A method, device, equipment and storage medium for parallel explicit sampling of geological modeling

Abstract

The present application provides a method, device, equipment and storage medium for parallel explicit sampling of geological modeling, the method comprising: collecting geological exploration drilling data to construct a point set, and training a fully connected neural network to obtain a stratigraphic classification model. Set the octree structure parameters, create a structure with a point set, and obtain stratigraphic attributes through the model. Recursively split the node, if the depth has not reached the maximum and the attributes are inconsistent, then split; if the maximum depth is reached or the attributes are consistent, then output. Until the node stack is empty, aggregate the sub-node data to generate three-dimensional visualization data. Based on traditional three-dimensional geological modeling, this method introduces adaptive octrees and parallel explicit sampling, solves the problem of insufficient efficiency of traditional grid sampling, and provides new ideas and methods for geological modeling.

Description

Technical Field

[0001] The present application relates to the field of geological modeling, and in particular to a method, device, equipment and storage medium for parallel explicit sampling of geological modeling. Background Art

[0002] In the field of three-dimensional geological modeling, geological modeling technology based on deep learning has gradually been developed and applied. Deep learning geological modeling technology often learns and fits the natural mapping relationship of underground three-dimensional strata distribution through neural networks, and finally obtains a space-based stratum classifier or a stratum attribute value regressor containing geological surface symbol distance information, which is used to determine the corresponding stratum conditions at each spatial point. To obtain a three-dimensional geological model product, it is necessary to carry out corresponding three-dimensional explicit sampling work based on the deep learning neural network.

[0003] The current mainstream explicit sampling scheme is to use a three-dimensional regular grid model at a given scale to perform sequential inference sampling on the deep learning stratigraphic classifier to obtain the stratigraphic attribute information corresponding to each grid space, or to use the moving tetrahedron / hexahedron method to perform explicit sampling of the surface shape on the deep learning geological surface regressor. However, the commonly used three-dimensional regular grid sampling method at this stage often consumes a lot of computing and storage resources in the process of explicit sampling modeling of the stratigraphic attribute classifier based on the neural network, and the regular grid model has poor overall smoothness compared to the three-dimensional vector model. When the grid size is not fine enough, it is difficult to accurately depict the surface morphology of the geological interface. In addition, after increasing the refinement of the grid size, the number of triangular facets of the model and the model storage volume will also increase exponentially. With limited computing resources, it is difficult to smoothly render large-scale, high-precision three-dimensional geological grid models. Summary of the invention

[0004] The purpose of the present application is to overcome the above-mentioned prior art and provide a geological modeling parallel explicit sampling method, device, equipment and storage medium.

[0005] The present application provides a parallel explicit sampling method for geological modeling, comprising:

[0006] Collecting stratum boundary spatial position points and stratum internal points with layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points with layer thickness information into a point set;

[0007] Using the point set to train a fully connected neural network to generate a formation classification model;

[0008] Setting a maximum depth and a minimum depth of an octree structure, and creating the octree structure according to the point set, wherein the octree structure includes a parent node and a child node;

[0009] Acquire the first stratum attribute of the parent node and the second stratum attribute of the center coordinates of the child node through the stratum classification model;

[0010] Push the parent node and child nodes of the octree structure into the node stack, and judge: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, then split the top node of the node stack into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, then output the top node of the node stack;

[0011] Recursively execute the previous step until the number of nodes in the node stack is 0;

[0012] Summarize the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0013] Optionally, collecting the stratum boundary spatial position points and the stratum internal points of the layer thickness information in the geological exploration borehole includes:

[0014] The original data of geological exploration boreholes are upsampled between borehole layers, the upsampling data interval number is set, and the internal points of the formation are sampled at intervals based on the spatial position points of the formation boundary in the borehole and the layer thickness information.

[0015] Optionally, the point set is preprocessed and then a fully connected neural network is trained to generate a stratum classification model, including:

[0016] Performing forward normalization processing on the spatial coordinate data in the point set;

[0017] Select an appropriate fully connected neural network architecture and randomly divide the normalized point set into training set, test set and validation set according to the preset ratio;

[0018] The Adam optimizer is used to train the neural network on the training set, and the cross entropy loss is used as the loss function. After the training loss and accuracy converge, all parameters of the neural network model are saved.

