Three-dimensional geological structure modeling method, device and system based on drilling information

Abstract

The invention provides a three-dimensional geological structure modeling method, device and system based on drilling information. The method comprises the following steps: S1, collecting drilling data, seismic exploration data and remote sensing image data of a target area; s2, constructing a neural network model composed of a simple full-connection module, a complex convolution residual module and a hierarchical integration module which are connected in sequence; s3, training the initial three-dimensional geologic model by adopting a joint loss function; and S4, in the network training process, a plurality of geological scene models are generated, and each scene model corresponds to a group of different network parameter combinations. According to the three-dimensional geologic structure modeling method, device and system based on drilling information, a spatial receptive field for feature extraction is expanded through a three-dimensional cavity convolution residual block of a complex convolution residual module according to a symmetric expansion rate, input features and deep convolution features are added, processing is carried out through a hierarchical integration module, and the three-dimensional geologic structure modeling method, device and system based on drilling information are obtained. And geologic structure correlation characteristics across drill holes can be effectively captured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of three-dimensional geological structure modeling, in particular to a three-dimensional geological structure modeling method based on drilling information, a device and a system thereof. BACKGROUND

[0002] Three-dimensional geological structure modeling is a technology for constructing a three-dimensional model of an underground geological body using drilling, seismic, remote sensing and other data, which is used to intuitively present the stratigraphic structure, lithological distribution and structural characteristics, and to provide a basis for resource exploration and engineering planning.

[0003] The existing technology mainly has two problems: first, the modeling accuracy of complex structural features in sparsely drilled areas, such as fault extension and stratigraphic pinch-out, is insufficient, traditional methods rely on single data or simple interpolation, and it is difficult to fuse the spatial correlation characteristics of multi-source data, resulting in interface distortion and structural discontinuity in the model; second, there is a lack of geological professional constraints and prediction uncertainty assessment, existing models do not embed prior knowledge such as stratigraphic occurrence and interface geometric position, which is easy to generate results that violate geological laws, and cannot quantify the prediction confidence, so the modeling reliability in deep or complex structural areas is low, therefore a three-dimensional geological structure modeling method based on drilling information and a system thereof are proposed. SUMMARY

[0004] In view of the technical problems in the background art, the present application provides a three-dimensional geological structure modeling method based on drilling information, a device and a system thereof, which effectively captures the complex structural correlation characteristics between sparse drill holes through multi-source data fusion and three-dimensional hollow convolution residual network, improves the modeling accuracy and avoids interface distortion; embeds geological prior constraints such as stratigraphic occurrence and geometric position using a joint loss function to ensure that the model conforms to geological laws; dynamically assesses the prediction confidence through a probabilistic model update to achieve the purpose of adaptive optimization of the model with new data.

[0005] In a first aspect, the embodiments of the present application provide for collecting historical geological data and real-time geological data of a target area; establishing a neural network model and training the neural network model to obtain an initial three-dimensional geological model through the historical geological data; setting a joint loss function and training the initial three-dimensional geological model through the joint loss function and the real-time geological data; wherein the setting of the joint loss function comprises constructing a geometric position constraint term and a structural occurrence constraint term; The geometric position constraint term is used to force the spatial distance between the predicted stratigraphic interface and the drilling exposure point to be within a preset error range. The structural occurrence constraint term is used to ensure that the predicted interface normal vector is consistent with the measured occurrence direction through gradient penalty.

[0006] Further, in the embodiment, constructing the geometric position constraint term comprises: constructing an implicit function based on the stratum interface points exposed by drilling to represent the stratum interface, calculating the signed distance value of the grid point to the nearest interface point by using the fast marching method, and incorporating the average value of the signed distance value into the loss function to limit the spatial deviation between the predicted stratum interface and the actual interface.

[0007] Further, in the embodiment, constructing the structure occurrence constraint term comprises: converting the measured stratum dip angle and dip direction into corresponding interface normal vector coordinate components, and generating a gradient penalty term by calculating the cosine similarity between the predicted interface normal vector and the measured normal vector, so as to force the occurrence direction of the predicted stratum interface to be consistent with the measured geological occurrence.

[0008] Further, in the embodiment, training the initial three-dimensional geological model by the joint loss function and the real-time geological data comprises: generating a plurality of geological scenario models by randomly inactivating neurons in a complex convolution residual module, inputting the real-time geological data into the plurality of geological scenario models, and calculating the posterior probability of each geological scenario model; screening the posterior probability of each geological scenario model, screening out the geological scenario model corresponding to the posterior probability higher than a set threshold, and updating the initial three-dimensional geological model according to the screened geological scenario model.

