A method for constructing a stratigraphic model by coupling a volume element segmentation algorithm with a Markov random field
Through the coupling of the coelement segmentation algorithm and Markov random field method, the three-dimensional strata modeling process is simplified, the randomness and manual intervention are reduced, the modeling efficiency and accuracy are improved, and the formation modeling model under finite drilling data is solved Accuracy issues.
Patent Information
- Application Number
- CN202410553420.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-07
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-05-07
AI Technical Summary
The existing three-dimensional strata modeling methods are difficult to accurately describe complex geological structures under limited drilling data, which have high randomness and low computing efficiency, and require a lot of manual intervention, which leads to the inability to accurately restore the real strata model.
Using coupled voxel segmentation algorithm and Markov random field method, a three-dimensional stratigraphic model is generated through data preprocessing, voxel segmentation and neighborhood system construction, simplifying the user input process, reducing randomness, improving computing efficiency, and reducing manual intervention .
The problem of continuous geological botanical spatial correlation assignment is effectively solved, the accuracy and efficiency of modeling is improved, and the generated three-dimensional stratigraphic model is closer to the real geological conditions, reducing the need for manual intervention.
Smart Images

Figure CN118504075B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of stratum structure simulation, and in particular to a stratum model construction method of coupling a volume element segmentation algorithm and a Markov random field. Background Art
[0002] Three-dimensional stratigraphic modeling is a key digital step in the hydrogeological simulation process. By comprehensively analyzing various limited topographic geology, lithology drilling, artificial logging, profile engineering survey results and other data, a digital three-dimensional stratigraphic model is generated using computer graphics processing technology. The built three-dimensional stratigraphic model is a set of integrated data bodies that can not only show users the real geological conditions in the form of intuitive graphic animations, but also provide users with stratigraphic data such as stratigraphic categories and structural distribution at any selected location in the model area, as well as geological data such as porosity, water storage, and anisotropy. The constructed three-dimensional stratigraphic model is required to accurately reflect the real geological conditions and regular changes in the modeling area to meet the data accuracy requirements of actual engineering projects and hydrogeological numerical simulations.
[0003] The main data for constructing a three-dimensional stratigraphic model is geological drilling data. However, due to the limitations of drilling costs and construction conditions, the drilling data that can be used for three-dimensional stratigraphic modeling is limited. In terms of spatial distribution, there are often problems such as small number, long intervals, and poor continuity. At the same time, there may also be problems such as large differences in stratigraphic structure between different boreholes, which makes it difficult to accurately describe the structure and spatial morphology of complex geological bodies, and even more difficult to construct a realistic and credible three-dimensional stratigraphic model.
[0004] At present, the geostatistical technology based on borehole data is the main method for generating stratigraphic models. Based on the assumption of regional stratigraphic structure continuity and heterogeneity, this method obtains a preliminary stratigraphic model through variogram calculation, and then relies on expert experience intervention and manual correction to improve the accuracy of the three-dimensional stratigraphic model. However, this process has the following two problems.
[0005] On the one hand, the geostatistical method can generate a variety of stratigraphic models based on the same drilling data, which is relatively random, and its computational efficiency and accuracy will be significantly reduced when the amount of data is large;
[0006] On the other hand, this method is relatively complex and often requires a lot of human intervention, requiring users to have sufficient expert knowledge. When the modeling scale is large, the amount of drilling data is larger and there are more repetitive situations in the vertical direction of the formation that need to be processed. The efficiency of manual intervention interpretation is low, and there is a situation where the real formation model cannot be accurately restored.
[0007] In summary, the problem faced in the process of 3D stratigraphic modeling is how to accurately and efficiently assign stratigraphic types to continuous geological units in the model space based on limited borehole data. Therefore, it is necessary to propose a fast and accurate method that can greatly reduce human intervention to solve the problem of spatial correlation assignment of continuous geological units. Summary of the invention
[0008] The purpose of the present invention is to provide a method for constructing a stratigraphic model by coupling a volume element segmentation algorithm and a Markov random field. On the one hand, it simplifies the user's input process of stratigraphic rules for the entire modeling area, effectively reduces the randomness of the modeling results, and improves the computational efficiency and accuracy; on the other hand, it effectively solves the problem of assigning spatial correlation values to continuous geological volume elements, avoids the problem of low efficiency of interpretation due to manual intervention, and the inability to accurately restore the real stratigraphic model.
