CN117611485B - Three-dimensional core permeability prediction method based on space-time diagram neural network - Google Patents
Three-dimensional core permeability prediction method based on space-time diagram neural network Download PDFInfo
- Publication number
- CN117611485B CN117611485B CN202410097446.4A CN202410097446A CN117611485B CN 117611485 B CN117611485 B CN 117611485B CN 202410097446 A CN202410097446 A CN 202410097446A CN 117611485 B CN117611485 B CN 117611485B
- Authority
- CN
- China
- Prior art keywords
- space
- representing
- core
- data
- dimensional
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000010586 diagram Methods 0.000 title claims abstract description 101
- 238000000034 method Methods 0.000 title claims abstract description 91
- 230000035699 permeability Effects 0.000 title claims abstract description 54
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 33
- 239000011148 porous material Substances 0.000 claims abstract description 37
- 230000007246 mechanism Effects 0.000 claims abstract description 30
- 230000011218 segmentation Effects 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 5
- 230000002708 enhancing effect Effects 0.000 claims abstract description 4
- 238000003062 neural network model Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 24
- 239000011435 rock Substances 0.000 claims description 24
- 230000009467 reduction Effects 0.000 claims description 15
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000005520 cutting process Methods 0.000 claims description 11
- 238000009826 distribution Methods 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- 239000012530 fluid Substances 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 230000001186 cumulative effect Effects 0.000 claims description 6
- 238000005315 distribution function Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 claims description 3
- 238000012512 characterization method Methods 0.000 claims description 3
- 238000002591 computed tomography Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005325 percolation Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000000547 structure data Methods 0.000 claims description 3
- 238000007667 floating Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011545 laboratory measurement Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The invention discloses a three-dimensional core permeability prediction method based on a space-time diagram neural network, which comprises the following steps of: performing three-dimensional reconstruction on the core sample image sequence of X-ray μCT scanning; filtering three-dimensional data noise of the core, enhancing contrast between a core entity and a hole seam, and executing automatic threshold segmentation to obtain binarized three-dimensional data; dividing the three-dimensional data volume into small-volume samples, and calculating the core permeability by using a Darcy formula to serve as a data tag; extracting a pore network from the three-dimensional core of each small-volume sample, and processing the pore network into space-time diagram representation data; and inputting the space-time diagram representation data into a space-time diagram neural network model, extracting the space-time characteristics of the core by using a diagram neural network with a multi-query attention mechanism and a gating logic unit module, and predicting the permeability by using a multi-layer perceptron. Compared with the prior art, the method has better physical science interpretation, simultaneously focuses on the permeation network channel, reduces the interference of data redundancy components, and is suitable for actual scene requirements.
Description
Technical Field
The invention relates to the technical field of rock permeability prediction, in particular to a three-dimensional core permeability prediction method based on a space-time diagram neural network.
Background
In the field of oil and gas exploration and development, rock permeability is an important parameter that has a critical impact on assessing the effectiveness and productivity of oil and gas exploration and development. In order to accurately evaluate the potential of oil and gas resources and the exploitation effect, accurately and efficiently evaluate the rock permeability, the method is a key for formulating a reasonable oil exploitation scheme, improving the recovery ratio and optimizing the production effect, and has very important value and significance.
With the continuous development of computer technology and numerical simulation methods, digital petrography and digital core permeability prediction technology are gradually mature. The physical and chemical characteristics of the rock are modeled and simulated by using the computer, so that the core permeability can be predicted more accurately, and more reliable technical support is provided for oil and gas exploration. Traditionally, the measurement of rock permeability typically requires expensive and time-consuming experimentation, requiring acquisition of core samples for laboratory measurements. However, this approach is costly and also involves risks and uncertainties. Therefore, the core permeability is predicted by using a computer simulation and prediction method, so that the cost and risk can be greatly reduced.
However, conventional permeability measurement methods have some limitations and challenges. Conventional methods of machine learning typically suffer from computational redundancy, limited to small sample predictions, and lack of fused physical knowledge. However, in a practical demand scenario, artificial intelligence algorithms that lack interpretability are difficult to approve and put into production. Further, the conventional method of machine learning is also limited by the constraint of hardware resources, and only samples with smaller data size can be predicted. Therefore, how to solve the problems of redundancy in calculation amount, limitation to small sample prediction and lack of fusion physical knowledge in the conventional permeability measurement is a urgent need to be solved.
The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.
Disclosure of Invention
The invention mainly aims to provide a three-dimensional core permeability prediction method based on a space-time diagram neural network, and aims to solve the technical problems of calculation redundancy, limitation to small sample prediction and lack of fusion physical knowledge existing in the traditional permeability measurement at present.
In order to achieve the above purpose, the invention provides a three-dimensional core permeability prediction method based on a space-time diagram neural network, which comprises the following steps:
s1: performing X-ray mu CT scanning on the core sample, and reconstructing a three-dimensional medium of an image sequence to obtain core three-dimensional data;
s2: preprocessing three-dimensional data of a core, filtering data noise, enhancing contrast between a core entity and a pore, executing automatic threshold segmentation, separating an internal entity of the core from a pore structure, and converting the internal entity of the core and the pore structure into binarized three-dimensional data;
s3: cutting the three-dimensional data of the core from any axial direction of X/Y/Z to obtain three-dimensional core cubes, and calculating core permeability of each three-dimensional core cube according to a Darcy theorem formula to obtain an extended data set as a data label;
s4: extracting network pore structure data of each three-dimensional rock core in the extended data set, and generating space-time diagram representation data;
s5: the time-space diagram representation data are input into a constructed time-space diagram neural network model, firstly, the spatial characteristics of a rock core are extracted based on a diagram neural network with a multi-query attention mechanism, then, the time sequence relationship of the rock core is captured by a time sequence network layer of a gating logic unit, and finally, the permeability is predicted through a multi-layer perceptron.