[0009] Further, in the embodiment, generating a plurality of geological scenario models by randomly inactivating neurons in a complex convolution residual module comprises: adopting a Monte Carlo random inactivation technology to randomly turn off neurons in the complex convolution residual module with a preset probability to generate a plurality of geological scenario models.

[0010] Further, in the embodiment, the randomly inactivated complex convolution residual module comprises a plurality of three-dimensional hollow convolution residual blocks, and the expansion rates of the plurality of three-dimensional hollow convolution residual blocks are symmetrically arranged in order from small to large and from large to small.

[0011] Further, in the embodiment, the historical geological data comprises drilling data, seismic exploration data and remote sensing image data.

[0012] In a second aspect, the embodiments of the present application provide a computer readable storage medium, which is a non-volatile storage medium or a non-transient storage medium, and has a computer program stored thereon, wherein the computer program is run by a processor to perform the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the embodiments.

[0013] In a third aspect, the embodiments of the present application provide a computer program product, comprising computer programs / instructions, characterized by that the computer programs / instructions, when executed by a processor, implement the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the above.

[0014] In a fourth aspect, the embodiments of the present application provide a three-dimensional geological structure modeling system based on drilling information, comprising a memory and a processor, and the memory stores a computer program capable of running on the processor, characterized by that the processor executes the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the above when running the computer program.

[0015] Beneficial effects: 1. The three-dimensional geological structure modeling method, device and system based on drilling information, which expands the spatial receptive field of feature extraction by setting a symmetric inflation rate for the three-dimensional hollow convolution residual block of the complex convolution residual module, adds the input features and deep convolution features through the jump connection, and then processes through the hierarchical integration module to realize multi-scale geological feature fusion, effectively capture the geological structure correlation features across the boreholes, and solve the problems of insufficient modeling precision, interface distortion and structure discontinuity of complex structure features in the sparse borehole area.

[0016] 2. The three-dimensional geological structure modeling method, device and system based on drilling information, which uses a joint loss function, constructs a geometric position constraint term based on a signed distance function, forces the predicted stratigraphic interface to maintain a reasonable spatial distance from the drilling exposure points, constructs a structure occurrence constraint term through gradient penalty to ensure that the predicted interface normal vector is consistent with the measured occurrence direction, and embeds geological prior knowledge; meanwhile, the probabilistic model updating process generates scenario models through random inactivation, selects and retains high-confidence models according to the posterior probability calculated based on new data, realizes dynamic updating and prediction uncertainty evaluation, and solves the problems of lack of geological professional constraints and inability to quantify the prediction confidence.

[0017] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following will specifically describe the embodiments of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings used in the present application. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor on the basis of these drawings.

[0019] Figure 1A flowchart of a three-dimensional geological structure modeling method based on drilling information is provided for the embodiments of the present application. Figure 2 A flowchart of a part of training an initial three-dimensional geological model is provided for the embodiments of the present application. Figure 3 A framework diagram of a complex convolution residual module part is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0020] The embodiments of the technical solutions of the present application will be described in detail below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, and cannot limit the protection scope of the present application.

[0021] 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 the present application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "include" and "have" and any variations thereof in the specification and claims of the present application and the above description of drawings are intended to cover non-exclusive inclusion.

[0022] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise explicitly and specifically limited.

[0023] In this paper, the term "embodiment" means that the specific features, structures or properties described in conjunction with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment to other embodiments. The skilled person in the art explicitly and implicitly understands that the embodiments described herein can be combined with other embodiments.

[0024] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are in an "or" relationship.

[0025] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two), and similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0026] In the description of the embodiments of the present application, the orientations or positional relationships indicated by the technical terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the embodiments of the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0027] In the description of the embodiments of the present application, unless otherwise explicitly specified and limited, the technical terms "mounting", "connection", "connection", "fixing" and the like should be understood broadly, for example, can be fixed connection, can also be detachable connection, or integral; can be mechanical connection, can also be electrical connection; can be directly connected, can also be indirectly connected through an intermediate medium, can be the internal communication of two elements or the interaction relationship of two elements. For those skilled in the art, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0028] Three-dimensional geological structure modeling is a technology for constructing a three-dimensional model of an underground geological body using drilling, seismic, remote sensing and other data, which is used to intuitively present the structure of strata, the distribution of lithology and the characteristics of structure, and to provide a basis for resource exploration, engineering planning and the like.