[0009] To achieve the above object, the present invention provides a formation model construction method by coupling a voxel segmentation algorithm and a Markov random field, comprising the following steps:
[0010] S1 data preprocessing, organizing the original drilling data in the modeling area into a spreadsheet format, and generating a 3D drilling point data set using a 3D point drilling model;
[0011] S2 constructs a three-dimensional geological volume segmentation algorithm to generate geological volumes and completes the initialization of stratum type assignment;
[0012] S3 constructs a neighborhood system between geological elements based on Markov random fields to generate a three-dimensional stratigraphic model.
[0013] Preferably, the process of data preprocessing in step S1 is as follows:
[0014] S1.1 Extract and organize all original borehole stratigraphic data in the modeling area, including borehole number, coordinate information, surface elevation, depth and lithology layer information, and organize them into a spreadsheet format according to the specifications;
[0015] S1.2 uses the borehole data model to read the spreadsheet information established in step S1, builds a geological borehole point database, and traverses all the strata of the boreholes and numbers them according to the order of the strata. Each borehole number is a unique index of the data set.
[0016] Preferably, step S2 constructs a three-dimensional geological volume segmentation algorithm to generate geological volumes, as shown below:
[0017] S2.1 Extract the elevation values of the surface and stratum soil of each borehole based on the elevation data of the surface and bottom layers of the borehole provided by the borehole dataset, and use the Kriging interpolation algorithm to construct the three-dimensional modeling area range;
[0018] S2.2 Use the minimum bounding box algorithm to divide the three-dimensional space into uniform cubic geological elements according to the specified rules;
[0019] S2.3 maps the drilling data into three-dimensional voxels to complete the deterministic voxel assignment, and initializes the assignment state of unknown voxels based on the continuity assumption.
[0020] Preferably, based on the drilling point data set obtained in step S1.2, the plane latitude and longitude position information of all the drilling holes and the depth information of each node are obtained, and the read information is stored in the format of a two-dimensional array, expressed as x, y, z coordinates; according to the minimum bounding box method, the AABB bounding box method is used to perform the range solution, as shown below:
[0021] S2.2.1 Initialize boundary parameters: Create six variables, representing the minimum values (min_x, min_y, min_z) and maximum values (max_x, max_y, max_z) in the X, Y, and Z directions, respectively, and initialize the above variables to positive infinity and negative infinity;
[0022] S2.2.2 Traverse each borehole: For each borehole in the data set, obtain its X, Y, and Z coordinates, and for each coordinate axis, update the minimum and maximum values in the corresponding direction;
[0023] S2.2.3 Consider the modeling area generated by kriging interpolation: determine the minimum and maximum coordinate values of the modeling area generated by kriging interpolation;
[0024] S2.2.4 Update boundaries: Whenever a smaller or larger coordinate value is found, update the corresponding minimum or maximum value;
[0025] S2.2.5 Quantity constraint: The quantity in the x and y directions is determined by the user input. The quantity in the z direction requires the thinnest formation thickness d of the drilling data set to be calculated first. min , and then through the already obtained boundaries max_z and min_z in the z direction, the number in the z direction is (max_z-min_z) / (d min / 2);
[0026] S2.2.6 Final result: Return six boundary values, combined with the number of voxels in each direction that have been determined, and finally generate a three-dimensional shape of (n x , n y , n z ) of the tetrahedron.
[0027] Preferably, step S3 constructs a neighborhood system between geological voxels as follows:
[0028] S3.1 Label the state of the voxels and construct the neighborhood system between the voxels: convert the state attribute information of all voxels into state attribute labels, forming the index set N of the voxels, the label set S of the stratum attributes and the label set L of the state space voxels;
[0029] S3.2 Based on the Markov random field model, the graph theory algorithm is used to construct the minimum neighborhood system for different voxels;
[0030] S3.3 establishes the energy function of the volume element and uses Gibbs sampling to update the state of the non-drilled intersection unit;
[0031] S3.4 introduces a penalty function to change the energy function after the voxel state is updated, and solves the optimal formation voxel state estimation that makes the system energy the lowest, and finally generates a three-dimensional formation model.
[0032] Preferably, step S3.2 constructs a minimum neighborhood system for all voxels according to the locality of the Markov random field model, and uses a graph theory algorithm to ensure that each minimum neighborhood system does not contain repeated voxels. The specific steps are as follows:
[0033] S3.2.1 Graph construction: The geospatial voxels are represented as nodes of the graph, and the relationships between voxels are represented as edges of the graph;
[0034] S3.2.2 Initialize the neighborhood system: For each voxel, initialize an empty neighborhood system set;
[0035] S3.2.3 Traverse the nodes of the graph: Start from any node in the graph and traverse the nodes one by one;
[0036] S3.2.4 For each node: obtain the neighboring nodes of the current node, check whether each neighboring node is already in the neighborhood system of the current node, and if the neighboring node is not in the neighborhood system of the current node, add it to the neighborhood system;
[0037] S3.2.5 Repeat: Repeat steps S3.2.3 and S3.2.4 until all nodes are traversed;
[0038] S3.2.6 Completion: After the traversal is complete, each node has a non-repeating neighborhood system that includes its adjacent nodes.