[0019] Optionally, setting a maximum depth and a minimum depth of an octree structure, and creating the octree structure according to the point set includes:

[0020] Based on the minimum and maximum values ​​of the preprocessed point set in three directions in the spatial rectangular coordinate system, the spatial positions of the eight vertices of the octree parent node are determined, and the spatial position of the octree parent node is the rectangular bounding box space of the point set.

[0021] Optionally, it also includes:

[0022] Determine the maximum number of processes that can be opened on the computer;

[0023] Calculate an initial octree split depth based on the number of processes;

[0024] Build a process pool based on the maximum number of processes, and evenly distribute the octree leaf nodes obtained by splitting to each process;

[0025] After each process completes sampling its child nodes, it sends the results back to the main process.

[0026] Optionally, summarizing the stratigraphic attributes and center coordinate data of all child nodes includes:

[0027] In the main process, the sub-node stratigraphic attributes and center coordinate data returned by all processes are summarized.

[0028] Optionally, generating three-dimensional visualization data includes:

[0029] Based on the summarized sub-node stratum attributes and center coordinate data, calculate each node attribute information, including: the vertex coordinates of the six surfaces of each node, the triangle face point set index, and texture information;

[0030] The attribute information is uniformly written into a universal three-dimensional visualization exchange format file to construct a three-dimensional geological model.

[0031] The present application also provides a geological modeling parallel explicit sampling device, comprising:

[0032] A collection module collects the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information in the geological exploration borehole, and constructs the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information into a point set;

[0033] A training module, using the point set to train a fully connected neural network to generate a formation classification model;

[0034] A setting module, which sets the maximum depth and the minimum depth of the octree structure, and creates the octree structure according to the point set, wherein the octree structure includes a parent node and a child node;

[0035] A prediction module, which obtains a first stratum attribute of the parent node and a second stratum attribute of the center coordinates of the child node through the stratum classification model;

[0036] A judgment module pushes the parent node and the child node of the octree structure into the node stack, and judges: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, the top node of the node stack is split into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, the top node of the node stack is output;

[0037] A recursive module recursively executes the previous step until the number of nodes in the node stack is 0;

[0038] The summary module summarizes the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0039] The present application also provides a geological modeling parallel explicit sampling device, comprising:

[0040] Memory;

[0041] A processor is used to retrieve a computer executable program of the above-mentioned geological modeling parallel explicit sampling method from the memory, and execute: collecting stratum boundary spatial position points and stratum internal points of layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points of layer thickness information into a point set; using the point set to train a fully connected neural network to generate a stratum classification model; setting the maximum depth and minimum depth of the octree structure, and creating the octree structure according to the point set, the octree structure including a parent node and a child node; obtaining the first stratum attribute of the parent node and the center of the child node through the stratum classification model The second stratigraphic attribute of the coordinate; the parent node and child nodes of the octree structure are pushed into the node stack, and it is judged that: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratigraphic attribute is inconsistent with the first stratigraphic attribute, the top node of the node stack is split into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratigraphic attribute is consistent with the first stratigraphic attribute, the top node of the node stack is output; recursively execute the previous step until the number of nodes in the node stack is 0; summarize the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0042] The present application also provides a storage medium, including: a computer executable program stored therein, wherein the computer executable program is used to be called by a processor to execute the steps of the above-mentioned geological modeling parallel explicit sampling method.

[0043] The beneficial effects of this application are:

[0044] The present application provides a parallel explicit sampling method for geological modeling, including: collecting stratum boundary spatial position points and stratum internal points with layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points with layer thickness information into a point set; using the point set to train a fully connected neural network to generate a stratum classification model; setting a maximum depth and a minimum depth of an octree structure, and creating the octree structure according to the point set, wherein the octree structure includes a parent node and a child node; obtaining a first stratum attribute of the parent node and a second stratum attribute of the center coordinates of the child node through the stratum classification model; The parent node and child node of the octree structure are pushed into the node stack, and it is judged that: if the depth of the top node of the node stack corresponding to the parent node has not reached the maximum depth, and the second stratigraphic attribute is inconsistent with the first stratigraphic attribute, the top node of the node stack is split into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratigraphic attribute is consistent with the first stratigraphic attribute, the top node of the node stack is output; recursively execute the previous step until the number of nodes in the node stack is 0; summarize the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data. Based on the traditional three-dimensional geological modeling technology, this application introduces an adaptive octree structure and a parallel explicit sampling method. It not only solves the shortcomings of the traditional grid sampling method in terms of efficiency, but also provides new ideas and methods for the development of the field of geological modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic diagram of the parallel explicit sampling process for geological modeling in this application. DETAILED DESCRIPTION

[0046] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, the embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0047] Please refer to Figure 1 As shown, the present application provides a parallel explicit sampling method for geological modeling, comprising:

[0048] S101, collecting stratum boundary spatial position points and stratum internal points with layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points with layer thickness information into a point set.