[0029] The existing technology mainly has two problems: one is that the modeling precision of complex structure characteristics in sparsely drilled areas, such as fault extension and strata pinch-out, is insufficient, and the traditional method relies on single data or simple interpolation, and it is difficult to fuse the spatial correlation characteristics of multi-source data, resulting in problems such as interface distortion and structure discontinuity in the model; the second is the lack of geological professional constraints and prediction uncertainty evaluation, the existing model does not embed prior knowledge such as strata occurrence and interface geometric position, and is easy to generate results that violate geological laws, and cannot quantify the prediction confidence, and the modeling reliability in deep or complex structure areas is low, and therefore a three-dimensional geological structure modeling method and system based on drilling information are proposed.

[0030] In view of the deficiencies of the prior art, the present application provides a three-dimensional geological structure modeling method and system based on drilling information, which effectively captures the complex structure correlation characteristics between sparse drill holes by multi-source data fusion and three-dimensional hollow convolution residual network, improves the modeling accuracy and avoids interface distortion; geological prior constraints such as stratigraphic occurrence and geometric position are embedded by using a joint loss function to ensure that the model conforms to geological laws; the prediction reliability is dynamically evaluated by a probabilistic model update to realize adaptive optimization of the model with new data, solving the problems of the prior art, such as insufficient modeling accuracy of complex structure characteristics in the sparse area of drill holes, such as fault extension and stratigraphic pinch-out, the traditional method relies on single data or simple interpolation, it is difficult to fuse the spatial correlation characteristics of multi-source data, resulting in interface distortion, structure discontinuity and other problems of the model, lack of geological professional constraints and prediction uncertainty evaluation, the existing model does not embed prior knowledge such as stratigraphic occurrence and interface geometric position, which is easy to generate results that violate geological laws, and cannot quantify the prediction reliability, and the modeling reliability is low in deep or complex structure area.

[0031] Please refer to Figure 1 , Figure 1 A flowchart of a three-dimensional geological structure modeling method based on drilling information is provided for the embodiments of the present application.

[0032] In a first aspect, as shown in Figure 1 The present application provides a three-dimensional geological structure modeling method based on drilling information, comprising the following steps: S1, collecting historical geological data and real-time geological data of a target area; S2, establishing a neural network model and training the neural network model to obtain an initial three-dimensional geological model by using the historical geological data; S3, setting a loss function and updating the initial three-dimensional geological model by using the real-time geological data.

[0033] Specifically, in the present embodiment, the historical geological data and real-time geological data of the target area are collected by the data acquisition and preprocessing component, and the historical geological data and real-time geological data of the target area are processed to output a standardized multi-source geological data set. A neural network model is established, the processed standardized multi-source geological data set is input into the neural network model, and the neural network model is trained to obtain an initial three-dimensional geological model. A loss function is set, the real-time geological data is input into the initial three-dimensional geological model, the initial three-dimensional geological model is trained by using the real-time geological data, the dynamic and probabilistic update of the three-dimensional geological model is realized, the three-dimensional geological model is automatically adjusted after the input of new drilling / geophysical data, the real underground structure is continuously approximated, and the personnel can more accurately analyze the geological conditions of the target area.

[0034] As Figure 1As shown, the application provides a three-dimensional geological structure modeling method based on drilling information, and the specific steps are as follows: S1, collecting historical geological data and real-time geological data of the target area; In this embodiment, drilling data, seismic exploration data and remote sensing image data of the target area are collected, the drilling data is cleaned to remove abnormal values and repeated data, and standardized processing is performed to generate a standard data set containing borehole coordinates, lithology stratification, stratum thickness and geophysical parameters.

[0035] Specifically, first, multi-source data acquisition and preprocessing are performed; borehole data of the target area is obtained through high-precision drilling equipment, each hole records accurate three-dimensional coordinates of the hole mouth, lithology stratification such as sandstone, shale, limestone, etc., 20 types are divided according to actual exploration, stratum thickness and geophysical parameters such as resistivity and density; three-dimensional seismic array is used for seismic exploration, covering the entire area, and the underground physical property data volume is obtained through pre-stack depth migration processing inversion, containing parameters such as P-wave velocity and S-wave velocity, with a resolution of 10 meters; remote sensing data is collected by unmanned aerial vehicle carrying multi-spectral camera, and features such as surface stratum outcrop boundary and structural line trend are obtained, and key geological information is extracted through image recognition technology.