[0039] Preferably, in step S3.3, the process of Gibbs sampling is as follows:
[0040] S3.3.1 Initialization: Based on the initial sample Assign an initial stratigraphic attribute to each non-deterministic voxel;
[0041] S3.3.2 performs iterative updates for the i loop, and for each dimension j, performs the following steps:
[0042] From the fully conditional distribution Extract in Indicates that after removing the sample of the jth dimension, the sample set is obtained
[0043] S3.3.3 Iterate until the number of cycles is reached or no more updates are made.
[0044] Therefore, the present invention adopts the above-mentioned coupled volume element segmentation algorithm and Markov random field stratigraphic model construction method. The user only needs to organize the drilling data into the standard format required by the algorithm and input the modeling area, and the three-dimensional geological volume element segmentation algorithm can automatically generate a three-dimensional stratigraphic model, which simplifies the user's input process of stratigraphic rules for the entire modeling area, effectively reduces the randomness of the modeling results, and improves the calculation efficiency and accuracy; using the Markov random field model, the spatial distribution of geological states in the modeling area is mined from limited original data, which effectively solves the problem of assigning spatial correlation of continuous geological volume elements, avoids the low efficiency of manual intervention interpretation, and the problem of being unable to accurately restore the real stratigraphic model.
[0045] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 It is a flow chart of the formation model construction method of the present invention coupling the volume element segmentation algorithm and the Markov random field;
[0047] Figure 2 It is a technical roadmap of the method for constructing a formation model by coupling a volume element segmentation algorithm and a Markov random field in the present invention;
[0048] Figure 3 It is a schematic diagram of the drilling layered structure of the formation model construction method of the present invention coupling the volume element segmentation algorithm and the Markov random field;
[0049] Figure 4 It is a schematic diagram of the drilling attributes of the formation model construction method of the present invention coupling the volume element segmentation algorithm and the Markov random field;
[0050] Figure 5 It is a schematic diagram of a deterministic voxel of a formation model construction method of the present invention coupling a voxel segmentation algorithm and a Markov random field;
[0051] Figure 6 It is a distribution diagram of each layer of deterministic voxels in the stratigraphic model building method of the present invention coupling the voxel segmentation algorithm and the Markov random field;
[0052] Figure 7It is a distribution diagram of each layer of deterministic voxels in the stratigraphic model building method of the present invention coupling the voxel segmentation algorithm and the Markov random field;
[0053] Figure 8 It is a distribution diagram of each layer of deterministic voxels in the stratigraphic model building method of the present invention coupling the voxel segmentation algorithm and the Markov random field;
[0054] Fig. 9 It is a neighborhood system schematic diagram of the method for constructing a stratum model by coupling a volume element segmentation algorithm and a Markov random field according to the present invention;
[0055] Fig.10 It is a three-dimensional stratigraphic structure result diagram of the stratigraphic model building method of the present invention coupling the volume element segmentation algorithm and the Markov random field. DETAILED DESCRIPTION
[0056] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0057] like Figure 1 and Figure 2 As shown, the method for constructing a stratum model by coupling a volume element segmentation algorithm and a Markov random field of the present invention comprises the following steps:
[0058] S1 data preprocessing: organize the original drilling data in the modeling area into a spreadsheet format, and use the 3D point drilling model to generate a 3D drilling point data set. The specific process is as follows:
[0059] S1.1 Extract and organize all original borehole stratigraphic data in the modeling area, including borehole number, coordinate information, surface elevation, depth and lithology layer information, and organize them into a spreadsheet format according to the specifications;
[0060] S1.2 uses the borehole data model to read the spreadsheet information established in step S1, builds a geological borehole point database, and traverses all the strata of the boreholes and numbers them according to the order of the strata. Each borehole number is a unique index of the data set.