[0049] Using geological exploration drilling data, a fully connected neural network model for the classification of stratum properties in the drilling study area is trained and constructed through deep learning technology. The fully connected neural network model predicts and classifies stratum properties, providing key data support for subsequent geological modeling.

[0050] First, the original data of geological exploration boreholes are collected. The original data includes the location, depth, and stratum distribution information of the boreholes, which are important basic information for geological modeling.

[0051] Then, the raw data is preprocessed, including data cleaning, format conversion, etc., to ensure the accuracy and consistency of the data.

[0052] Next, we perform upsampling between drilling layers. Upsampling is a data augmentation technique that improves the generalization ability of the model by increasing the number of data points.

[0053] In this application, the number of upsampled data intervals is set to n, and then the internal points of the formation are sampled at intervals based on the spatial position points of the formation boundary in the borehole and the layer thickness information. In this way, the density of data points is increased while maintaining the distribution characteristics of the formation, providing richer data support for subsequent training.

[0054] Finally, the original borehole spatial point data and the upsampled spatial points are constructed into a point set p. The point set p includes not only the three-dimensional spatial coordinates of the point, but also the formation attribute information corresponding to the point. The point set p will be used as the input data of the fully connected neural network model to train the model and predict the formation attributes.

[0055] The specific operation process of constructing a point set p containing rich stratum attribute information is as follows:

[0056] First, determine the number of upsampled data intervals n. The selection of the number of intervals n needs to be determined based on the actual situation of geological exploration drilling data and modeling requirements to ensure that the upsampled data points can fully reflect the distribution characteristics of the strata.

[0057] Then, based on the spatial position points of the formation boundaries in the borehole and the layer thickness information, the internal points of the formation are sampled at intervals. This step is achieved through an interpolation algorithm, which can generate a series of new data points inside the formation based on the formation boundaries and layer thickness information to increase the density of data points.

[0058] Finally, the original borehole spatial point data and the upsampled spatial points are constructed into a point set p. The point set p not only contains the three-dimensional spatial coordinates of the point, but also contains the stratum attribute information corresponding to the point. The information is the key data required for the subsequent training of the fully connected neural network model.

[0059] S102: Using the point set to train a fully connected neural network to generate a formation classification model.

[0060] Determine the range of the point set p in the spatial rectangular coordinate system. Specifically, calculate the minimum value (Xmin, Ymin, Zmin) and maximum value (Xmax, Ymax, Zmax) of the point set p in the three directions of X, Y, and Z. The values ​​define the minimum rectangular bounding box that contains all the point sets p.

[0061] Next, in order to convert the spatial coordinate data X, Y, Z in the point set p into a format more suitable for neural network processing, the coordinates are forward normalized.

[0062] Normalization is a common data preprocessing technique that can convert data of different magnitudes to the same magnitude, thereby speeding up the training of neural networks and improving the accuracy of the model. The normalization formula is shown in formula (1-3):

[0063]

[0064] In this step, an appropriate fully connected neural network architecture is selected. Since the goal is to predict formation properties, the input vector length of the neural network should be set to 3 (corresponding to the spatial coordinates in the X, Y, and Z directions), and the output vector length should be set to 1 (corresponding to the predicted value of the formation property).

[0065] Next, the normalized point set p is randomly divided into training set, test set and validation set in a ratio of 8:1:1. This is to be able to evaluate the performance of the model during the training process and ensure that the model has good generalization ability.

[0066] Then, the neural network is trained on the training set using the Adam optimizer. The Adam optimizer is an optimization algorithm based on gradient descent, which has the feature of adaptive learning rate adjustment and can train the model more efficiently.

[0067] During the training process, cross entropy loss (CrossEntropyLoss) is used as the loss function to measure the difference between the model's predicted value and the actual value. Cross entropy loss is one of the commonly used loss functions in classification problems, and it can handle multi-classification problems well.