[0036] When cleaning the drilling data, first, abnormal values are identified through statistical analysis, for example, there are obvious unreasonable maximum or minimum values in the stratum thickness, which are removed through manual review; repeated data is compared with the hole mouth coordinates and lithology stratification information, and is quickly checked and deleted by using hash table; during the standardization processing, the continuous variables such as borehole coordinates and stratum thickness are scaled to the interval of 0-1 according to the proportion, and the lithology stratification adopts one-hot encoding, for example, sandstone corresponds to only one position of 1 in the 20-dimensional vector, and the rest are 0, and finally a standard data set containing 128-dimensional features is generated, including 3-dimensional coordinates, 20-dimensional lithology encoding, 10-dimensional geophysical parameters and derived features; the seismic and remote sensing data are unified to the local engineering coordinate system through the coordinate conversion module, to ensure the spatial alignment of multi-source data.

[0037] In this embodiment, the drilling data provides the vertical absolute reference, the lithology classification (20 types) and the physical property parameters constitute the geological entity "fingerprint" library, the three-dimensional seismic realizes the horizontal continuous coverage, the P / S wave velocity inversion reveals the spatial variation law of the deep structure, the unmanned aerial vehicle remote sensing constrains the shallow geological geometric topological relationship through the structural line trend and the outcrop boundary, and the three-dimensional coordinates (geometric positioning) + lithology encoding (material composition) + physical property parameters (physical response) constitute the geological entity triple, so that the subsequent deep learning model can learn the spatial structure law, the lithology combination mode and the physical property correlation characteristics at the same time.

[0038] S2, establishing a neural network model and training the neural network model through historical geological data to obtain an initial three-dimensional geological model; In the embodiment, a neural network model composed of a simple full connection module, a complex convolution residual module and a hierarchical integration module connected in sequence is constructed; the standard data set after spatial registration, the subsurface physical property data volume obtained by seismic inversion and the surface geological features extracted by remote sensing are input into the neural network, multi-scale geological feature fusion and prediction are realized through processing of each module, and an initial three-dimensional geological model is output.

[0039] Specifically, the neural network model is constructed, the simple full connection module receives 128-dimensional input features, the first full connection layer is provided with 256 neurons, the weight matrix is initialized as a small random value to avoid saturation of the activation function, a Softplus activation function is used to smoothly convert negative values to positive values, and weak features in the data are retained, for example, nonlinear amplification of low resistivity shale features enhances the model's ability to identify lithological boundaries; the second full connection layer outputs a 128-dimensional basic geological feature vector, which fuses preliminary correlation information of lithology and spatial position, such as high-density limestone corresponding to high resistivity, and the trend of spatial position and formation thickness, such as the change rule of formation thickness with increasing depth, to provide more representative input for the subsequent convolution module, for example, when processing two drill holes far apart, the implicit relationship between lithological combination and spatial distance can be effectively extracted, and the model's prediction ability for stratigraphic pinch-out phenomena is improved. After the establishment of the neural network model, the standard data set after spatial registration, the subsurface physical property data volume obtained by seismic inversion and the surface geological features extracted by remote sensing are input into the neural network, multi-scale geological feature fusion and prediction are realized through processing of each module, and an initial three-dimensional geological model is output.

[0040] In the embodiment, as shown in Figure 3 The complex convolution residual module includes a plurality of three-dimensional hollow convolution residual blocks, i.e., three-dimensional hollow convolution residual blocks 1, …, n, and the dilation rates of the plurality of three-dimensional hollow convolution residual blocks are symmetrically arranged in order from small to large and from large to small.

[0041] Specifically, the complex convolution residual module is composed of seven three-dimensional hollow convolution residual blocks connected in series, and the dilation rates are arranged in the order of 1, 2, 4, 8, 4, 2 and 1 to form a symmetric expansion and contraction structure; among them, the first four blocks gradually expand the receptive field to capture long-distance structural trends across drill holes, such as fault extension direction, and the last three blocks shrink and focus on local features, such as formation dip change; each residual block first performs hollow convolution, and then adds the input features and the convolution result through jump connection to ensure the fusion of shallow position information and deep structural features, for example, directly superimposing drill hole coordinate features and fault occurrence features to avoid loss of deep network information; the hierarchical integration module performs global average pooling and multi-layer full connection processing on the multi-scale features output by the complex convolution residual module, and finally outputs the formation interface depth, the 20-class lithology classification probability and the structural occurrence, i.e., the dip angle and the trend.