Optionally, the step S1 specifically includes:
s11: registering and cutting the two-dimensional image slice sequence scanned by the X-ray μCT, firstly aligning the two-dimensional image slices in the same coordinate system, then determining the direction information of the image sequence, and finally cutting to the same size;
s12: converting the two-dimensional image slice sequence after registration cutting into voxel three-dimensional data;
s13: and carrying out three-dimensional reconstruction on the voxelized three-dimensional data, and combining all voxels together to generate three-dimensional reconstruction data of the internal structure of the core.
Optionally, the step S2 specifically includes:
s21: carrying out Gaussian smoothing filtering on the obtained three-dimensional data of the core, and carrying out convolution operation by using a Gaussian filter of 7 x 7 to obtain the smoothed three-dimensional data of the core;
s22: filtering and denoising the smoothed three-dimensional data of the rock core, creating a noise template according to the type and the intensity of the noise, and applying the noise template to the three-dimensional data of the rock core to obtain the three-dimensional data after denoising;
s23: performing histogram equalization on the three-dimensional data after noise reduction, firstly calculating a histogram of the three-dimensional data after noise reduction to obtain a cumulative distribution function, then performing gray level mapping through a mapping function, and finally performing gamma transformation function adjustment to obtain core three-dimensional data after contrast enhancement;
s24: and performing automatic threshold segmentation on the core three-dimensional data after contrast enhancement, firstly normalizing a data gray level distribution histogram, then calculating probability distribution of gray levels, entropy of each gray level and entropy of left and right sides, secondly calculating weighted average entropy of each gray level, finally finding out a gray level corresponding to the maximum value, and binarizing an image by using the gray level as a threshold value to obtain binarized three-dimensional data.
Optionally, the noise reduction processing formula is as follows:
in the method, in the process of the invention,is the pixel value in the output image, representing the pixel value after noise reduction, < >>Is a pixel value in the input image, representing the original pixel value, λ is a super-parameter, representing the intensity of noise reduction;
the formula of the histogram equalization is as follows:
in the method, in the process of the invention,representing cumulative distribution function->Representing a mapping function->The number of pixels representing the jth gray level in the original image, N representing the total number of pixels,/and->Represents the sum of the number of pixels from the 0 th gray level to the i th gray level, L represents the number of gray levels, ">Representing a numerical processing function for rounding floating point numbers;
the formula of the gamma transformation function is as follows:
in the method, in the process of the invention,representing a mapping relation of gamma transformation, wherein x represents a pixel value in original data, and Γ represents a gamma value;
the automated thresholding formula is as follows:
in the method, in the process of the invention,representing the probability distribution of each gray level, +.>A pixel number representing an ith gray level, N representing a total pixel number of data;
in the method, in the process of the invention,an entropy representing each gray level, i representing the gray level;
in the method, in the process of the invention,representing left entropy, i ranges from 0 to t-1;
in the method, in the process of the invention,representing right entropy, i ranges from t to 255;
in the method, in the process of the invention,representing a weighted average entropy, t representing a current gray level;
where t represents an optimal threshold, where t ranges from 0 to 255.
Optionally, the step S3 specifically includes:
s31: taking Chinese standard time as a random seed, randomly setting a starting point position, a segmentation size and a stepping length of a sliding window, sliding and segmenting an original three-dimensional data body in any axial direction by X/Y/Z, firstly, sliding the sliding window from the random starting point position, moving the sliding window by the set stepping length, then axially intercepting data in the direction according to the set segmentation size, and finally obtaining a data set;
s32: according to the Darcy theorem formula, core permeability is calculated for the data set, and the mathematical formula is expressed as follows:
wherein Q represents the rate of fluid passing through the porous medium, k represents the permeability coefficient of the porous medium, a represents the cross-sectional area, Δh represents the head difference between two points of fluid flow, and L represents the percolation path length;
s33: and randomly scrambling all the data, and then carrying out hierarchical sampling to obtain an expanded data set.
Optionally, the step S4 specifically includes:
s41: connectivity analysis is carried out on the expanded data set, and isolated and non-connected hole points are removed;
s42: calculating a three-dimensional pore network structure of the core through a network pore model, extracting the length, the connected throat holes, the average radius and the area data of the throat of the core as attribute characteristics of edges, and extracting the relative position, coordination number, average radius and area size of the throat of the core as attribute characteristics of nodes;
s43: constructing a space-time diagram according to a three-dimensional network pore structure of a coreFirstly, starting from a core boundary in a permeation calculation direction, then gradually scanning a pore network structure in an area and accumulating point-edge sets, and recording the accumulated point-edge sets to a space-time diagram every time a fixed length passes>Finally, the set of all scans is +.>As a space-time diagram->The data, relational formula is:
in the method, in the process of the invention,representing space-time diagram data, ">Time-space diagram data indicating the jth scan, and t indicates the number of scans.