[0068] Finally, when the training loss and accuracy converge, all parameters of the neural network model are saved to a local file for subsequent use. The parameters include key information such as the weights and biases of the neural network, which determine the predictive ability of the model.

[0069] S103. Set a maximum depth and a minimum depth of an octree structure, and create the octree structure according to the point set, where the octree structure includes a parent node and child nodes.

[0070] Initialize the adaptive octree grid field and establish a basic octree structure to prepare for subsequent recursive splitting and sampling.

[0071] First, set the maximum depth Dmax and minimum depth Dmin of the octree.

[0072] These two parameters are crucial for controlling the refinement of the octree. The maximum depth Dmax determines the deepest level that the octree can split to, thus affecting the minimum sampling unit size of the model. The minimum depth Dmin limits the starting level of the octree split, preventing unnecessary over-splitting. By setting these two parameters reasonably, the accuracy of the model and the consumption of computing resources can be balanced according to actual needs.

[0073] Next, based on the calculated minimum and maximum values ​​(Xmin, Ymin, Zmin and Xmax, Ymax, Zmax) of the point set p in the spatial rectangular coordinate system, the spatial positions of the eight vertices of the octree parent node are determined. The vertices will form a rectangular bounding box space surrounding the point set p, which is the initial parent node space of the octree. The above steps are the basis for establishing the octree structure, which ensures that subsequent splitting and sampling can be performed within the correct spatial range.

[0074] S104, obtaining a first stratum attribute of the parent node and a second stratum attribute of the center coordinates of the child node through the stratum classification model;

[0075] The stratum attribute of the parent node of the octree is predicted using the trained fully connected neural network stratum classification model. The step is to assign a stratum attribute label (stratum attribute) to each node of the octree so that in the subsequent splitting and sampling process, the sampling accuracy can be dynamically adjusted according to the difference in stratum attributes. Through this step of prediction, we can preliminarily understand the stratum distribution in the area where the point set p is located, providing basic information for subsequent adaptive sampling.

[0076] Create a node stack S to manage the nodes of the octree.

[0077] A stack is a last-in, first-out (LIFO) data structure, which is very suitable for handling this recursive splitting situation. The initial octree node (i.e., the root node) is pushed into the stack S as the starting point of the recursive splitting.

[0078] The top node of the stack is pre-split into an octree, and the spatial coordinates of the eight child nodes after the pre-split are normalized to predict the formation attributes.

[0079] Next, the top node (i.e., the node currently to be processed) is taken out from the stack S and the octree is pre-split. Pre-splitting means temporarily dividing the current node into eight child nodes.

[0080] For the eight child nodes after pre-splitting, their spatial coordinates are calculated and normalized using the above-mentioned spatial coordinate normalization method. Normalization is an important step to ensure that the neural network input data is within the same scale range, which helps to improve the prediction accuracy of the neural network.

[0081] Then, the fully connected neural network model trained in the process is used to predict the stratigraphic properties of the center coordinates of the eight sub-nodes. This step is the core application of deep learning in geological modeling. It uses the generalization ability of neural networks to predict the stratigraphic properties of unknown areas based on the stratigraphic distribution laws learned from geological exploration drilling data.

[0082] After completing the formation attribute prediction, it will be decided whether to actually split the current node into eight child nodes according to the splitting conditions (such as the node depth does not reach the maximum depth Dmax and the child node label (formation attribute) is inconsistent with the parent node label (formation attribute)). If the splitting conditions are met, the child node is pushed into the stack S for subsequent processing; if the splitting conditions are not met, the node is retained as a leaf node, and its formation attribute and spatial coordinate information are output.

[0083] Through this step, adaptive recursive split sampling of octree nodes is realized, and the sampling accuracy is dynamically adjusted according to the actual situation of the geological interface, thereby effectively reducing the consumption of computing and storage resources while ensuring the accuracy and detail of the model.

[0084] S105. Push the parent node and child nodes of the octree structure into the node stack, and judge: if the depth of the top node of the node stack corresponding to the parent node has not reached the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, then split the top node of the node stack into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node has reached the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, then output the top node of the node stack; recursively execute the previous step until the number of nodes in the node stack is 0.