[0042] Specifically, in the complex convolutional residual module, the first residual block has an expansion rate of 1, which is a standard 3×3×3 convolution, capturing local features within a 10-meter radius of the borehole, such as the internal homogeneity of a single lithology. The second block has an expansion rate of 2, with the convolution kernel sampling at 2 grid points, expanding the receptive field to 20 meters, capturing small-scale structural relationships between adjacent boreholes, such as the lateral extension of thin sandstone layers. This pattern continues until the fourth block, with an expansion rate of 8, reaches a receptive field of 80 meters, capable of capturing the fault strike or the overall trend of large folds between distant boreholes. The internal operation process of each residual block is as follows: the input features are first extracted through a hollow convolutional layer, then the data distribution is adjusted through a batch normalization layer to avoid gradient vanishing, then nonlinearity is introduced through the ReLU activation function, and finally, it is added to the original input features via skip connections. If the dimensions are inconsistent, the number of channels is adjusted through a 1×1 convolution. For example, if the input features are 128-dimensional and the convolution output is 256-dimensional, the original input is first subjected to a 1×1 convolution to increase the dimensionality, and then added to the convolution result to ensure the effectiveness of feature fusion.

[0043] This design significantly expands the feature extraction range without substantially increasing computational load, making it particularly suitable for sparse borehole regions. For example, in a certain exploration area, the average borehole spacing is 50 meters. Traditional convolution cannot capture cross-hole features, but this module effectively identifies small faults between two boreholes through convolution operations with an expansion rate of 8, increasing the fault modeling accuracy from 60% to 85%. The skip connection mechanism ensures that shallow location information, such as borehole coordinates, is directly integrated with deep structural features, such as fault attitude, avoiding feature loss due to the increase in the number of layers in the deep network. For example, when dealing with overturned folds, it can simultaneously preserve the spatial location of the strata and the complex dip direction, improving the model's ability to model geological interfaces with multiple Z values.

[0044] S3. Set the loss function and train and update the initial three-dimensional geological model using real-time geological data.

[0045] In this embodiment, as Figure 2 As shown, a joint loss function is used to train the initial 3D geological model. During network training, multiple geological scenario models are generated through neurons in the randomly deactivated complex convolutional residual module. Each scenario model corresponds to a different set of network parameter combinations. When real-time geological data (i.e., new drilling data or geophysical data) is acquired, the posterior probability of each scenario model is calculated. The posterior probabilities of each geological scenario model are screened, and geological scenario models with posterior probabilities higher than a set threshold are selected, while low-probability models are eliminated. The initial 3D geological model is then dynamically updated based on the selected geological scenario models.

[0046] In embodiments, the joint loss function comprises a geometric position constraint term and a structural attitude constraint term, wherein the geometric position constraint term is constructed based on a signed distance function to enforce the spatial distance between the predicted stratigraphic interface and the drilling exposure points within a preset error range; the structural attitude constraint term is used to ensure the consistency between the predicted interface normal vector and the measured attitude direction by gradient penalty.

[0047] Specifically, in the present embodiment, the geometric position constraint term: constructs an implicit function to represent the stratigraphic interface based on the drilling-exposed stratigraphic interface points, calculates the signed distance value of the grid point to the nearest interface point using the fast marching method, and incorporates the average value of the distance into the loss function to limit the spatial deviation between the predicted stratigraphic interface and the actual interface. Directly using the most reliable drilling exposure point data, the discrete point constraint is converted into a constraint on the continuous geological interface through the signed distance function. This loss term forces the stratigraphic interface predicted by the neural network (usually the zero level surface of the signed distance function) to pass through or be extremely close to these known drilling points, ensuring the absolute accuracy of the model at known control points.

[0048] Specifically, in the present embodiment, the geometric position constraint term: constructs an implicit function to represent the stratigraphic interface based on the drilling-exposed stratigraphic interface points, calculates the signed distance value of the grid point to the nearest interface point using the fast marching method, and incorporates the average value of the distance into the loss function to limit the spatial deviation between the predicted stratigraphic interface and the actual interface. Directly using the most reliable drilling exposure point data, the discrete point constraint is converted into a constraint on the continuous geological interface through the signed distance function. This loss term forces the stratigraphic interface predicted by the neural network (usually the zero level surface of the signed distance function) to pass through or be extremely close to these known drilling points, ensuring the absolute accuracy of the model at known control points.

