threshold; if the probability density value is greater than the random threshold, the random discrete point is determined as the target discrete point; if the probability density value is less than or equal to the random threshold, the random discrete point is determined not to be the target discrete point.

[0158] By using a random threshold, diverse discrete points are retained, thereby generating diverse sample images. Retaining random discrete points with probability density values ​​greater than the random threshold ensures that the generation process of the target discrete point set follows structured geological laws, while the random threshold ensures that the target discrete points conform to the inherent randomness and heterogeneity within the geological body.

[0159] S503. Convert the target discrete point set into a regular grid image using an interpolation algorithm, and batch adjust the thickness data of each point on the regular grid image to generate a sample image library.

[0160] The target discrete point set is converted into a grid image. Each discrete point in the grid image includes thickness data determined based on a thickness mapping function. The thickness data of other points in the grid image is determined by an interpolation algorithm to obtain a regular grid image. The thickness data of each point in the regular grid image is then adjusted in batches to generate a sample image library.

[0161] For example, Figure 6 A flowchart illustrating an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions provided in this application embodiment. Figure 3 ,like Figure 6 As shown, the method mainly includes the following stages: sample generation stage, preliminary prediction stage, and optimization stage.

[0162] The sample generation stage comprises four phases: sedimentary information parameterization, Monte Carlo sampling, spatial interpolation and parameter adjustment, and generation of a sample image library. By acquiring sedimentary information from multiple morphological units within a predefined region, the sedimentary information of each morphological unit is parameterized, generating a spatial probability density function. Monte Carlo sampling generates a target discrete point set consistent with geological genesis, and through spatial interpolation and parameter adjustment, a structured sample image library with tens of thousands of entries is generated in batches. This sample image library represents a digital and structured expression of geological knowledge, providing massive prior pattern constraints for subsequent learning.

[0163] Preliminary prediction stage: includes selecting target sample images, U-net training, and outputting initial sand body thickness distribution predictions. Figure 3 The process involves several stages. Based on sample well data, target sample images matching geological patterns are selected from a sample image library and input into a feature-enhanced U-net network with an attention mechanism for training. In this process, well data serves as a hard constraint, while the geological pattern embodied in the image structure serves as a soft constraint. The two work together and iteratively learn to ultimately generate a preliminary, geologically plausible initial sand body thickness distribution prediction map.

[0164] Optimization phase: This includes constructing a graph structure that integrates address rules, graph neural network optimization, and outputting the final predicted sand body thickness distribution. Figure 3The process involves several stages. Using the initial sand body thickness distribution prediction map obtained in the preliminary prediction stage as input, a graph structure integrating spatial proximity and attribute similarity relationships is constructed. A graph convolutional neural network (GNN) is then used for feature propagation and topology optimization between nodes. The actual thickness data from well points serves as direct supervision, driving the network parameter updates of the GNN. Simultaneously, the graph structure itself encodes geological spatial association rules, ensuring that the optimization process maintains spatial continuity and conforms to geological topological logic while fitting well points. The final output is a sand body thickness distribution prediction map with strong geological consistency and high well point matching accuracy.

[0165] This application provides an intelligent prediction method for sand body thickness distribution under seismic-free sparse well network conditions. By acquiring sedimentary information from multiple configuration units, parameterizing the sedimentary information of each configuration unit, constructing a thickness mapping function, and determining a target discrete point set, the method converts the target discrete point set into a regular grid image using an interpolation algorithm. It then batch-adjusts the thickness data of each point on the regular grid image to generate a sample image library, transforming sedimentary information into a thickness mapping function. This solves the problem of insufficient inter-well information and addresses the issue of low prediction accuracy due to insufficient sample data in existing technologies. It provides a wealth of prior knowledge for model training, improving the accuracy and reliability of sand body thickness distribution prediction.

[0166] Figure 7 A sand body thickness distribution prediction method provided in this application embodiment Figure 1 ,like Figure 7 As shown, Figure 8 A sand body thickness distribution prediction method provided in this application embodiment Figure 2 . Figure 7 and Figure 8 The multiple black dots in the graph represent well points, and the color scale is used to indicate the numerical range of sand body thickness: 0~17 meters, showing and revealing the geometry and thickness variation of the sand body.