[0061] S2 constructs a 3D geological volume segmentation algorithm to generate geological volumes and completes the initialization of stratum type assignment, as shown below:
[0062] S2.1 Extract the elevation values of the surface and stratum soil of each borehole based on the elevation data of the surface and bottom layers of the borehole provided by the borehole dataset, use the Kriging interpolation algorithm to realize the top and bottom interpolation within the modeling area, and construct the three-dimensional modeling area range;
[0063] S2.2 Use the minimum bounding box algorithm to divide the three-dimensional space into uniform cubic geological elements according to the specified rules;
[0064] Based on the drilling point data set obtained in step S1.2, the plane latitude and longitude position information of all drilling holes and the depth information of each node are obtained. The read information is stored in the format of a two-dimensional array, with three numbers in each row, representing the x, y, and z coordinates respectively. According to the minimum bounding box method, the AABB bounding box method is used to perform this range solution, as shown below:
[0065] S1.2.1 Initialize boundary parameters: Create six variables to represent the minimum values (min_x, min_y, min_z) and maximum values (max_x, max_y, max_z) in the X, Y, and Z directions, respectively, and initialize these variables to positive infinity and negative infinity;
[0066] S1.2.2 Traverse each borehole: For each borehole in the data set, obtain its X, Y, and Z coordinates, and for each coordinate axis, update the minimum and maximum values in the corresponding direction;
[0067] S1.2.3 Consider the modeling area generated by kriging interpolation: determine the minimum and maximum coordinate values of the modeling area generated by kriging interpolation;
[0068] S1.2.4 Update boundaries: Whenever a smaller or larger coordinate value is found, update the corresponding minimum or maximum value;
[0069] S1.2.5 Quantity constraints: The quantities in the x and y directions are determined by the user input. The quantity in the z direction requires the thinnest formation thickness d of the drilling data set to be calculated first. min , and then through the already obtained boundaries max_z and min_z in the z direction, the number in the z direction is (max_z-min_z) / (d min / 2);
[0070] S1.2.6 Final result: Return six boundary values, combined with the number of voxels in each direction that have been determined, and finally generate a three-dimensional shape of (n x , n y , n z ) of the tetrahedron.
[0071] S2.3 maps the drilling data into three-dimensional voxels to complete the deterministic voxel assignment, and initializes the assignment state of unknown voxels based on the continuity assumption.
[0072] After generating voxels using the AABB bounding box method based on step S1.2, the spatial coordinate information of the drilling data is normalized to make it consistent with the coordinate system of the voxel grid. The data of each drilling point is then assigned to the voxel closest to the spatial position as the state value of the voxel, and these known voxels are used as deterministic voxels. There is no need to iteratively update the state in subsequent calculations. For voxels with unknown states, the initial state is set using a distance-weighted method based on the identified stratigraphic pattern and the assumption of stratigraphic continuity.
[0073] Assuming that the stratigraphic types have a certain degree of continuity in space, unknown elements that are closer are more likely to be similar to nearby stratigraphic types. Define the distance calculation formula between unknown elements and known geological information. Commonly used distance measurement methods include Euclidean distance, Manhattan distance, or other geologically appropriate distance indicators.
[0074] Assuming the coordinates of the unknown voxel are (x, y, z) and the coordinates of the known geological information are (x', y', z'), the distance can be expressed as:
[0075]
[0076] Among them, the state of the nearest known voxel is taken as the initialization state of the unknown voxel.
[0077] S3 constructs a neighborhood system between geological voxels based on Markov random fields and generates a three-dimensional stratigraphic model as shown below:
[0078] S3.1 Label the voxel states and construct a neighborhood system between voxels: convert the state attribute information of all voxels into state attribute labels to form the index set N of the voxels, the label set S of the stratum attributes and the label set L of the state space voxels.
[0079] According to the needs of Markov random field, all spatial voxel states are labeled. Assume that N = {i|1, 2, ..., n} is the voxel index set, where n = n x ×n y ×n z ; S is the label set of all stratigraphic attributes, and the number of S is equal to the stratigraphic type of known drilling data; L is the label set of state space voxels, which mainly considers the impact of spatial position on stratigraphic attribute labels. The stratigraphic attributes are used as labels of the state space, and the spatial position, i.e., the voxel index, is used as the label of the observation space.