Optionally, the step S43 specifically includes:
s431: extracting a space-time diagramFirstly, starting according to the permeability calculation direction, scanning three-dimensional boundary from a rock core, and then carrying out scanning on the scanned nodes together with +.>Node addition to space-time diagram +.>In the middle, if the time-space diagram is->Connected with the pore edge of the newly added node, the time-space diagram is +.>Is used as the starting point of the edge, and the newly scanned node is used as the end point of the edge until the whole space-time diagram is generated +.>Wherein->A set of time-space diagrams representing t times of the whole scan,/->Space-time diagram representing the i-1 th scan,/->A space-time diagram representing the ith scan;
s432: extracting a space-time diagramNode attribute characteristics such that the relative position, coordination number, average radius and area size of the laryngeal aperture of the network pore structure of step S42 are extracted as a space-time diagram +.>Node attribute characteristics;
s433: extracting a space-time diagramThe edge attribute characteristics are such that the length, the continuous throat holes, the average radius and the area of the throat of the network pore structure of the step S4 are extracted as a space-time diagram +.>Edge attribute characteristics;
s434: time space diagramCarrying out normalization operation on nodes and edges of the graph;
s435: generating different adjacency matricesSo that normalized time-space diagram +.>Edge attribute feature generation ++>The relational formula is:
in the method, in the process of the invention,representing a set of space-time diagram adjacency matrices, +.>Adjacency matrix representing the ith scan, +.>The number of nodes in the space-time diagram in the ith scan is shown, and t is the number of scans.
Optionally, the step S5 specifically includes:
s51: space-time diagramThe data input is based on a graphic neural network layer of a multi-query attention mechanism, and spatial features are extracted;
s52: the spatial features extracted by the graphic neural network layer based on the multi-query attention mechanism are input to the gating logic unit time sequence network layer to extract time sequence features, so that fused time-space feature data are obtained;
s53: and (3) passing the space-time characteristic data through a multi-layer perceptron to obtain the predicted three-dimensional core permeability.
Optionally, the step S51 specifically includes:
s511: calculating a space-time diagramImportance scores of neighbors of the medium nodes, calculation of multi-query attention mechanism is performed on each node, and the importance scores are calculated by adjacency matrix +.>The method comprises the steps of obtaining a current node degree vector as query, and obtaining importance scores of neighbor nodes by using a node feature matrix as a key and a value, wherein the multi-query attention scoring formula is as follows:
in the method, in the process of the invention,score representing multiple query attention mechanisms of node u,/->Representing tensor stitching function,/->Representing a space-time diagram->Lower u node edge attribute->Score of each attention mechanism, +.>Representing each of/>Weight value of score->，/>，/>Respectively representing query, key and value, +.>Representing normalized exponential function, ++>Representing the dimension of the input vector;
s512: updating the graph characterization node of the current network layer, sampling d nodes with highest scores by using a graph sampling aggregation mechanism, aggregating the scores and the nodes to generate node characteristics of the current network layer, wherein a related formula of information propagation is as follows:
where Z represents the multi-query attention mechanism score for all nodes v,representing all the neighboring nodes of node v,
where M represents the top d maximum scoring node sets,node representing the score, < >>Representing the number of sets Z, +.>Meaning greater than arbitrary->The scoring element of the individual nodes, the first d scores of the maximum score,
in the method, in the process of the invention,representing hidden state of layer i node v, sigma represents activation function, < >>Represents a mean function>Representing the hidden state of the layer 1 node v,/->Is the hidden state of node u of layer l-1.
The invention has the beneficial effects that:
(1) The three-dimensional core permeability prediction algorithm flow designed by the invention applies the latest graph neural network leading edge technology in the deep learning field, reduces the expensive and time-consuming experimental cost compared with the traditional permeability prediction method, fuses the physical knowledge constraint of the core, and increases the interpretability and scientificity of the related field;
(2) Adopting a multi-query attention mechanism strategy, collecting pore structure information with larger surrounding important influence degree, carrying out a causal convolution simulation fluid permeation process on time sequence, combining a time sequence neural network to store fluid propagation information, and fitting a permeability principle, so that the reliability of results is improved;
(3) The designed graph neural network structure meets the actual scene requirement, and compared with the operation and prediction errors generated by a large amount of redundant space information by a common deep learning model, the graph neural network structure has the advantages that physical factors which actually influence the permeability are more emphasized, and the graph neural network structure can be applied to the prediction of the rock permeability in a large scale under the same resource, and meets the actual market requirement.
Drawings
Fig. 1 is a flow chart of a three-dimensional core permeability prediction method based on a space-time diagram neural network in an embodiment of the invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The embodiment of the invention provides a three-dimensional core permeability prediction method based on a space-time diagram neural network, and referring to fig. 1, fig. 1 is a flow diagram of an embodiment of the three-dimensional core permeability prediction method based on a space-time diagram neural network.