[0167] For example, consider a shallow-water delta region to be predicted. Within this region, there are 45 actual drilled wells (3-5 km apart).

[0168] The measured well point data from 45 actual drilled wells were input into the sand body thickness distribution prediction model to generate a predicted sand body thickness distribution map for the area to be predicted. For example, it could be like... Figure 7 or Figure 8 The diagram shows the predicted thickness distribution of the sand body.

[0169] Optionally, the sand body thickness distribution prediction model includes a first sub-model and a second sub-model. The first sub-model performs a preliminary planar sand body thickness prediction for the shallow-water delta region to be predicted, obtaining an initial sand body thickness distribution prediction map, such as... Figure 7As shown, the second sub-model optimizes the output of the first sub-model to obtain the final predicted sand body thickness distribution map, as shown. Figure 8 As shown, the predicted sand body has continuous boundaries, a reasonable geological structure, and is highly consistent with the understanding of sedimentological interpretation.

[0170] Figure 9 This is a schematic diagram of the structure of an intelligent prediction device for sand body thickness distribution under seismic sparse well network conditions provided in an embodiment of this application, as shown below. Figure 9 As shown, the intelligent prediction device 9 for sand body thickness distribution under seismic sparse well network conditions provided in this embodiment includes: a second acquisition module 901, a generation module 902, a first acquisition module 903, a processing module 904, and a training module 905.

[0171] The second acquisition module 901 is used to acquire measured well point data of the area to be predicted.

[0172] The generation module 902 is used to input the measured well point data of the area to be predicted into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted.

[0173] The first acquisition module 903 is used to acquire multiple sets of sample well point data;

[0174] The processing module 904 is used to determine the target sample image corresponding to the sample well point data from multiple sample images in a pre-generated sample image library for each group of sample well point data; the sample image library includes multiple sample images, which are generated by parameterizing the depositional information of multiple configuration units;

[0175] Training module 905 is used to train the initial prediction model based on the sample well point data of each group and its corresponding target sample image, to obtain the sand body thickness distribution prediction map output by the model, and to iteratively update the model parameters through the loss function until the training termination condition is met.

[0176] The loss function includes: well point matching loss term and structural similarity loss term; the well point matching loss term is related to the deviation between the predicted sand body thickness value output by the model and the measured sand body thickness value of the sample well point data input to the model, and the structural similarity loss term is related to the similarity between the distribution map predicted by the model and the target sample image input to the model; the predicted sand body thickness value output by the model is extracted from the predicted sand body thickness distribution map.

[0177] In one possible implementation, the model includes: a first sub-model and a second sub-model; the sand body thickness distribution prediction map includes an initial sand body thickness distribution prediction map and a final sand body thickness distribution prediction map; the input of the first sub-model is sample well point data, and the output is the initial sand body thickness distribution prediction map; the loss function of the first sub-model includes: a well point matching loss term and a structural similarity loss term; the input of the second sub-model is the initial sand body thickness distribution prediction map output by the first sub-model, and the output is the optimized final sand body thickness distribution prediction map; the loss function of the second sub-model includes: a well point matching loss term.

[0178] In one possible implementation, the training module 905 is further configured to, based on the target sample images corresponding to the well point data of each group of samples, output an initial sand body thickness distribution prediction map through a first sub-model, and train the first sub-model based on the loss function of the first sub-model; discretize the initial sand body thickness distribution prediction map into a node network, determine the connection relationship and weight between nodes based on each node in the node network, and obtain a weighted heterogeneous map; through a second sub-model, perform feature propagation and topology optimization on the weighted heterogeneous map to generate a final sand body thickness distribution prediction map, and train the second sub-model based on the loss function of the second sub-model.