[0080] S3.2 According to the requirements of the Markov random field model, the graph theory algorithm is used to construct the minimum neighborhood system for different voxels;
[0081] According to the locality of the Markov random field model, a minimum neighborhood system is constructed for all voxels, and the neighborhood system is used to characterize the correlation between voxels. In the Markov random field model, the elements in the voxel set N are interconnected through the neighborhood system, and the neighborhood system is defined as O = {O i |i∈S}. Among them, O i It refers to the minimum neighborhood system in this three-dimensional voxel space. To avoid redundancy in subsequent calculations, a graph theory algorithm is used to ensure that each minimum neighborhood system does not contain repeated voxels. The specific steps are as follows:
[0082] S3.2.1 Construct a graph: Represent geospatial voxels as nodes of the graph, and the relationships between voxels (e.g., horizontal and vertical relationships) as edges of the graph;
[0083] S3.2.2 Initialize the neighborhood system: For each voxel, initialize an empty neighborhood system set;
[0084] S3.2.3 Traverse the nodes of the graph: Start from any node in the graph and traverse the nodes one by one;
[0085] S3.2.4 For each node: obtain the neighboring nodes of the current node, check whether each neighboring node is already in the neighborhood system of the current node. If the neighboring node is not in the neighborhood system of the current node, add it to the neighborhood system;
[0086] S3.2.5 Repeat: Repeat steps S3.2.3 and S3.2.4 until all nodes are traversed;
[0087] S3.2.6 Completion: After the traversal is complete, each node has a non-repeating neighborhood system that includes its adjacent nodes.
[0088] S3.3 establishes the energy function of the volume element and uses Gibbs sampling to update the state of the non-drilled intersection unit;
[0089] According to the existing three-dimensional voxel space and the existing deterministic voxel information, the prior probability distribution of the Markov random field model is constructed. According to the determined borehole voxel, there are S types of stratum attributes of the voxel according to step S3.1, and there are n types of depth ranges according to step S2.3. z There are n deterministic voxels in total.
[0090] It is known that the formation attribute is a discrete random variable whose value range is {1, 2, …, s}, where s is the number of formation types. The probability vector of the formation attribute in each interval is p i =(pi1 ,p i2 ,…,p is ), where p ij represents the probability that the formation attribute is of the jth type in the i-th depth interval. In this way, the distribution of formation attributes in each depth interval can be expressed as a multinomial distribution:
[0091] P(stratum attribute|depth interval)=Multinomial(p i )
[0092] Over the entire depth range, considering that there are n deterministic voxels, the distribution of stratigraphic attributes in each depth interval can be given a Dirichlet prior:
[0093]
[0094] Among them, α 1 , α 2 , …, α s can be considered as the parameter distribution of stratigraphic properties at each depth.
[0095] In the Markov random field model, it is assumed that the state of each voxel depends only on the state of its neighboring voxels, and is independent of the states of other voxels. The Markov random field model of stratum voxels is represented by an undirected graph with a grid structure, where each node represents a voxel and each edge represents the correlation between two adjacent voxels. This assumption can be expressed by the following formula:
[0096] p(x i |x i-1 ,y)=p(x|x N(i) ,y
[0097] Among them, p(x i |x i-1 ,y) represents the conditional probability of the state of the i-th voxel given the states of all other voxels and the drilling data; p(x|x N(i) ,y) represents the conditional probability of the state of the i-th element given the state of the neighborhood elements of the i-th element and the drilling data; N(i) represents the set of neighborhood elements of the i-th element.
[0098] Event S=s represents a joint event (S i =s,i∈N), where s = {s,…,s n} represents the stratum attribute label of S (e.g. sand, clay, silt, etc.). For the realization of random field S, S = {s = {s,…,s n}|s i∈L,i∈L} are all possible stratum attribute labels in the modeling area. If the following conditions are met, then S is called the MRF of N for the neighborhood system O:
[0099] P(S=s)>0,s∈S
[0100]
[0101] where N-{i} is the set of voxels excluding i. According to the Hammersley-Clifford theorem, S is a Markov random field over N with respect to a neighborhood O if and only if the configuration of S obeys the Gibbs distribution over N with respect to a neighborhood O.
[0102] A clique is a subgraph in an undirected graph where any two nodes are connected by an edge. The maximum clique is a clique to which no more nodes can be added. In this method, the minimum neighborhood system corresponds to the maximum clique. In the application of Markov models, a potential function is defined to represent the correlation of the variables. The potential function is represented by the following formula:
[0103]
[0104] Among them, ψ C (S i ,S j ) represents the potential function of the neighborhood system; S i and S j represents different voxels in the neighborhood system; distance(i,j) represents the spatial distance between voxels i and j. σ is a parameter that controls the shape of the Gaussian function; similarity(S i ,S j ) is a measure of voxel state similarity.