[0085] In the process of adaptive recursive split sampling of octree nodes, it is necessary to constantly judge whether the top node of the stack meets the conditions for further splitting. This judgment is based on two key factors:

[0086] Node depth: First, check whether the depth of the top node of the stack has reached the preset maximum depth Dmax. If it has reached or exceeded Dmax, it means that the node is deep enough and does not need to be split further.

[0087] Consistency of child node and parent node labels (stratum attributes): Secondly, compare the child node label (stratum attribute) of the top node with the parent node label (stratum attribute). If there is a situation where the child node label (stratum attribute) is inconsistent with the parent node label (stratum attribute), it indicates that the geological interface in the area where the node is located has changed, and more refined sampling is needed to capture this change. Therefore, in this case, the node needs to be split to further refine the sampling.

[0088] If the top node of the stack satisfies both of the above conditions (i.e., the node depth does not reach Dmax and there is a child node label (stratum attribute) that is inconsistent with the parent node label (stratum attribute)), the top node of the stack is split into eight child nodes, and the child nodes are sequentially pushed into the node stack S for subsequent processing. If the splitting condition is not met, the top node of the stack is regarded as a leaf node that has completed sampling, and it is output or processed subsequently.

[0089] After the judgment and splitting operations are completed, the above steps are recursively repeated to continue processing the next node in the stack S. This process will continue until the number of nodes in the stack S is reduced to 0, that is, all nodes have completed the splitting or sampling processing.

[0090] By recursively executing the steps, the octree structure is ensured to be adaptively split according to the actual situation of the geological interface, so that finer sampling is used in key areas (such as stratum boundaries) and coarser sampling is used inside the stratum, so as to achieve the purpose of optimizing the allocation of computing resources and improving the accuracy and smoothness of the model.

[0091] Furthermore, determine the maximum number of processes that the computer can open .

[0092] This number is limited by the computer's hardware configuration and the current system resource usage. Once the maximum number of processes is determined, it can be calculated based on a predetermined formula, namely, formula (4):

[0093]

[0094] The split depth determines the level to which the octree should be pre-split before parallel sampling begins. , which can ensure that in the subsequent parallel sampling process, each process can be assigned an appropriate amount of computing tasks, thereby achieving effective utilization of computing resources.

[0095] Next, a process pool is constructed based on the determined maximum number of processes Np. The process pool will be used to execute the sampling tasks of the octave tree leaf nodes in parallel. Once the process pool is constructed, the octave tree leaf nodes obtained by splitting need to be evenly distributed to each process. This distribution process needs to ensure that each process can obtain a roughly equal number of leaf nodes to avoid imbalance in computing load. In order to achieve this goal, various load balancing algorithms, such as polling, hashing, etc., can be used to ensure the uniform distribution of leaf nodes.

[0096] After each process is assigned to its leaf node, they can start to perform sampling tasks in parallel. Each process will independently sample the leaf node to which it is assigned and calculate the corresponding stratigraphic properties and other relevant information. Once the sampling is completed, each process will send its results back to the main process. The main process is responsible for collecting the results returned by all processes and integrating them together for the subsequent construction of the 3D geological model.

[0097] Through parallel sampling technology, the sampling efficiency can be significantly improved and the time required to build a 3D geological model can be reduced. At the same time, since each process performs the sampling task independently, this method also has good scalability and fault tolerance. Even if a process fails or the computing resources are insufficient, other processes can still continue to perform the sampling task, thus ensuring the smooth progress of the entire process.

[0098] S106: Summarize the stratum attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0099] After completing the multi-process parallel octree stratum sampling, each process handles the sampling work of a part of the octree leaf nodes and obtains the stratum attributes and center coordinate data of the leaf nodes. In order to build a complete three-dimensional geological model, the data scattered in each process needs to be aggregated.

[0100] Specifically, the main process is responsible for receiving data from each sub-process and integrating the data into a unified data set. The data set contains the stratigraphic attributes and center coordinate information of all leaf nodes, which is the basis for the subsequent construction of a three-dimensional geological model.

[0101] After obtaining the stratigraphic attributes and center coordinate data of all leaf nodes, a three-dimensional geological model is constructed based on the data. To achieve this goal, the data needs to be converted into a universal three-dimensional visualization exchange format for subsequent visualization and analysis.

[0102] Specifically, according to the center coordinates and stratum attributes of each leaf node, the vertex coordinates, triangle point set index and texture information of the six surfaces of the node are calculated. Then, the information is written uniformly into a common 3D visualization exchange format file, such as OBJ, STL and other formats. The file will contain all the geometry and attribute information of the 3D geological model, which can be used for subsequent 3D visualization, analysis and further processing.