In this embodiment, the three-dimensional core permeability prediction method based on the space-time diagram neural network includes the following steps:
s1, performing X-ray mu CT scanning on a core sample, and reconstructing a three-dimensional medium of an image sequence to obtain core three-dimensional data;
s2, preprocessing the three-dimensional data of the core, filtering data noise, enhancing the contrast ratio between the entity of the core and the pore space, executing automatic threshold segmentation, separating the entity inside the core from the pore space structure, and converting the entity inside the core and the pore space structure into binary three-dimensional data;
s3, slicing the binarized three-dimensional data body, namely slicing the original three-dimensional data body from any one axial direction of X/Y/Z to obtain three-dimensional core cubes, calculating core permeability as a data label according to a Darcy theorem formula for each three-dimensional core cube, and then obtaining an expansion data set;
s4, extracting network pore structure data of each three-dimensional rock core in the expansion data set, and generating space-time diagram representation data;
s5, inputting the space-time diagram representation data into the constructed space-time diagram neural network model, firstly extracting the spatial characteristics of the core based on the diagram neural network of the multi-query attention mechanism, then entering a time sequence network layer of a gating logic unit to capture the time sequence relation of the core, and finally predicting the permeability through a multi-layer perceptron.
Specifically, the specific implementation principle of three-dimensional core permeability prediction based on the space-time diagram neural network is described in detail below.
In this example, the step S1 includes the following sub-steps:
s11, performing registration cutting on a two-dimensional image slice sequence scanned by X-ray mu CT, firstly aligning two-dimensional image slices in the same coordinate system, then determining the direction information of the image sequence, and finally cutting to the same size;
s12, converting the two-dimensional image slice sequence after registration cutting into voxel three-dimensional data;
and S13, carrying out three-dimensional reconstruction on the voxelized three-dimensional data, and combining all voxels together to generate three-dimensional reconstruction data of the internal structure of the core.
In this example, the step S2 includes the following sub-steps:
s21, performing Gaussian smoothing filtering on the obtained preliminary three-dimensional data of the core, and performing convolution operation by using a Gaussian filter of 7 x 7 to obtain the smoothed three-dimensional data of the core;
s22, filtering and denoising the smoothed three-dimensional data of the rock core, creating a noise template according to the type and the intensity of the noise, and applying the noise template to the three-dimensional data of the rock core to obtain the three-dimensional data after denoising;
s23, carrying out histogram equalization on the three-dimensional data after noise reduction, firstly calculating a histogram of the three-dimensional data after noise reduction to obtain a cumulative distribution function, then carrying out gray level mapping through a mapping function, and finally carrying out gamma transformation function adjustment to obtain core three-dimensional data after contrast enhancement;
and S24, performing automatic threshold segmentation on the core three-dimensional data after contrast enhancement, firstly normalizing a data gray level distribution histogram, then calculating probability distribution of gray levels, entropy of each gray level and entropy of left and right sides, secondly calculating weighted average entropy of each gray level, finally finding out a gray level corresponding to the maximum value, and binarizing the image by using the gray level as a threshold value to obtain binarized three-dimensional data.
Further, the noise reduction calculation formula is as follows:
wherein g (x, y, z) is a pixel value in the output image, represents a pixel value after noise reduction, f (x, y, z) is a pixel value in the input image, represents an original pixel value, λ is an over-parameter, and represents the intensity of noise reduction;
the formula of the histogram equalization is as follows:
wherein the number of pixels of the jth gray level in the original image is nj, the total number of pixels is N, the cumulative distribution function is CDF, and the mapping function is MAP,representing the sum of the number of pixels from the 0 th gray level to the i th gray level, L being the number of gray levels;
the formula of the gamma transformation function is as follows:
wherein x represents a pixel value in the original data, and Γ represents a gamma value;
the automated thresholding formula is as follows:
where p (i) represents a probability distribution for each gray level, h (i) represents the number of pixels of the i-th gray level, and N represents the total number of pixels of the data;
where H (i) represents entropy of each gray level, i represents gray level;
wherein HL (t) represents left entropy, i ranges from 0 to t-1;
where HR (t) represents right entropy and i ranges from t to 255;
where Hv (t) represents the weighted average entropy and t represents the current gray level;
where t represents an optimal threshold, where t ranges from 0 to 255.
In this example, the step S3 includes the following sub-steps:
s31, providing a random segmentation method, taking CST (China Standard Time) milliseconds as random seeds, setting random starting point positions, segmentation size and stepping length of a sliding window, and carrying out sliding segmentation on an original three-dimensional data body in any axial direction of X/Y/Z to obtain a plurality of three-dimensional rock core cube data with smaller volumes;
s32, calculating the core permeability according to a Darcy theorem formula, wherein the mathematical formula is expressed as follows:
where Q is the rate of fluid through the porous medium, k is the permeability coefficient of the porous medium, A is the cross-sectional area, Δh is the head difference between the two points of fluid flow, and L is the percolation path length;
s33, randomly disturbing all data, and performing hierarchical sampling to obtain an expanded data set.
In this example, the step S4 includes the following sub-steps:
s41, performing connectivity analysis on the expanded data set, and removing isolated and non-connected hole points;
s42, calculating through a network pore model to obtain a three-dimensional pore network structure of the core, extracting the length, the connected throat holes, the average radius and the area data of the throat of the core as attribute characteristics of edges, and extracting the relative position, coordination number, average radius and area size of the throat of the core as attribute characteristics of nodes;
s43, constructing a space-time diagram according to the three-dimensional network pore structure of the coreFirstly, starting from a core boundary in a permeation calculation direction, then gradually scanning a pore network structure in an area and accumulating point-edge sets, and recording the accumulated point-edge sets to a space-time diagram every time a fixed length passes>Finally, the set of all scans is +.>As a space-time diagram->The data, relational formula is:
in the method, in the process of the invention,representing space-time diagram data, ">Time-space diagram data indicating the jth scan, and t indicates the number of scans.