[0105] The Gibbs distribution of the potential function refers to the probability distribution of the state variables of the random field under the condition of given observation data. It is determined by the energy function of the potential function. The smaller the energy function, the greater the probability. The relationship between the Gibbs distribution and the potential energy function can be expressed by the following formula:
[0106]
[0107] Among them, s is a vector representing the state of all voxels; x is a vector representing the attribute labels of all voxels; P(s|x) represents the conditional probability of the voxel state under the condition of given drilling data, E(s|x) is the total energy function, and Z(s) is the exponential sum of the energy functions of all possible voxel states, as shown below:
[0108]
[0109] Minimizing the energy function means finding the optimal voxel state under the condition of given observation data so that the energy function reaches the minimum value, as shown below:
[0110]
[0111] Among them, s * represents the optimal voxel state, argmin s Indicates the state that minimizes the energy function, argmax s It means taking the state that maximizes the conditional probability.
[0112] In the Markov random field model, in order to solve the optimal solution of the Markov random field model, that is, the lowest energy function, the state of the non-deterministic voxel will be updated. The process of Gibbs sampling is as follows:
[0113] S3.3.1 Initialization: Based on the initial sample Assign an initial stratigraphic attribute to each non-deterministic voxel;
[0114] S3.3.2 performs iterative updates for the i loop, and for each dimension j, performs the following steps:
[0115] From the fully conditional distribution Extract in Indicates that after removing the sample of the jth dimension, the sample set is obtained
[0116] S3.3.3 Iterate until the number of cycles is reached or no more updates are made.
[0117] S3.4 introduces a penalty function to change the energy function after the voxel state is updated, and solves the optimal formation voxel state estimation that makes the system energy the lowest, and finally generates a three-dimensional formation model.
[0118] Based on the above-mentioned Gibbs sampling, state mutations are involved in the process of updating the state of the voxel. State mutations refer to the drastic changes in the state of the voxel at adjacent locations, such as a sudden change from fill to silt, or from silty clay to silt. This state mutation is usually inconsistent with geological laws, so a penalty term needs to be added to the energy function to increase the energy of the state mutation, thereby meeting the goal of minimizing the system energy.
[0119] One method of penalizing energy functions for state mutations is to use the penalty function method, which is a method for transforming constrained optimization problems into unconstrained optimization problems. The basic idea of the penalty function method is to add a penalty function related to the constraints to the original objective function, so that when the constraints are violated, the value of the penalty function is large, thereby increasing the value of the objective function; when the constraints are met, the value of the penalty function is small or zero, thereby not affecting the value of the objective function.
[0120] A commonly used penalty function is the quadratic penalty function, which uses the square of the constraint as the penalty function, as shown below:
[0121]
[0122] Among them, s is a vector representing the state of all voxels; f(s) is the original objective function, that is, the total energy function; σ is a positive penalty factor; c i (s) is the i-th constraint, which is a function of s; when c i When (s) = 0, it means that the constraint condition is satisfied and the penalty function is zero; when c i When (s)≠0, it means that the constraint is violated, the penalty function is positive, and as c i (s) increases.
[0123] The degree of state mutation is taken as a constraint, as follows:
[0124] c i (s)=|s i -s i+1 |
[0125] Among them, s i and i+1 Indicates the state of two adjacent voxels, |s i -s i+1 | represents the degree of state mutation. The larger it is, the more drastic the state mutation is. This constraint is added to the energy function to form a penalty function, as shown below:
[0126]
[0127] Among them, E(s,x) is the original energy function, s is a vector representing all drilling data, and σ is a positive penalty factor, which can be adjusted according to the specific situation of the problem.
[0128] In the formation model, the penalty factor for state mutation in the horizontal direction should be smaller than the penalty factor for state mutation in the vertical direction. In this way, when a state mutation occurs, the value of the penalty function will increase, thereby increasing the energy of the state mutation and reducing the probability of state mutation; when the state mutation does not occur, the value of the penalty function will be zero, thus not affecting the original energy function, so that the probability of unchanged state remains unchanged; combined with the energy function representation of the above neighborhood system, similarity (S i ,S j ) is used as the penalty function control factor.
[0129] Example
[0130] like Figure 1 and Figure 2 As shown, the formation model construction method of the coupled voxel segmentation algorithm and Markov random field of the present invention is used to perform corresponding modeling for the spatial coordinates of 29 wells and formation drillability data in a certain place, including the following steps:
[0131] 1. Sort out the original drilling data in the modeling area. Figure 3 The judgment and identification of the strata in the borehole column chart shown is used to establish an electronic table including information such as borehole number, plane coordinate data, depth, thickness, and stratum attributes.
[0132] 2. Construct a geological borehole database and build a borehole data model. Read the spreadsheet information established in S1.1 and generate a hierarchical borehole point data set according to the order of the borehole numbers.