[0103] The data obtained from multi-process parallel sampling is converted into a common three-dimensional visualization format, providing strong support for subsequent geological modeling and analysis.

[0104] The present application also provides a geological modeling parallel explicit sampling device, comprising:

[0105] A collection module collects the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information in the geological exploration borehole, and constructs the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information into a point set;

[0106] A training module, using the point set to train a fully connected neural network to generate a formation classification model;

[0107] A setting module, which sets the maximum depth and the minimum depth of the octree structure, and creates the octree structure according to the point set, wherein the octree structure includes a parent node and a child node;

[0108] A prediction module, which obtains a first stratum attribute of the parent node and a second stratum attribute of the center coordinates of the child node through the stratum classification model;

[0109] A judgment module pushes the parent node and the child node of the octree structure into the node stack, and judges: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, the top node of the node stack is split into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, the top node of the node stack is output;

[0110] A recursive module recursively executes the previous step until the number of nodes in the node stack is 0;

[0111] The summary module summarizes the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0112] The present application also provides a geological modeling parallel explicit sampling device, comprising:

[0113] Memory;

[0114] A processor is used to retrieve a computer executable program of the above-mentioned geological modeling parallel explicit sampling method from the memory, and execute: collecting stratum boundary spatial position points and stratum internal points of layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points of layer thickness information into a point set; using the point set to train a fully connected neural network to generate a stratum classification model; setting the maximum depth and minimum depth of the octree structure, and creating the octree structure according to the point set, the octree structure including a parent node and a child node; obtaining the first stratum attribute of the parent node and the center of the child node through the stratum classification model The second stratigraphic attribute of the coordinate; the parent node and child nodes of the octree structure are pushed into the node stack, and it is judged that: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratigraphic attribute is inconsistent with the first stratigraphic attribute, the top node of the node stack is split into eight child nodes; if the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratigraphic attribute is consistent with the first stratigraphic attribute, the top node of the node stack is output; recursively execute the previous step until the number of nodes in the node stack is 0; summarize the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data.

[0115] The present application also provides a storage medium, including: a computer executable program stored therein, wherein the computer executable program is used to be called by a processor to execute the steps of the above-mentioned geological modeling parallel explicit sampling method.

Claims

1. A parallel explicit sampling method for geological modeling, characterized in that: include: Collecting stratum boundary spatial position points and stratum internal points with layer thickness information in geological exploration boreholes, and constructing the stratum boundary spatial position points and stratum internal points with layer thickness information into a point set; Using the point set to train a fully connected neural network to obtain a formation classification model; Setting a maximum depth and a minimum depth of an octree structure, and creating the octree structure according to the point set, wherein the octree structure includes a parent node and a child node; Acquire the first stratum attribute of the parent node and the second stratum attribute of the center coordinates of the child node through the stratum classification model; Pushing the parent node and the child node of the octree structure into a node stack, and judging: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, splitting the top node of the node stack into eight child nodes; If the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, the top node of the node stack is output; Recursively execute the previous step until the number of nodes in the node stack is 0; Summarize the stratigraphic attributes and center coordinate data of all subnodes to generate three-dimensional visualization data; It also includes: determining the maximum number of processes that can be opened by the computer; calculating the initial octree splitting depth according to the number of processes; building a process pool based on the maximum number of processes, and evenly distributing the octree leaf nodes obtained by splitting to each process; after each process completes the sampling of its child nodes, it sends the results back to the main process.

2. The parallel explicit sampling method for geological modeling according to claim 1, characterized in that: Collect the spatial position points of the stratum boundary and the internal points of the stratum thickness information in the geological exploration borehole, including: The original data of geological exploration boreholes are upsampled between borehole layers, the upsampled data interval number is set, and the internal points of the formation are sampled at intervals based on the spatial position points of the formation boundary and the layer thickness information in the borehole; The fully connected neural network is trained after the point set preprocessing to obtain a formation classification model, including: Performing forward normalization processing on the spatial coordinate data in the point set; Select an appropriate fully connected neural network architecture and randomly divide the normalized point set into training set, test set and validation set according to the preset ratio; The Adam optimizer is used to train the neural network on the training set, and the cross entropy loss is used as the loss function. After the training loss and accuracy converge, all parameters of the neural network model are saved.