In this example, the step S43 includes the following sub-steps:
s431, extracting space-time diagramFirstly, starting according to the permeability calculation direction, scanning three-dimensional boundary from a rock core, and then carrying out scanning on the scanned nodes together with +.>Node addition to space-time diagram +.>In the middle, if the time-space diagram is->Connected with the pore edge of the newly added node, the time-space diagram is +.>Is used as the starting point of the edge, and the newly scanned node is used as the end point of the edge until the whole space-time diagram is generated +.>Wherein->A set of time-space diagrams representing t times of the whole scan,/->Space-time diagram representing the i-1 th scan,/->A space-time diagram representing the ith scan;
s432, extracting space-time diagramNode attribute characteristics such that the relative position, coordination number, average radius and area size of the laryngeal aperture of the network pore structure of step S42 are extracted as a space-time diagram +.>Node attribute characteristics;
s433, extracting space-time diagramThe edge attribute characteristics are such that the length, the continuous throat holes, the average radius and the area of the throat of the network pore structure of the step S4 are extracted as a space-time diagram +.>Edge attribute characteristics;
s434, time space diagramCarrying out normalization operation on nodes and edges of the graph;
s435, generating different adjacency matrixesSo that normalized time-space diagram +.>Edge attribute feature generation ++>The relational formula is:
in the method, in the process of the invention,representing a set of space-time diagram adjacency matrices, +.>Adjacency matrix representing the ith scan, +.>The number of nodes in the space-time diagram in the ith scan is shown, and t is the number of scans.
In this example, the step S5 includes the following sub-steps:
s51, space-time diagram is formedThe data input is based on a graphic neural network layer of a multi-query attention mechanism, and spatial features are extracted;
s52, inputting the spatial features extracted by the graphic neural network layer based on the multi-query attention mechanism into a time sequence network layer of a gating logic unit to extract time sequence features, so as to obtain fused space-time feature data;
and S53, passing the space-time characteristic data through a multi-layer perceptron to obtain the predicted three-dimensional core permeability.
In this example, the step S51 includes the following sub-steps:
s511, calculating a space-time diagramImportance scores of neighbors of the medium nodes, calculation of multi-query attention mechanism is performed on each node, and the importance scores are calculated by adjacency matrix +.>The method comprises the steps of obtaining a current node degree vector as query, and obtaining importance scores of neighbor nodes by using a node feature matrix as a key and a value, wherein the multi-query attention scoring formula is as follows:
in the method, in the process of the invention,score representing multiple query attention mechanisms of node u,/->Representing tensor stitching function,/->Representing a space-time diagram->Lower u node edge attribute->Score of each attention mechanism, +.>Representing->Weight value of score->，/>，/>Respectively representing query, key and value, +.>Representing normalized exponential function, ++>Representing the dimensions of the input vector.
S512, updating graph characterization nodes of the current network layer, sampling d nodes with highest scores by using a graph sampling aggregation mechanism, aggregating the scores with the nodes, and generating node characteristics of the current network layer, wherein a related formula of information propagation is as follows:
where Z represents the multi-query attention mechanism score for all nodes v,representing all the neighboring nodes of node v,
where M represents the top d maximum scoring node sets,nodes representing the score, |Z| represents the number of sets Z, |N ++>Meaning greater than arbitrary->The scoring element of the individual nodes, the first d scores of the maximum score,
in the method, in the process of the invention,representing hidden state of layer i node v, sigma represents activation function, < >>Represents a mean function>Representing the hidden state of the layer 1 node v,/->Is the hidden state of node u of layer l-1.
Therefore, the three-dimensional core permeability prediction method based on the space-time diagram neural network is provided, the latest diagram neural network front technology in the deep learning field is applied, the multi-query attention mechanism strategy is adopted, pore structure information with larger surrounding important influence degree is collected, the space-time characteristics of the core are extracted, the physical knowledge constraint of the core is fused, the interpretability and scientificity of the related field are improved, compared with the conventional machine learning model, the physical factors which actually influence the permeability are more focused, the method can be applied to the prediction of the rock permeability in a large scale under the same resource, and the market actual demand is more met.