[0049] The structural attitude constraint term: converts the measured attitude data (i.e. stratigraphic dip and trend) into corresponding interface normal vector coordinates, and generates a gradient penalty term by calculating the cosine similarity between the predicted interface normal vector and the measured normal vector, thereby forcing the attitude direction of the predicted stratigraphic interface to be consistent with the measured geological attitude. This loss term acts on the gradient of the predicted implicit field (signed distance function), and mathematically, the gradient of the signed distance function at its zero level surface (i.e. the predicted interface) is the normal vector direction of the interface. This constraint term penalizes the difference between the predicted interface normal vector and the measured attitude direction (calculated from the trend dip angle), forcing the model to predict the local geometric shape (inclination direction) of the geological interface at the attitude measurement point to be consistent with the measured value.

[0050] For example, the structural occurrence constraint is directed to the dip and trend of the stratum, for example, the measured dip of a stratum is 45 degrees, and the trend is 90 degrees, that is, the east direction, the normal vector is calculated by the trigonometric function, and 0, 1 / 2, and 1 / 2 are obtained. The normal vector of the predicted interface is obtained by calculating the interface gradient, and the cosine similarity of the two needs to be higher than 0.98, and the corresponding angle deviation is not more than 5 degrees, otherwise a gradient penalty term is generated to force the model to adjust the predicted occurrence; in the fold development area, the constraint can effectively avoid the model to generate an occurrence that violates the geological law, for example, to prevent the dip of the inverted fold from being predicted in the reverse direction, so that the prediction accuracy of the occurrence is improved by 30%, and the stratum inclination direction in the three-dimensional model constructed is ensured to be consistent with the actual geological Figure 1 direction.

[0051] In the embodiment, the joint loss function (geometric position constraint + structural occurrence constraint) combines two different types of but crucial geological observation data (point position, surface direction) into a single optimization target. The network needs to minimize the two loss terms during the training process, not only to fit the spatial position of the borehole point, but also to respect the structural form law of the stratum during the fitting process, and force the model to meet the requirements of accurate spatial position and local structural geometry at the same time. Significantly improve the geological rationality, prediction accuracy (especially in sparse data area), generalization ability of the model, and finally guarantee the effectiveness of the scenario model screening in the dynamic updating process, so that the final geological model updating result is more reliable and more consistent with the actual geology.

[0052] In the embodiment, the neurons of the complex convolution residual module adopt the Monte Carlo random inactivation technology, and the neurons in the complex convolution residual module are randomly turned off at a preset probability, for example, a certain neuron is temporarily turned off at a probability of 50%, and the neuron does not participate in calculation in this iteration, and an independent scenario model is generated every time the forward propagation is performed, and 100 models with different parameter combinations are accumulated. The parameters of each model, such as the convolution kernel weight and the full connection layer bias, are stored separately to form a model library for subsequent evaluation of new data.

[0053] When new drilling data is acquired, such as 10 new drill holes, revealing pinch-out of a certain formation, the fitting loss of each scenario model to the new data is calculated one by one, and the smaller the loss, the stronger the model's ability to explain the new data; based on Bayes' theorem, assuming that the prior probabilities of all models are the same, the posterior probability is proportional to the likelihood function and inversely proportional to the loss, and the models with a posterior probability higher than 70% are screened out, and usually the top 20 high-confidence models are retained; after eliminating low-probability models, the parameters of the retained models are weighted and averaged, and the weight is the respective posterior probability, to generate updated model parameters; for example, after new data is added in a certain fault area, the prediction error of the original model for the fault pinch-out position is 10 meters, and after updating, the error is reduced to 4 meters by fusing the parameters of 3 high-probability models, and the model's ability to describe the details of the local structure is significantly improved; this mechanism supports the model to quickly integrate new data information without retraining, and is especially suitable for scenarios where exploration data is gradually enriched, ensuring that the model always remains adaptable to the latest geological understanding.

[0054] In a second aspect, the embodiments of the present application provide a computer readable storage medium, which is a non-volatile storage medium or a non-transitory storage medium, and has stored thereon a computer program, which, when executed by a processor, performs the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the above aspects.