3. The parallel explicit sampling method for geological modeling according to claim 1, characterized in that: Setting the maximum depth and the minimum depth of the octree structure, and creating the octree structure according to the point set, including: Based on the minimum and maximum values ​​of the preprocessed point set in three directions in the spatial rectangular coordinate system, the spatial positions of the eight vertices of the octree parent node are determined, and the octree parent node space is the rectangular bounding box space of the point set.

4. The parallel explicit sampling method for geological modeling according to claim 1, characterized in that: The stratigraphic attributes and center coordinate data of all child nodes are summarized as follows: In the main process, the sub-node stratigraphic attributes and center coordinate data returned by all processes are summarized.

5. The parallel explicit sampling method for geological modeling according to claim 1, characterized in that: The generating of three-dimensional visualization data comprises: Based on the summarized sub-node stratum attributes and center coordinate data, calculate each node attribute information, including: vertex coordinates of the six surfaces of each node, triangle face point set index, and texture information; The attribute information is uniformly written into a universal three-dimensional visualization exchange format file to construct a three-dimensional geological model.

6. A parallel explicit sampling device for geological modeling, characterized in that: include: A collection module collects the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information in the geological exploration borehole, and constructs the spatial position points of the stratum boundary and the internal points of the stratum with layer thickness information into a point set; A training module, using the point set to train a fully connected neural network to obtain a formation classification model; A setting module, which sets the maximum depth and the minimum depth of the octree structure, and creates the octree structure according to the point set, wherein the octree structure includes a parent node and a child node; A prediction module, which obtains a first stratum attribute of the parent node and a second stratum attribute of the center coordinates of the child node through the stratum classification model; A judgment module pushes the parent node and the child node of the octree structure into a node stack, and judges: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratum attribute is inconsistent with the first stratum attribute, then splits the top node of the node stack into eight child nodes; If the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, the top node of the node stack is output; A recursive module recursively executes the previous step until the number of nodes in the node stack is 0; The summary module summarizes the stratigraphic attributes and center coordinate data of all subnodes to generate three-dimensional visualization data; It also includes: determining the maximum number of processes that can be opened by the computer; calculating the initial octree splitting depth according to the number of processes; building a process pool based on the maximum number of processes, and evenly distributing the octree leaf nodes obtained by splitting to each process; after each process completes the sampling of its child nodes, it sends the results back to the main process.

7. A geological modeling parallel explicit sampling device, characterized in that: include: Memory; A processor, configured to retrieve from the memory a computer executable program of a parallel explicit sampling method for geological modeling as described in any one of claims 1 to 5, and execute: collecting spatial position points of stratum boundaries and internal points of stratums with layer thickness information in geological exploration boreholes, and constructing the spatial position points of stratum boundaries and internal points of stratums with layer thickness information into a point set; The fully connected neural network is trained using the point set to obtain a stratigraphic classification model; the maximum depth and the minimum depth of the octree structure are set, and the octree structure is created according to the point set, wherein the octree structure includes a parent node and a child node; the first stratigraphic attribute of the parent node and the second stratigraphic attribute of the center coordinates of the child node are obtained through the stratigraphic classification model; the parent node and the child node of the octree structure are pushed into a node stack, and it is judged that: if the depth of the top node of the node stack corresponding to the parent node does not reach the maximum depth, and the second stratigraphic attribute is inconsistent with the first stratigraphic attribute, the top node of the node stack is split into eight child nodes; If the depth of the top node of the node stack corresponding to the parent node reaches the maximum depth, or the second stratum attribute is consistent with the first stratum attribute, the top node of the node stack is output; Recursively execute the previous step until the number of nodes in the node stack is 0; summarize the stratigraphic attributes and center coordinate data of all child nodes to generate three-dimensional visualization data; which also includes: determining the maximum number of processes that can be opened by the computer; calculating the initial octree splitting depth according to the number of processes; building a process pool based on the maximum number of processes, and evenly distributing the octree child nodes obtained by splitting to each process; after each process completes the sampling of its child nodes, it sends the result back to the main process.

8. A storage medium, characterized in that: include: A computer executable program is stored, and the computer executable program is used to be called by a processor to execute the steps of a parallel explicit sampling method for geological modeling as described in any one of claims 1 to 5.

Images