It is appreciated that in the description herein, reference to the terms "one embodiment," "another embodiment," "other embodiments," or "first through nth embodiments," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (5)
1. The three-dimensional core permeability prediction method based on the space-time diagram neural network is characterized by comprising the following steps of:
s1: performing X-ray mu CT scanning on the core sample, and reconstructing a three-dimensional medium of an image sequence to obtain core three-dimensional data;
s2: preprocessing three-dimensional data of a core, filtering data noise, enhancing contrast between a core entity and a pore, executing automatic threshold segmentation, separating an internal entity of the core from a pore structure, and converting the internal entity of the core and the pore structure into binarized three-dimensional data;
s3: cutting the three-dimensional data of the core from any axial direction of X/Y/Z to obtain three-dimensional core cubes, and calculating core permeability of each three-dimensional core cube according to a Darcy theorem formula to obtain an extended data set as a data label;
s4: extracting network pore structure data of each three-dimensional rock core in the extended data set, and generating space-time diagram representation data; the step S4 specifically includes:
s41: connectivity analysis is carried out on the expanded data set, and isolated and non-connected hole points are removed;
s42: calculating a three-dimensional pore network structure of the core through a network pore model, extracting the length, the connected throat holes, the average radius and the area data of the throat of the core as attribute characteristics of edges, and extracting the relative position, coordination number, average radius and area size of the throat of the core as attribute characteristics of nodes;
s43: constructing a space-time diagram according to a three-dimensional network pore structure of a coreFirstly, starting from a core boundary in a permeation calculation direction, then gradually scanning a pore network structure in an area and accumulating point-edge sets, and recording the accumulated point-edge sets to a space-time diagram every time a fixed length passes>Finally, the set of all scans is +.>As a space-time diagram->The data, relational formula is:
in the method, in the process of the invention,representing space-time diagram data, ">Space-time diagram data representing the jth scan, t representing the number of scans;
the step S43 specifically includes:
s431: extracting a space-time diagramFirstly, starting according to the permeability calculation direction, scanning three-dimensional boundary from a rock core, and then carrying out scanning on the scanned nodes together with +.>Node addition to space-time diagram +.>In the middle, if the time-space diagram is->Connected with the pore edge of the newly added node, the time-space diagram is +.>As the start of an edge, the newly scanned node as the end of an edge until the entire space-time diagram is generatedWherein->A set of time-space diagrams representing t times of the whole scan,/->Space-time diagram representing the i-1 th scan,/->A space-time diagram representing the ith scan;
s432: extracting a space-time diagramNode attribute characteristics such that the relative position, coordination number, average radius and area size of the laryngeal aperture of the network pore structure of step S42 are extracted as a space-time diagram +.>Node attribute characteristics;
s433: extracting a space-time diagramThe edge attribute characteristics are such that the length, the continuous throat holes, the average radius and the area of the throat of the network pore structure of the step S4 are extracted as a space-time diagram +.>Edge attribute characteristics;
s434: time space diagramCarrying out normalization operation on nodes and edges of the graph;
s435: generating different adjacency matricesSo that normalized time-space diagram +.>Edge attribute feature generation ++>The relational formula is:
in the method, in the process of the invention,representing a set of space-time diagram adjacency matrices, +.>Adjacency matrix representing the ith scan, +.>The real number domain is represented by the number,the number of nodes of the space-time diagram in the ith scanning is represented, and t represents the number of scanning times;
s5: the method comprises the steps of inputting space-time diagram representation data into a constructed space-time diagram neural network model, firstly extracting space characteristics of a rock core based on a diagram neural network of a multi-query attention mechanism, then entering a sequential network layer of a gating logic unit to capture a sequential relationship of the rock core, and finally predicting permeability through a multi-layer perceptron;
the step S5 specifically includes:
s51: space-time diagramThe data input is based on a graphic neural network layer of a multi-query attention mechanism, and spatial features are extracted;
s52: the spatial features extracted by the graphic neural network layer based on the multi-query attention mechanism are input to the gating logic unit time sequence network layer to extract time sequence features, so that fused time-space feature data are obtained;
s53: the space-time characteristic data is processed by a multi-layer perceptron to obtain predicted three-dimensional core permeability;
the step S51 specifically includes:
s511: calculating a space-time diagramImportance scores of neighbors of the medium nodes, calculation of multi-query attention mechanism is performed on each node, and the importance scores are calculated by adjacency matrix +.>The method comprises the steps of obtaining a current node degree vector as query, and obtaining importance scores of neighbor nodes by using a node feature matrix as key and value, wherein a scoring formula of a multi-query attention mechanism is as follows:
in the method, in the process of the invention,a score representing the multi-query attention mechanism of node u, multi-query attention (·) representing the multi-query attention algorithm, +_j->Representing tensor stitching function,/->Representing a space-time diagram->Lower u node edge attribute->Score of each attention mechanism, +.>Representing->Weight value of score->，/>，/>Respectively representing a query, a key and a value,representing normalized exponential function, ++>Representing the dimension of the input vector;
s512: updating the graph characterization node of the current network layer, sampling d nodes with highest scores by using a graph sampling aggregation mechanism, aggregating the scores and the nodes to generate node characteristics of the current network layer, wherein a related formula of information propagation is as follows:
wherein Z represents multiple views of all nodes vThe polling attention mechanism score is a score,all neighbor nodes, z, representing node v i A score representing a multi-query attention mechanism;
where M represents the top d maximum scoring node sets,node representing the score, < >>Representing the number of sets Z, +.>Meaning greater than arbitrary->Scoring elements of individual nodes, i.e. d scores before the maximum score, u i A corresponding node representing a multi-query attention mechanism score;
in the method, in the process of the invention,representing hidden state of layer i node v, sigma represents activation function, < >>Represents a mean function>Representing the hidden state of the layer 1 node v,/->Is the hidden state of node u of layer l-1, W represents a learnable weight.
2. The three-dimensional core permeability prediction method based on a space-time diagram neural network as set forth in claim 1, wherein the step S1 specifically includes:
s11: registering and cutting the two-dimensional image slice sequence scanned by the X-ray μCT, firstly aligning the two-dimensional image slices in the same coordinate system, then determining the direction information of the image sequence, and finally cutting to the same size;
s12: converting the two-dimensional image slice sequence after registration cutting into voxel three-dimensional data;
s13: and carrying out three-dimensional reconstruction on the voxelized three-dimensional data, and combining all voxels together to generate three-dimensional reconstruction data of the internal structure of the core.