[0179] In one possible implementation, the first acquisition module 903 is further configured to acquire deposition information of multiple configuration units; the processing module 904 is further configured to parameterize the deposition information of each configuration unit, construct a thickness mapping function, and determine the target discrete point set; convert the target discrete point set into a regular grid image through an interpolation algorithm, and batch adjust the thickness data of each point in the regular grid image to generate a sample image library.

[0180] In one possible implementation, the thickness mapping function includes a planar distribution function and a spatial probability density function. The processing module 904 is also used to parameterize the deposition information of the configuration unit, construct the planar distribution function of the configuration unit, and the planar distribution function is used to indicate the mapping relationship between the coordinates of each discrete point and the thickness data. The planar distribution function is normalized to obtain the spatial probability density function. Based on multiple spatial probability density functions, probability density filtering is performed to determine the target discrete point set.

[0181] In one possible implementation, the processing module 904 is further configured to construct a thickness variation function of the configuration unit in different directions based on the deposition information of the configuration unit; and to combine the thickness variation functions in different directions to obtain the planar distribution function of the configuration unit.

[0182] In one possible implementation, the processing module 904 is further configured to: arbitrarily determine multiple random discrete points for each configuration unit; perform probability density filtering on the multiple random discrete points based on the spatial probability density function of the configuration unit to determine the target discrete point; and determine the target discrete point set based on the target discrete points of the multiple configuration units.

[0183] In one possible implementation, the processing module 904 is further configured to calculate the probability density value of each random discrete point based on the coordinate information and spatial probability density function of the random discrete point; determine whether the probability density value is greater than a random threshold; if so, determine the random discrete point as the target discrete point.

[0184] This embodiment provides an intelligent prediction device for sand body thickness distribution under seismic sparse well network conditions. It can execute the intelligent prediction method for sand body thickness distribution under seismic sparse well network conditions provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0185] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 10 As shown, the electronic device 100 provided in this embodiment includes at least one processor 1001 and a memory 1002. Optionally, the device 100 further includes a communication component 1003. The processor 1001, memory 1002, and communication component 1003 are connected via a bus 1004.

[0186] In a specific implementation, at least one processor 1001 executes computer execution instructions stored in memory 1002, causing at least one processor 1001 to perform the above-described method.

[0187] The specific implementation process of processor 1001 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0188] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0189] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0190] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0191] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0192] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0193] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0194] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0195] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0196] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0197] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0198] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0199] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0200] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for intelligent prediction of sand body thickness distribution under seismic-free sparse well network conditions, characterized in that, include: Obtain measured well point data for the area to be predicted; The measured well point data of the area to be predicted are input into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted. The training process of the sand body thickness distribution prediction model is as follows: Acquire multiple sets of sample well point data; For each group of sample well point data, a target sample image corresponding to the sample well point data is determined from multiple sample images in a pre-generated sample image library; The multiple sample image libraries are generated by parameterizing the deposition information of multiple configuration units; Based on the sample well point data and their corresponding target sample images of each group, the initial prediction model is trained to obtain the sand body thickness distribution prediction map output by the model, and the model parameters are iteratively updated through the loss function until the training termination condition is met. The loss function includes: a well point matching loss term and a structural similarity loss term; The well point matching loss term is related to the deviation between the predicted sand body thickness output by the model and the measured sand body thickness of the sample well point data input to the model. The structural similarity loss term is related to the similarity between the distribution map predicted by the model and the target sample image input to the model. The predicted sand body thickness output by the model is extracted from the predicted sand body thickness distribution map.

2. The method according to claim 1, characterized in that, The model includes: a first sub-model and a second sub-model; the sand body thickness distribution prediction map includes an initial sand body thickness distribution prediction map and a final sand body thickness distribution prediction map; The input of the first sub-model is the sample well point data, and the output is the initial sand body thickness distribution prediction map. The loss function of the first sub-model includes: well point matching loss term and structural similarity loss term. The input to the second sub-model is the initial sand body thickness distribution prediction map output by the first sub-model, and the output is the optimized final sand body thickness distribution prediction map. The loss function of the second sub-model includes a well point matching loss term.