[0133] 3. Use the borehole data model to read the point data set data, start the initial stratum pattern recognition, complete the initial borehole stratum attribute extraction, and complete the three-dimensional distribution of the borehole stratum data.
[0134] 4. Based on the modeling area input by the user and the elevation range of the surface and bottom soils provided by the drilling point dataset, the three-dimensional space of the modeling range is generated using Kriging interpolation. Using the AABB bounding box method combined with the three-dimensional space range generated by Kriging interpolation, the minimum value (min_x, min_y, min_z) and maximum value (max_x, max_y, max_z) of the 29 input drilling data in the X, Y and Z directions are found to define this three-dimensional space experience.
[0135] The number of voxels in the X and Y directions is defined as 50. In order to ensure that the generated 3D stratigraphic model can meet the thinnest stratigraphic requirements, the thinnest stratigraphic unit in the 29 borehole data input this time is 3m. Half of the thickness of the thinnest stratigraphic unit is taken as the height of the voxel in the Z direction, and the number of voxels generated in the Z direction is 60. The spatial shape of the cube voxel adaptively generated in the 3D space where the final borehole is located is 50*50*60.
[0136] 5. Based on the number of generated voxels, the spatial extent was adaptively processed, the geographic information was normalized, and the three-dimensional boreholes were mapped to the defined voxel space. A total of 1459 voxels were determined.
[0137] There are 10 types of stratigraphic attributes in the 29 boreholes input this time: fill, clay, silty clay, silty clay mixed with silt, silt, silt mixed with silty clay, silt mixed with silty sand, silt sand, silt mixed with silt, and silty sand. Figure 4 As shown. Map the existing drill holes into the voxel space to obtain deterministic voxels, such as Figure 5 The distribution of deterministic voxels in each layer in the z direction is shown in Figure 6 , Figure 7 and Figure 8 As shown in Figure 2, the initialization of the undetermined voxel state is completed based on the continuity assumption.
[0138] 6. Complete attribute labeling for all voxels. The set of stratum attributes S is "fill": 1; "clay": 2; "silty clay": 3; "silty clay mixed with silty soil": 4; "silty soil": 5; "silty soil mixed with silty clay": 6; "silty soil mixed with silty sand": 7; "silty sand": 8; "silty sand mixed with silty soil": 9; "silty fine sand": 10. Voxel index set N = {i|1, 2, ..., 15000}. State space voxel L = {Li = (s, n i ), i|1, 2, …, 15000};
[0139] 7. Construct the neighborhood system set O of the voxel according to step S3.2, such as Fig. 9 As shown; using the Markov random field model, the neighborhood system is used as the minimum calculation unit to represent the potential function. According to the number of stratum types in different depth spaces of deterministic voxels, the prior distribution of stratum types is constructed. According to this prior distribution, Gibbs sampling is used to update the voxel state. The number of iterations is set to 5000 times. After each iterative update, the total energy of the stratum unit system is recalculated according to the formula. Finally, the stratum voxel state estimate with the lowest system energy in the iterative process is taken as the optimal solution, and the following is obtained: Fig.10 The three-dimensional stratigraphic model shown.
[0140] Therefore, the present invention adopts the above-mentioned coupled voxel segmentation algorithm and Markov random field stratigraphic model construction method, which, on the one hand, simplifies the user's input process of stratigraphic rules for the entire modeling area, effectively reduces the randomness of the modeling results, and improves computational efficiency and accuracy; on the other hand, it effectively solves the problem of spatial correlation assignment of continuous geological voxels, avoids the problem of low efficiency of manual intervention interpretation, and the inability to accurately restore the real stratigraphic model.