[0055] In a third aspect, the embodiments of the present application provide a computer program product, which includes computer programs / instructions, and when executed by a processor, implements the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the above aspects.

[0056] In a fourth aspect, the embodiments of the present application provide a three-dimensional geological structure modeling system based on drilling information, which includes a memory and a processor, and the memory has stored thereon a computer program executable on the processor, and when the processor executes the computer program, performs the steps of the three-dimensional geological structure modeling method based on drilling information according to any one of the above aspects.

[0057] It should be noted that the present application is not limited to the above-mentioned embodiments. The above-mentioned embodiments are only examples, and embodiments having the same technical idea and playing the same role and effect within the scope of the technical solution of the present application are all included in the technical scope of the present application. In addition, within the scope of the main idea of the present application, various modifications of the embodiments, combination of part of the constituent elements in the embodiments to construct other ways can also be included in the scope of the present application.

Claims

1. A three-dimensional geological structure modeling method based on drilling information, characterized in that, Includes the following steps: Collect historical and real-time geological data of the target area; A neural network model is established, and an initial three-dimensional geological model is obtained by training the neural network model using the historical geological data. A joint loss function is constructed, and the initial three-dimensional geological model is trained using the joint loss function and the real-time geological data; The construction of the joint loss function includes: constructing a geometric position constraint term and constructing an attitude constraint term; The geometric position constraint term is used to force the predicted spatial distance between the formation interface and the drilling exposure point to be within a preset error range. The constructed attitude constraint term is used to ensure, through gradient penalty, that the predicted interface normal vector is consistent with the measured attitude direction.

2. The three-dimensional geological structure modeling method based on drilling information according to claim 1, characterized in that, The construction of the geometric position constraint term includes: constructing an implicit function based on the formation interface points exposed by drilling to represent the formation interface; using the fast traverse method to calculate the signed distance value from the grid point to the nearest interface point; and incorporating the average value of the signed distance value into the loss function to limit the spatial deviation between the predicted formation interface and the actual interface.

3. The three-dimensional geological structure modeling method based on drilling information according to claim 1, characterized in that, The construction of the structural attitude constraint term includes: converting the measured stratigraphic dip angle and dip direction into the corresponding interface normal vector coordinate components, and generating a gradient penalty term by calculating the cosine similarity between the predicted interface normal vector and the measured normal vector, thereby forcing the predicted stratigraphic interface attitude direction to be consistent with the measured geological attitude.

4. The three-dimensional geological structure modeling method based on drilling information according to claim 1, characterized in that, The training of the initial three-dimensional geological model using the joint loss function and the real-time geological data includes: The initial three-dimensional geological model is trained using the joint loss function. During the training of the initial three-dimensional geological model, multiple geological scenario models are generated by neurons in the randomly deactivated complex convolutional residual module. The real-time geological data is input into the multiple geological scenario models, and the posterior probability of each geological scenario model is calculated. The posterior probabilities of each geological scenario model are filtered to select the geological scenario models whose posterior probabilities are higher than a set threshold, and the initial three-dimensional geological model is updated based on the selected geological scenario models.

5. The three-dimensional geological structure modeling method based on drilling information according to claim 4, characterized in that, The generation of multiple geological scenario models through neurons in a randomly deactivated complex convolutional residual module includes: The Monte Carlo random deactivation technique is used to randomly shut down neurons in complex convolutional residual modules with a preset probability to generate multiple geological scenario models.

6. The three-dimensional geological structure modeling method based on drilling information according to claim 1, characterized in that, The random deactivated complex convolution residual module includes multiple three-dimensional dilated convolution residual blocks, and the dilation rates of the multiple three-dimensional dilated convolution residual blocks are symmetrically arranged in order from small to large and from large to small.

7. The three-dimensional geological structure modeling method based on drilling information according to claim 1, characterized in that, The historical geological data includes drilling data, seismic exploration data, and remote sensing image data.

8. A computer-readable storage medium, wherein the computer-readable storage medium is a non-volatile storage medium or a non-transient storage medium, and a computer program is stored thereon, characterized in that, When the computer program is run by the processor, it executes the steps of the three-dimensional geological structure modeling method based on drilling information as described in any one of claims 1 to 7.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the three-dimensional geological structure modeling method based on drilling information as described in any one of claims 1 to 7.

10. A three-dimensional geological structure modeling system based on drilling information, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the three-dimensional geological structure modeling method based on drilling information as described in any one of claims 1 to 7.

Images