3. The method according to claim 2, characterized in that, The initial prediction model is trained based on the sample well point data and their corresponding target sample images of each group to obtain the sand body thickness distribution prediction map output by the model, including: Based on the sample well point data and their corresponding target sample images of each group, the first sub-model outputs the initial sand body thickness distribution prediction map, and the first sub-model is trained based on the loss function of the first sub-model. The initial sand body thickness distribution prediction map is discretized into a node network. Based on each node in the node network, the connection relationship and weight between the nodes are determined to obtain a weighted heterogeneous map. The second sub-model performs feature propagation and topology optimization on the weighted heterogeneous graph to generate the final sand body thickness distribution prediction map, and trains the second sub-model based on the loss function of the second sub-model.

4. The method according to claim 1, characterized in that, The training method also includes: Obtain depositional information for multiple configuration units; For each of the aforementioned configuration units, the deposition information of the configuration unit is parameterized, a thickness mapping function is constructed, and a target discrete point set is determined; The target discrete point set is converted into a regular grid image using an interpolation algorithm, and the thickness data of each point in the regular grid image is adjusted in batches to generate a sample image library.

5. The method according to claim 4, characterized in that, The thickness mapping function includes a planar distribution function and a spatial probability density function. The parameterization of the deposition information of the configuration unit, the construction of the thickness mapping function, and the determination of the target discrete point set include: The deposition information of the configuration unit is parameterized to construct the planar distribution function of the configuration unit. The planar distribution function is used to indicate the mapping relationship between the coordinates of each discrete point and the thickness data. Normalize the plane distribution function to obtain the spatial probability density function; The target discrete point set is determined by probability density screening based on multiple spatial probability density functions.

6. The method according to claim 5, characterized in that, The parameterization of the deposition information of the configuration unit and the construction of the planar distribution function of the configuration unit include: Based on the deposition information of the configuration unit, a thickness variation function of the configuration unit in different directions is constructed; By combining the thickness variation functions in different directions, the planar distribution function of the configuration unit is obtained.

7. The method according to claim 5, characterized in that, The step of determining the target discrete point set by performing probability density screening based on multiple spatial probability density functions includes: For each of the aforementioned configuration units, arbitrarily determine multiple random discrete points; Based on the spatial probability density function of the configuration unit, the probability density of the plurality of random discrete points is filtered to determine the target discrete point; The target discrete point set is determined based on the target discrete points of multiple configuration units.

8. The method according to claim 7, characterized in that, The step of using the spatial probability density function of the configuration unit to perform probability density filtering on the plurality of random discrete points to determine the target discrete point includes: For each of the aforementioned random discrete points, the probability density value of the random discrete point is calculated based on the coordinate information of the random discrete point and the spatial probability density function; Determine whether the probability density value is greater than a random threshold; If so, the random discrete point is determined as the target discrete point.

9. A smart prediction device for sand body thickness distribution under seismic-free sparse well network conditions, characterized in that, include: The second acquisition module is used to acquire measured well point data of the area to be predicted. The generation module is used to input the measured well point data of the area to be predicted into the pre-trained sand body thickness distribution prediction model to generate a sand body thickness distribution prediction map of the area to be predicted. The first acquisition module is used to acquire multiple sets of sample well point data; The processing module is used to determine, for each group of sample well point data, a target sample image corresponding to the sample well point data from multiple sample images in a pre-generated sample image library; the sample image library includes multiple sample images, which are generated by parameterizing the depositional information of multiple configuration units; The training module is used to train the initial prediction model based on the sample well point data and the corresponding target sample image of each group, to obtain the sand body thickness distribution prediction map output by the model, and to iteratively update the model parameters through the loss function until the training termination condition is met. The loss function includes: a well point matching loss term and a structural similarity loss term; The well point matching loss term is related to the deviation between the predicted sand body thickness output by the model and the measured sand body thickness of the sample well point data input to the model. The structural similarity loss term is related to the similarity between the distribution map predicted by the model and the target sample image input to the model. The predicted sand body thickness output by the model is extracted from the predicted sand body thickness distribution map.

10. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

Images