3. The three-dimensional core permeability prediction method based on a space-time diagram neural network as set forth in claim 1, wherein the step S2 specifically includes:
s21: carrying out Gaussian smoothing filtering on the obtained three-dimensional data of the core, and carrying out convolution operation by using a Gaussian filter of 7 x 7 to obtain the smoothed three-dimensional data of the core;
s22: filtering and denoising the smoothed three-dimensional data of the rock core, creating a noise template according to the type and the intensity of the noise, and applying the noise template to the three-dimensional data of the rock core to obtain the three-dimensional data after denoising;
s23: performing histogram equalization on the three-dimensional data after noise reduction, firstly calculating a histogram of the three-dimensional data after noise reduction to obtain a cumulative distribution function, then performing gray level mapping through a mapping function, and finally performing gamma transformation function adjustment to obtain core three-dimensional data after contrast enhancement;
s24: and performing automatic threshold segmentation on the core three-dimensional data after contrast enhancement, firstly normalizing a data gray level distribution histogram, then calculating probability distribution of gray levels, entropy of each gray level and entropy of left and right sides, secondly calculating weighted average entropy of each gray level, finally finding out a gray level corresponding to the maximum value, and binarizing an image by using the gray level as a threshold value to obtain binarized three-dimensional data.
4. The three-dimensional core permeability prediction method based on the space-time diagram neural network as set forth in claim 3, wherein:
the noise reduction processing formula is as follows:
in the method, in the process of the invention,is the pixel value in the output image, representing the pixel value after noise reduction, < >>Is a pixel value in the input image, representing the original pixel value, λ is a super-parameter, representing the intensity of noise reduction;
the formula of the histogram equalization is as follows:
in the method, in the process of the invention,representing cumulative distribution function->Representing a mapping function->The number of pixels representing the jth gray level in the original image, N representing the total number of pixels,/and->Represents the sum of the number of pixels from the 0 th gray level to the i th gray level, L represents the number of gray levels, ">Representing a numerical processing function for rounding floating point numbers;
the formula of the gamma transformation function is as follows:
in the method, in the process of the invention,representing a mapping relation of gamma transformation, wherein x represents a pixel value in original data, and Γ represents a gamma value;
the automated thresholding formula is as follows:
in the method, in the process of the invention,representing the probability distribution of each gray level, +.>A pixel number representing an ith gray level, N representing a total pixel number of data;
in the method, in the process of the invention,an entropy representing each gray level, i representing the gray level;
in the method, in the process of the invention,representing left entropy, i ranges from 0 to t-1;
in the method, in the process of the invention,representing right entropy, i ranges from t to 255;
in the method, in the process of the invention,representing a weighted average entropy, t representing a current gray level;
where t represents an optimal threshold, where t ranges from 0 to 255.
5. The three-dimensional core permeability prediction method based on a space-time diagram neural network as set forth in claim 1, wherein the step S3 specifically includes:
s31: taking Chinese standard time as a random seed, randomly setting a starting point position, a segmentation size and a stepping length of a sliding window, sliding and segmenting an original three-dimensional data body in any axial direction by X/Y/Z, firstly, sliding the sliding window from the random starting point position, moving the sliding window by the set stepping length, then axially intercepting data in the direction according to the set segmentation size, and finally obtaining a data set;
s32: according to the Darcy theorem formula, core permeability is calculated for the data set, and the mathematical formula is expressed as follows:
wherein Q represents the rate of fluid passing through the porous medium, k represents the permeability coefficient of the porous medium, a represents the cross-sectional area, Δh represents the head difference between two points of fluid flow, and L represents the percolation path length;
s33: and randomly scrambling all the data, and then carrying out hierarchical sampling to obtain an expanded data set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410097446.4A CN117611485B (en) | 2024-01-24 | 2024-01-24 | Three-dimensional core permeability prediction method based on space-time diagram neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410097446.4A CN117611485B (en) | 2024-01-24 | 2024-01-24 | Three-dimensional core permeability prediction method based on space-time diagram neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117611485A CN117611485A (en) | 2024-02-27 |
| CN117611485B true CN117611485B (en) | 2024-04-02 |
Family
ID=89953904
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410097446.4A Active CN117611485B (en) | 2024-01-24 | 2024-01-24 | Three-dimensional core permeability prediction method based on space-time diagram neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117611485B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101556703A (en) * | 2009-05-16 | 2009-10-14 | 中国石油大学(华东) | Method for establishing network model based on serial section image |
| CN103492874A (en) * | 2011-03-04 | 2014-01-01 | 兰德马克绘图国际公司 | Systems and methods for determining fluid mobility in rock samples |
| CN110135311A (en) * | 2019-05-06 | 2019-08-16 | 重庆科技学院 | A kind of hole based on three-dimensional Core Scanning Image and pore throat identifying system and method |