[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A method for constructing a stratum model by coupling a volume element segmentation algorithm and a Markov random field, characterized in that: The following steps are involved: S1 data preprocessing: organize the original drilling data in the modeling area into a spreadsheet format, and use the 3D point drilling model to generate a 3D drilling point data set. The specific process is as follows: S1.1 Extract and organize all original borehole stratigraphic data in the modeling area, including borehole number, coordinate information, surface elevation, depth and lithology layer information, and organize them into a spreadsheet format according to the specifications; S1.2 Use the borehole data model to read the established spreadsheet information, construct a geological borehole point data set, and traverse all the strata of the boreholes and number them according to the order of the old and new strata. Each borehole number is a unique index of the data set; S2 constructs a 3D geological volume segmentation algorithm to generate geological volumes and completes the initialization of stratum type assignment. The specific process is as follows: S2.1 Extract the elevation values of the surface and stratum soil of each borehole based on the elevation data of the surface and bottom layers of the borehole provided by the borehole dataset, and use the Kriging interpolation algorithm to construct the three-dimensional modeling area range; S2.2 Use the minimum bounding box algorithm to divide the three-dimensional space into uniform cubic geological elements according to the specified rules; S2.3 maps the drilling data into three-dimensional voxels to complete deterministic voxel assignment, and initializes the assignment state of unknown voxels based on the continuity assumption; S3 constructs a neighborhood system between geological elements based on Markov random fields and generates a three-dimensional stratigraphic model. The specific process is as follows: S3.1 Label the state of the voxels and construct the neighborhood system between the voxels: convert the state attribute information of all voxels into state attribute labels, forming the index set N of the voxels, the label set S of the stratum attributes and the label set L of the state space voxels; S3.2 Based on the Markov random field model, the graph theory algorithm is used to construct the minimum neighborhood system for different voxels; S3.3 establishes the energy function of the volume element and uses Gibbs sampling to update the state of the non-drilled intersection unit; S3.4 introduces a penalty function to change the energy function after the voxel state is updated, and solves the optimal formation voxel state estimation that makes the system energy the lowest, and finally generates a three-dimensional formation model.
2. The method for constructing a stratum model by coupling a voxel segmentation algorithm and a Markov random field according to claim 1, characterized in that: Based on the drilling point data set obtained in step S1.2, the plane latitude and longitude position information of all drilling holes and the depth information of each node are obtained, and the read information is stored in the format of a two-dimensional array, represented as x, y, z coordinates; according to the minimum bounding box method, the AABB bounding box method is used to perform this range solution, as shown below: S2.2.1 Initialize boundary parameters: Create six variables, representing the minimum values (min_x, min_y, min_z) and maximum values (max_x, max_y, max_z) in the X, Y, and Z directions, respectively, and initialize the above variables to positive infinity and negative infinity; S2.2.2 Traverse each borehole: For each borehole in the data set, obtain its X, Y, and Z coordinates, and for each coordinate axis, update the minimum and maximum values in the corresponding direction; S2.2.3 Consider the modeling area generated by kriging interpolation: determine the minimum and maximum coordinate values of the modeling area generated by kriging interpolation; S2.2.4 Update boundaries: Whenever a smaller or larger coordinate value is found, update the corresponding minimum or maximum value; S2.2.5 Quantity constraint: The quantity in the x and y directions is determined by the user input. The quantity in the z direction requires the thinnest formation thickness d of the drilling data set to be calculated first. min , and then through the already obtained boundaries max_z and min_z in the z direction, the number in the z direction is (max_z-min_z) / (d min / 2); S2.2.6 Final result: Return six boundary values, combined with the number of voxels in each direction that have been determined, and finally generate a three-dimensional shape of (n x , n y , n z ) of the tetrahedron.
3. The method for constructing a stratum model by coupling a voxel segmentation algorithm and a Markov random field according to claim 1, characterized in that: Step S3.2 constructs a minimum neighborhood system for all voxels based on the locality of the Markov random field model, and uses a graph theory algorithm to ensure that each minimum neighborhood system does not contain duplicate voxels. The specific steps are as follows: S3.2.1 Graph construction: The geospatial voxels are represented as nodes of the graph, and the relationships between voxels are represented as edges of the graph; S3.2.2 Initialize the neighborhood system: For each voxel, initialize an empty neighborhood system set; S3.2.3 Traverse the nodes of the graph: Start from any node in the graph and traverse the nodes one by one; S3.2.4 For each node: obtain the neighboring nodes of the current node, check whether each neighboring node is already in the neighborhood system of the current node, and if the neighboring node is not in the neighborhood system of the current node, add it to the neighborhood system; S3.2.5 Repeat: Repeat steps S3.2.3 and S3.2.4 until all nodes are traversed; S3.2.6 Completion: After the traversal is complete, each node has a non-repeating neighborhood system that includes its adjacent nodes.
4. The method for constructing a stratum model by coupling a voxel segmentation algorithm and a Markov random field according to claim 1, characterized in that: In step S3.3, the Gibbs sampling process is as follows: S3.3.1 Initialization: Based on the initial sample Assign an initial stratigraphic attribute to each non-deterministic voxel; S3.3.2 performs iterative updates for the i loop, and for each dimension j, performs the following steps: From the fully conditional distribution Extract in Indicates that after removing the sample of the jth dimension, the sample set is obtained S3.3.3 Iterate until the number of cycles is reached or no more updates are made.