| CA3062555A1 (en) * | 2018-07-18 | 2020-01-18 | China University Of Mining And Technology | Porous medium permeability prediction method based on machine image intelligent learning |
| CN112070774A (en) * | 2020-09-16 | 2020-12-11 | 西南石油大学 | Neural network optimization method for shale digital core image segmentation |
| CN114547978A (en) * | 2022-02-22 | 2022-05-27 | 中国石油大学(华东) | Heterogeneous graph neural network-based multi-layer system water injection amount splitting method and system |
| CN114819096A (en) * | 2022-05-10 | 2022-07-29 | 深延科技(北京)有限公司 | Model training method and device, electronic equipment and storage medium |
| CN116152259A (en) * | 2023-04-23 | 2023-05-23 | 西南石油大学 | Reservoir permeability calculation method based on graph neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11409015B2 (en) * | 2020-06-12 | 2022-08-09 | Saudi Arabian Oil Company | Methods and systems for generating graph neural networks for reservoir grid models |
| US20230097859A1 (en) * | 2021-09-30 | 2023-03-30 | Saudi Arabian Oil Company | Method and system for determining coarsened grid models using machine-learning models and fracture models |
-
2024
- 2024-01-24 CN CN202410097446.4A patent/CN117611485B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101556703A (en) * | 2009-05-16 | 2009-10-14 | 中国石油大学(华东) | Method for establishing network model based on serial section image |
| CN103492874A (en) * | 2011-03-04 | 2014-01-01 | 兰德马克绘图国际公司 | Systems and methods for determining fluid mobility in rock samples |
| CA3062555A1 (en) * | 2018-07-18 | 2020-01-18 | China University Of Mining And Technology | Porous medium permeability prediction method based on machine image intelligent learning |
| CN110135311A (en) * | 2019-05-06 | 2019-08-16 | 重庆科技学院 | A kind of hole based on three-dimensional Core Scanning Image and pore throat identifying system and method |
| CN112070774A (en) * | 2020-09-16 | 2020-12-11 | 西南石油大学 | Neural network optimization method for shale digital core image segmentation |
| CN114547978A (en) * | 2022-02-22 | 2022-05-27 | 中国石油大学(华东) | Heterogeneous graph neural network-based multi-layer system water injection amount splitting method and system |
| CN114819096A (en) * | 2022-05-10 | 2022-07-29 | 深延科技(北京)有限公司 | Model training method and device, electronic equipment and storage medium |
| CN116152259A (en) * | 2023-04-23 | 2023-05-23 | 西南石油大学 | Reservoir permeability calculation method based on graph neural network |
Non-Patent Citations (5)
| Title |
|---|
| NTformer: Near Time Transformer for multi-step prediction of wellbead pressure in facturing operations;Yuan Zhong;《Geoenergy Science and Engineering》;20231231;第231卷(第2023期);1-16 * |
| Rapid estimation of pemeability from digital rock using 3D convollution neural network;Jin Hong等;《Computational Geosciences》;20200527;第24卷;1523-1539 * |
| 基于产出序列时空特征的图神经网络煤层气产量预测研究;高云云;《中国优秀硕士学位论文全文数据库 (基础科学辑)》;20230215(第2023（02）期);A011-919 * |
| 邬光辉等.四川盆地川中古隆起北斜坡震旦系走滑断裂分布及其勘探意义.《周刚 * |
| 钟原 天然气工业》.2023,第43卷(第07期),35-43. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117611485A (en) | 2024-02-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086773B (en) | Fault plane identification method based on full convolution neural network | |
| Zhu et al. | Intelligent logging lithological interpretation with convolution neural networks | |
| Monegaglia et al. | Automated extraction of meandering river morphodynamics from multitemporal remotely sensed data | |
| CN112232391B (en) | Dam crack detection method based on U-net network and SC-SAM attention mechanism | |
| CN108052966A (en) | Remote sensing images scene based on convolutional neural networks automatically extracts and sorting technique | |
| Canaz Sevgen et al. | An improved RANSAC algorithm for extracting roof planes from airborne lidar data | |
| Zhao et al. | Extracting buildings from and regularizing boundaries in airborne lidar data using connected operators | |
| CN111985552B (en) | Method for detecting diseases of thin strip-shaped structure of airport pavement under complex background | |
| CN111582093A (en) | Automatic small target detection method in high-resolution image based on computer vision and deep learning | |
| Geman et al. | Locating texture and object boundaries | |
| Chen et al. | A new process for the segmentation of high resolution remote sensing imagery | |
| CN113505670B (en) | Remote sensing image weak supervision building extraction method based on multi-scale CAM and super-pixels | |
| CN115205692B (en) | Typical feature intelligent identification and extraction method based on generation of countermeasure network | |
| CN114296075A (en) | Ground penetrating radar image artificial intelligence identification method and device | |
| Valliammal et al. | A novel approach for plant leaf image segmentation using fuzzy clustering | |
| CN110363178B (en) | Airborne laser point cloud classification method based on local and global depth feature embedding | |
| CN112836614B (en) | High-resolution remote sensing image classification method based on residual error network and transfer learning | |
| CN114913434A (en) | High-resolution remote sensing image change detection method based on global relationship reasoning | |
| CN115019163A (en) | City factor identification method based on multi-source big data | |
| CN113689453A (en) | Method, device and equipment for automatically identifying well logging image cracks and storage medium | |
| CN112784777A (en) | Unsupervised hyperspectral image change detection method based on antagonistic learning | |
| CN117611485B (en) | Three-dimensional core permeability prediction method based on space-time diagram neural network | |
| CN116258877A (en) | Land utilization scene similarity change detection method, device, medium and equipment | |
| CN116051966A (en) | Pen and stone image recognition method based on deep learning network and model training method thereof | |
| CN113506253B (en) | Sandstone pore detection method based on multilayer multi-core learning and region combination |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |