CN116245013A - Geological prediction model construction method, modeling method, equipment and storage medium - Google Patents
Geological prediction model construction method, modeling method, equipment and storage medium Download PDFInfo
- Publication number
- CN116245013A CN116245013A CN202211710956.1A CN202211710956A CN116245013A CN 116245013 A CN116245013 A CN 116245013A CN 202211710956 A CN202211710956 A CN 202211710956A CN 116245013 A CN116245013 A CN 116245013A
- Authority
- CN
- China
- Prior art keywords
- geological
- map
- graph
- node
- deep learning
- 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.)
- Pending
Links
- 238000010276 construction Methods 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 title claims description 47
- 238000012549 training Methods 0.000 claims abstract description 77
- 238000013135 deep learning Methods 0.000 claims abstract description 60
- 230000006870 function Effects 0.000 claims abstract description 25
- 238000000605 extraction Methods 0.000 claims abstract description 23
- 239000013598 vector Substances 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000013507 mapping Methods 0.000 claims abstract description 15
- 230000002776 aggregation Effects 0.000 claims abstract description 7
- 238000004220 aggregation Methods 0.000 claims abstract description 7
- 238000005070 sampling Methods 0.000 claims description 32
- 238000012795 verification Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 12
- 230000015654 memory Effects 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 4
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 2
- 238000004590 computer program Methods 0.000 description 8
- 238000013136 deep learning model Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000000873 masking effect Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 241001192665 Anous Species 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000003673 groundwater Substances 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000003208 petroleum Substances 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/04—Constraint-based CAD
Abstract
The invention provides a construction method of a map deep learning geological prediction model, which comprises the following steps: s1, constructing a sample data set, and carrying out graph structuring processing on the sample data set to form a geological node graph structural grid as a training data set; s2, constructing a graph deep learning geological prediction model based on a GNN-transporter model, wherein the graph deep learning geological prediction model comprises a GNN module, a transporter geological element graph global feature extraction module and a fully-connected classification mapping module which are sequentially connected, the GNN module generates embedded vectors for all graph nodes, and the transporter geological element graph global feature extraction module is used for further encoding the embedded vectors output by the GNN module; s3, a loss function is set, a map deep learning geological prediction model is trained, spatial correlation among geological nodes is intuitively constructed through a map structure, spatial relations of neighbor geological nodes are represented through GNN aggregation, a transform module is used for effectively excavating a spatial distribution mode of geological elements, and local and global spatial correlation of the geological elements can be considered.
Description
Technical Field
The invention belongs to the field of mapping science and technology, and particularly relates to a construction method of a map deep learning geological prediction model, a three-dimensional geological modeling method, computer equipment and a storage medium.
Background
Three-dimensional geologic models which are required to accurately represent underground geologic structures are used as supports for geologic environment evaluation, deep geologic resource exploration (ore, groundwater, petroleum, natural gas and the like), underground space development and utilization, large-scale engineering construction and the like, and the three-dimensional geologic models become important bases for developing geologic space analysis, geologic phenomenon interpretation and geologic process simulation. The three-dimensional geologic model is a mathematical model reflecting the distribution characteristics of geologic elements in a three-dimensional space, combines tools such as space information management, geologic interpretation, space analysis and prediction, geoscience statistics, graphic visualization and the like through computer technology, and is used for geologic analysis. Compared with two-dimensional geological mapping achievements widely used in traditional geological work, the three-dimensional geological model can completely and intuitively reflect the shape and space relation of a geological structure in a three-dimensional space, and the model achievements are more accurate and reasonable.
At present, three-dimensional geological modeling methods have been developed in a wide variety, and the three-dimensional geological modeling methods can be divided into an explicit modeling method and an implicit modeling method from the construction process of a model. The explicit modeling method requires a large amount of manual interaction, modeling results depend on expert interpretation, reproducibility is poor, automation degree is low, and artificial modeling topology errors are easy to occur. Implicit geologic modeling methods implicitly mine spatially distributed features from geologic data, solve for related geologic modeling parameters, but are largely limited by the quality of the geologic data and the complexity of the geologic environment, especially in geologic structure environments that face high variability, face sparse and uneven geologic sampling, and are prone to over-fitting, yielding unreasonable geologic modeling.
In recent years, an artificial intelligence method opens up a new development direction for a three-dimensional geological modeling technology. The intelligent three-dimensional geological modeling method based on deep learning is researched, the three-dimensional geological modeling is developed to be intelligent, hidden geological features are further excavated from massive geological data, anisotropic expression of geological structures is achieved, and understanding of geological processes is greatly facilitated. However, the current geological modeling method based on deep learning and machine learning is difficult to consider the correlation between geological local space and global space, is limited by a sample data set and a deep learning model, is difficult to effectively learn and extract a geological element distribution characteristic mode expected by geological specialists, and is difficult to generate an accurate and reasonable geological model; and limited by the variety of available geologic data, it is difficult to fully constrain existing geologic sampling data in practical geologic modeling projects.
Disclosure of Invention
The invention provides a construction method of a map deep learning geological prediction model and a three-dimensional geological modeling method based on the model, so as to solve the problems that the existing three-dimensional geological modeling method is difficult to consider the correlation of local and global spaces of geological elements and difficult to generate geological modeling with good quality under the constraint of limited geological data.
In order to achieve the above and other related objects, the present invention provides a method for constructing a map deep learning geological prediction model, comprising the following steps:
s1, constructing a sample data set, and carrying out graph structuring processing on the sample data set to form a geological node graph structural grid as a training data set, wherein the training data set comprises training data, verification data and test data;
s2, constructing a graph deep learning geological prediction model based on a GNN-Transformer model, wherein the graph deep learning geological prediction model comprises a GNN module, a Transformer geological element graph global feature extraction module and a fully-connected classification mapping module which are sequentially connected, the GNN module is used for generating embedded vectors for all graph nodes, and the Transformer geological element graph global feature extraction module is used for further encoding the embedded vectors output by the GNN module;
s3, setting a loss function for the map deep learning geological prediction model, and training the map deep learning geological prediction model based on the training data set and the loss function, wherein the training is performed based on the training data to achieve parameter updating, and when the loss function on the verification data is not lowered any more, the training is stopped, so that the trained map deep learning geological prediction model is obtained.
Preferably, in the step S1, performing a graph structuring process on the sample data set to form a geological node graph structural grid includes the following steps:
s11, regularly sampling the sample data set, wherein the step of sampling comprises the following steps: dispersing a geological modeling sample into space scattered points, carrying out category coding according to the prior knowledge of the existing geology and the stratum age from new to old and the natural number from small to large, wherein each stratum category corresponds to a natural number i, and taking a stratum category corresponding value i of a graph node as a graph node characteristic;
s12, constructing a preliminary geological node grid: matching the scattered points in the S11 to the positions of grid nodes on the regular grid units with preset sizes, and endowing the stratum category corresponding values i of the scattered points with corresponding grid nodes so as to construct a preliminary geological node grid;
s13, setting a virtual geological structure in the preliminary geological node grid, reserving original attributes of grid nodes related to the virtual geological structure in the geological node grid to serve as known data, performing mask masking processing on the attributes of other grid nodes to serve as test data, and regarding most of the grid nodes as training data and the rest as verification data according to a preset proportion for the known data;
s14, performing triangulation processing on the regular grid nodes, and constructing edge connection among the grid nodes, so that each grid node is positioned on the graph structure to obtain a final geological node graph structure grid.
Preferably, the construction of the map deep learning geological prediction model comprises the following steps:
s21, constructing a GNN module, and based on graph neighborhood information of the training data set, generating embedded vectors for each graph node, wherein the embedded vectors are used for representing the spatial relationship among geological elements in a modeling space;
s22, constructing a global feature extraction module of a transducer geological element map, and further encoding an embedded vector output by the GNN module based on the global feature extraction module of the transducer geological element map to obtain high-dimensional features representing the spatial features of the geological element;
s23, constructing a stratum category-oriented fully-connected classification mapping module to map the high-dimensional characteristic representation to stratum categories contained in the geological sample.
Further, the GNN module includes a sub-sampling unit and a space graph convolution unit, where the sub-sampling unit is configured to randomly sample an oversized graph of the geological node graph structural grid into a sub-graph with a fixed scale and smaller, and set the number of neighbor sampling points for each graph convolution layer l, so as to aggregate l-order neighborhood information of each node into an embedded vector of a graph node with a specified dimension; the space map convolution unit is used for aggregating characteristic information of neighborhood geological nodes and comprises a plurality of space map convolution layers, space coordinates of each map node are reserved by the space map convolution as map node neighborhood characteristic aggregation weight factors, residual connection is used between layers, and larger receptive fields are obtained layer by layer to obtain global information.
Further, the global feature extraction module of the transform geological element map comprises a stacked normalization layer, a multi-head attention layer and a feedforward neural network layer, and the dimension output by the GNN module is equal to the embedding dimension of the map nodes processed and input by the global feature extraction module of the transform geological element map.
Further, the fully-connected classification mapping module comprises a plurality of fully-connected layers, the input dimension of the fully-connected layers is the output dimension of the transform geological factor graph global feature extraction module, and the output dimension of the fully-connected layers is adjusted according to the stratum category number of the training task so as to map the output result to the appointed category of the prediction task.
Further, the loss function is a model mixed loss function based on stratum category and stratum distribution of the geological element map, and the map node category cross entropy loss function I is directly constructed by predicting stratum category for each node 0 Constructing graph similarity loss l for batch sampling graph nodes 1 To minimize the difference between the predicted formation distribution of each batch of randomly sampled graph nodes and the formation distribution in the real modeling space, wherein the model mixing loss function is:
the invention also provides a three-dimensional geological modeling method based on the map deep learning geological prediction model, which comprises the following steps:
(A) Acquiring real geological data, and carrying out graph structuring processing on the real geological data to convert the real geological data into geological node graph structural grids;
(B) Constructing a map deep learning geological prediction model pre-training workflow, taking the map deep learning geological prediction model as a main body as a pre-training model, storing corresponding pre-training model parameters after training through samples with large data volume in a pre-training stage, loading parameters of a corresponding pre-training layer as a multiplexing model on a prediction application task, and carrying out fine adjustment on model parameters so as to predict unknown node stratum categories on a geological node map structural grid.
The invention also provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to realize the steps of the three-dimensional geological modeling method based on the map deep learning geological prediction model.
The invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program realizes the steps of the three-dimensional geological modeling method based on the map deep learning geological prediction model when being executed by a processor.
The method for constructing the map deep learning geological prediction model intuitively constructs the spatial association between the spatial nodes by using the map structure, characterizes the spatial relationship of the neighborhood nodes by GNN aggregation, effectively excavates the spatial distribution mode of geological elements by using a transducer layer, and can give consideration to the local and global spatial correlation of the geological elements.
Meanwhile, the three-dimensional geological modeling method, the computer equipment and the storage medium based on the map deep learning geological prediction model, which are disclosed by the invention, adopt a pre-training strategy when the map deep learning geological prediction model is used for prediction, overcome the problem caused by sample data sparseness to a certain extent, effectively avoid the over-fitting of the model, and obtain the three-dimensional geological modeling which accords with geological laws and observations better under the condition of realizing less geological sampling input, thereby improving the application value of the model.
Specifically, geological sampling data is processed into map structure data through space triangulation, stratum prediction in a three-dimensional space is converted into map node stratum classification problems of a map neural network, map node neighborhood information is aggregated by using a GNN network model to generate map node feature embedded vectors, a map convolution form based on spatial relations is constructed, the influence of different geological nodes on the neighborhood is considered implicitly, the learning capacity of the geological feature is improved by combining a transducer lifting model, and a map layer is fully connected through stratum classification, so that stratum classification results of map node stratum classification results are output, and stratum classification of each triangular splitting point in space is obtained. When the model is used for prediction, the model is pre-trained on a large number of geological model sample data sets to obtain sufficient geological priori, and model parameter fine adjustment is performed on a task data set with a smaller scale to obtain the three-dimensional geological model based on the graph grid. The spatial correlation among the spatial nodes is intuitively constructed through the graph structure, the spatial distribution mode of geological properties is effectively mined by using a transducer layer, and the problem caused by sample data sparseness can be overcome to a certain extent by adopting a pre-training strategy, and the overfitting of the model is effectively avoided.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of an embodiment of a method for constructing a map deep learning geological prediction model according to the present invention;
FIG. 2 is a schematic flow chart of an embodiment of a method for constructing a map deep learning geological prediction model according to the present invention;
FIG. 3 is a schematic flow chart of an embodiment of a three-dimensional geologic modeling method based on a map deep learning geologic prediction model according to the present invention;
FIG. 4 is a schematic diagram of a computer device according to an embodiment of the invention;
element reference numeral description 101, electronic device; 102. a memory; 103. a processor(s),
Detailed Description
Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the following embodiments and features in the embodiments may be combined with each other without conflict.
It should be noted that the illustrations provided in the following embodiments merely illustrate the basic concept of the present invention by way of illustration, and only the components related to the present invention are shown in the drawings and are not drawn according to the number, shape and size of the components in actual implementation, and the form, number and proportion of the components in actual implementation may be arbitrarily changed, and the layout of the components may be more complicated.
Example 1
Referring to fig. 1, the invention discloses a method for constructing a map deep learning geological prediction model, which comprises the following steps:
s1, constructing a geological node map structural grid of a sample data set to serve as a training data set, wherein the training data set comprises training data, verification data and test data;
in a preferred embodiment, in step S1, a geologic structure model of a specified geologic event is randomly synthesized as a sample dataset for geologic modeling; in this embodiment, the Noddy geologic model sample simulation tool may be utilized to randomly synthesize a geologic structure model of a specified geologic event, for example, a Noddy geologic model sample simulation tool may be utilized to randomly generate a geologic event such as a fold, fault, etc. as a sample dataset;
the invention discloses a construction method of a map deep learning geological prediction model, which aims to construct a map deep learning model based on GNN-transform combination, so that after geological events are obtained as sample data, the sample data are further processed to be converted into a geological node map structural grid as a training data set of the map deep learning model. The construction of the geological node map structural grid comprises the following steps: and carrying out regular sampling and discretization mapping processing on the sample data set to obtain a preliminary geological node grid, carrying out graph structuring unified processing to obtain a geological node graph structure grid, and randomly shielding part of node features on the preliminary geological node grid to divide the data set. In this embodiment, as a preferred scheme, the construction of the geological node map structure grid specifically includes the following steps:
s11, regularly sampling the sample data set, wherein the step of sampling comprises the following steps: and dispersing the geological modeling sample into space scattered points, carrying out category coding according to the prior knowledge of the existing geology and the stratum age from new to old by natural numbers from small to large, wherein each stratum category corresponds to a natural number i, and taking a stratum category value i of a graph node as a graph node characteristic.
S12, constructing a preliminary geological node grid: matching the scattered points in the S11 to the positions of grid nodes on the regular grid units with preset sizes, and endowing the stratum category corresponding values i of the scattered points with corresponding grid nodes so as to construct a preliminary geological node grid;
after the scattered points are obtained, the scattered points are required to be matched to the positions of grid nodes on the regular grid units, the size of the grid units can be set according to the implementation requirement, after the size of the grid units is determined, in the embodiment, the scattered points and the grid nodes can be matched by using a nearest point matching algorithm, the distance between the scattered points and the adjacent grid nodes is calculated, then one scattered point closest to the grid nodes is used as a matching object of the grid nodes, the stratum category corresponding value i of the scattered points is given to the corresponding grid nodes, namely, the parameters of the grid nodes are (x, y, z and i), in addition, when the grid nodes are matched to the same scattered point, the single scattered point can be assigned to the grid nodes, and the preliminary geological node grid is formed.
S13, setting a virtual geological structure in the preliminary geological node grid, reserving original attributes of grid nodes related to the virtual geological structure in the geological node grid to serve as known data, performing mask masking processing on the attributes of other grid nodes to serve as test data, and regarding most of the grid nodes as training data and the rest as verification data according to a preset proportion for the known data;
after the preliminary geological node grid is generated, in order to truly restore the data condition in the practical geological modeling application, in this embodiment, a virtual geological structure such as a section line position and a virtual drilling coordinate is further set in the preliminary geological node grid to further simulate the use condition of real geological data, after the preliminary geological node grid is inserted into the virtual geological structure, the attribute of the grid node included in the virtual geological structure such as the section line position and the virtual drilling coordinate is kept unchanged, the attribute of other grid nodes in the preliminary geological node grid is masked, that is, the parameters of the grid nodes included in the virtual geological structure are still (x, y, z, i), the parameters of the other grid nodes are set to be (x, y, z, 0) to serve as test data, and meanwhile, the grid nodes with the parameters of (x, y, z, i) are further segmented into training data and verification data according to a preset proportion.
In addition, a virtual geological structure may be set according to the scattered point sampling proportion, for example, the data set may be randomly segmented according to the proportion for the regular grid node data mapped in step S12, for example, the proportion of the training data, the test data and the verification data may be set to 6:2:2, so as to mask.
S14, performing triangulation processing on the regular grid nodes, and constructing edge connection among the grid nodes, so that each grid node is positioned on the graph structure to obtain a final geological node graph structure grid.
In this embodiment, a constrained Delaunay triangulation algorithm is used to triangulate the regular grid nodes in S13, and edge connection between the grid nodes is constructed, so that each grid node is located on a graph structure, and the graph structures are connected, so that the graph neural network can directly process the geological node graph structure grid, where. The grid nodes are structured by the graph and are collectively called as graph nodes.
S2, constructing a graph deep learning geological prediction model based on a GNN-converter structure, wherein the graph deep learning geological prediction model comprises a GNN module, a converter geological element graph global feature extraction module and a full-connection classification mapping module which are sequentially connected.
As a preferred scheme, as shown in fig. 2, the map deep learning geological prediction model construction includes the following steps:
s21, constructing a GNN module, and based on graph neighborhood information of the training data set, generating embedded vectors for each graph node, wherein the embedded vectors are used for representing the spatial relationship among geological elements in a modeling space;
in step S21, the GNN module is used as a geological element map node feature aggregation module, and is configured to aggregate map neighborhood information, generate an embedded vector for a map node, and expand a model receptive field layer by layer, so as to implicitly extract a spatial relationship between geological elements in a modeling space.
In an embodiment, the GNN module includes a sub-sampling unit and a space graph convolution unit, where the sub-sampling unit is configured to randomly sample an oversized graph of a geological node graph structure grid into sub-graphs with a fixed scale and smaller sizes, and set the number of neighbor sampling points for each graph convolution layer l, so as to aggregate the l-order neighborhood information of each node into an embedded vector of a graph node with a specified dimension; the space diagram convolution unit is used for aggregating characteristic information of the neighborhood of the diagram nodes, and acquiring more global information by acquiring larger receptive fields layer by layer.
In this embodiment, for the oversized graph composed of three-dimensional geological element nodes, considering the characteristic of huge number of nodes, the sub-sampling unit adopts a sub-sampling strategy to randomly sample the oversized graph of the training data set into sub-graphs with a fixed scale and smaller scale, so as to input data into the space graph convolution unit in batches for training, wherein the number of neighbor sampling points is set for each graph convolution layer, so that the l-order neighborhood information of each node is aggregated into a graph node embedding vector with a specified dimension.
The space map convolution unit is used for constructing a map convolution method based on geological element neighborhood space relations so as to guide a model to extract more effective space relation representation, and the space map convolution unit is expanded to a deep layer to aggregate high-order neighborhood information of map nodes so as to pay attention to the space relation representation among geological elements at a far distance. In this embodiment, the spatial map convolution unit includes a plurality of spatial map convolution layers, where the spatial map convolution layers are formed by stacking a map convolution layer, a normalization layer, and an activation layer, and residual connection is used between the spatial map convolutions, so that a larger receptive field is obtained layer by layer, so as to obtain global information.
Further, in this embodiment, the following formula may be used to perform graph convolution of the geological element neighborhood spatial relationship:
wherein p is i Representing the spatial coordinates of node i, W and b are learnable parameters, ε is a minimum value, N i Is the neighborhood of node i.
The graph node characteristics are aggregated by introducing a space graph convolution operation, the space graph convolution reserves the space coordinate of each graph node, the space coordinate is used as a fixed characteristic representation of the graph node, the characteristic is not changed in a cross-layer mode, the characteristic is not directly involved in node characteristic convolution, in order to enable a model to learn and construct better graph node characteristic representation, the distance, the elevation difference and the direction angle between adjacent nodes in a geological element graph are used as weight training factors for the feature aggregation of the adjacent nodes in the graph, and the graph node characteristic embedding vector considering geological anisotropy is generated.
S22, constructing a global feature extraction module of the transform geological element map, and further encoding the embedded vector output by the GNN module based on the global feature extraction module of the transform geological element map to obtain high-dimensional features.
The transform geological factor map global feature extraction module uses a transform coding layer to enhance model learning capability to further obtain high-dimensional features characterizing the geological factor spatial features from the input of the GNN module, which compromise global and local geological anisotropy.
In this embodiment, as a preferred scheme, as shown in fig. 2, the global feature extraction module of the transform geological element map includes a stacked normalization layer, a multi-head attention layer and a feedforward neural network layer, where the multi-head attention layer implicitly focuses on the influence of different neighboring nodes on the center node and focuses on the global feature contained in the input sub-graph feature sequence, and the dimension output by the GNN module is equal to the embedded dimension of the map node processed and input by the global feature extraction module of the transform geological element map.
S23, constructing a stratum category-oriented fully-connected classification mapping module, and mapping the high-dimensional characteristic representation to stratum categories contained in the geological sample.
In this embodiment, as a preferred solution, the fully-connected classification mapping module includes a plurality of fully-connected layers, an input dimension of which is an output dimension of the transform geological factor map global feature extraction module, and the output dimension is adjusted according to the number of stratum categories of the training task, so as to map an output result to a specified category of the prediction task.
S3, setting a loss function for the map deep learning geological prediction model, and training the map deep learning geological prediction model based on the training data set and the loss function, wherein training and parameter updating are performed based on training data, and when the loss function on verification data is not lowered any more, training is stopped, so that a trained map deep learning geological prediction model is obtained.
In this embodiment, as a preferred scheme, the loss function is a model mixture loss function based on the stratum category and stratum distribution of the geological element map, which directly constructs a map node category cross entropy loss function l for each node prediction stratum category 0 Constructing graph similarity loss l for batch sampling graph nodes 1 To minimize the difference between the predicted formation distribution of each batch of randomly sampled graph nodes and the formation distribution in the real modeling space, wherein the model mixing loss function is:
in this embodiment, the super parameter lambda may be set 0 ＝0.5，λ 1 The training data set obtained in the step S1 is input into a geological factor map deep learning model for training, wherein the model is trained on training data to update model parameters, and training is stopped when the loss of the model on verification data is no longer reduced; the model directly outputs accuracy on the test set data for use in evaluating the accuracy of the model.
Example two
As shown in fig. 3, the invention further provides a three-dimensional geological modeling method based on the map deep learning geological prediction model, which comprises the following steps:
(A) Acquiring real geological data, and carrying out graph structuring processing on the real geological data to convert the real geological data into geological node graph structural grids;
in the step (a), firstly, a three-dimensional geological modeling sample data set is constructed by real geological drilling, profile data, scattered points and the like to be modeled, and is subjected to graph structuring processing, and is converted into a geological node graph structural grid, and the specific conversion step is as described in the first embodiment, and comprises the following steps:
s11, regularly sampling the sample data set, wherein the step of sampling comprises the following steps: and dispersing the geological modeling sample into space scattered points, carrying out category coding according to the prior knowledge of the existing geology and the stratum age from new to old and the natural number from small to large, wherein each stratum category corresponds to a natural number i, and taking a stratum category corresponding value i of a graph node as a graph node characteristic.
S12, constructing a preliminary geological node grid: matching the scattered points in the S11 to the positions of grid nodes on the regular grid units with preset sizes, and endowing the stratum category corresponding values i of the scattered points to corresponding grid nodes so as to construct a preliminary geological node grid;
after the scattered points are obtained, the scattered points are required to be matched to the positions of grid nodes on a regular grid unit, the size of the grid unit can be set according to the implementation requirement, after the size of the grid unit is determined, in the embodiment, the scattered points and the grid nodes can be matched by using a nearest-neighbor point matching algorithm, the distance between the scattered points and the adjacent grid nodes is calculated, then one scattered point closest to the grid nodes is used as a matching object of the grid nodes, and the stratum category corresponding value i of the scattered points is given to the corresponding grid nodes, namely, the parameters of the grid nodes are (x, y, z, i), in addition, when the grid nodes are matched to the same scattered point, the single scattered point can be assigned to the grid nodes, so that a preliminary geological node grid is formed.
S13, setting a virtual geological structure in the preliminary geological node grid, reserving original attributes of grid nodes related to the virtual geological structure in the geological node grid to serve as known data, performing mask masking processing on the attributes of other grid nodes to serve as test data, and regarding most of the grid nodes as training data and the rest as verification data according to a preset proportion for the known data;
after the preliminary geological node grid is generated, in order to truly restore the data condition in the practical geological modeling application, in this embodiment, a virtual geological structure such as a section line position and a virtual drilling coordinate is further set in the preliminary geological node grid to further simulate the use condition of real geological data, after the preliminary geological node grid is inserted into the virtual geological structure, the attribute of the grid node included in the virtual geological structure such as the section line position and the virtual drilling coordinate is kept unchanged, the attribute of other grid nodes in the preliminary geological node grid is masked, that is, the parameters of the grid nodes included in the virtual geological structure are still (x, y, z, i), the parameters of the other grid nodes are set to be (x, y, z, 0) to serve as test data, and meanwhile, the grid nodes with the parameters of (x, y, z, i) are further segmented into training data and verification data according to a preset proportion.
In addition, a virtual geological structure may be set according to the scattered point sampling proportion, for example, the data set may be randomly segmented according to the proportion for the regular grid node data mapped in step S12, for example, the proportion of the training data, the test data and the verification data may be set to 6:2:2, so as to mask.
S14, performing triangulation processing on the regular grid nodes, and constructing edge connection among the grid nodes, so that each grid node is positioned on the graph structure to obtain a final geological node graph structure grid.
In this embodiment, a constrained Delaunay triangulation algorithm is used to triangulate the regular grid nodes in S13, and edge connection between the grid nodes is constructed, so that each grid node is located on a graph structure, and the graph structures are connected, so that the graph neural network can directly process the geological node graph structure grid, where. The grid nodes are structured by the graph and are collectively called as graph nodes.
(B) Constructing a map deep learning geological prediction model pre-training workflow, taking the map deep learning geological prediction model in the first embodiment as a main body as a pre-training model, storing corresponding pre-training model parameters after training through samples with large data volume in the pre-training stage, loading parameters of a corresponding pre-training layer as a multiplexing model on a prediction application task, and carrying out fine tuning on model parameters so as to predict unknown node stratum types on a geological node map structural grid and output corresponding three-dimensional geological modeling.
In the step (B), the GNN-transducer model is firstly used as a pre-training main model for learning a space relation mode in a large number of geological model samples, in a prediction task data set, a pre-training model part only carries out parameter fine adjustment, and parameters of the full-connection classification mapping module in the step S23 are trained, so that the model faces to a prediction task of a specified geological sample data set. Specifically, firstly, a map deep learning geological prediction model pre-training workflow is constructed, GNN-transducer is used as a main body to serve as a pre-training model, corresponding pre-training model parameters are stored after training is carried out through samples with large data volume in a pre-training stage, on a prediction application task, parameters of a corresponding pre-training layer are loaded to serve as a multiplexing model, and the learning rate of the multiplexing model is adjusted to be one tenth of that in the pre-training stage, and fine adjustment of model parameters is carried out.
According to the invention, the spatial correlation between the spatial nodes is intuitively constructed by using the graph structure, the spatial relation of the neighborhood nodes is represented by the aggregation of the GNN modules, the spatial distribution mode of geological elements is effectively excavated by using the transducer module, meanwhile, the pre-training strategy is adopted, the problem caused by sample data sparseness is overcome to a certain extent, the overfitting of the model is effectively avoided, and the three-dimensional geological modeling which accords with geological rules and observation is obtained under the condition of less geological sampling input, so that the application value is improved.
Example III
Referring to fig. 4, in an embodiment, the present invention further provides an electronic device including one or more memories 102 and a processor 103, and may further include a computer program stored in the memory 102 and executable on the processor 103, for example, the program of the three-dimensional geological modeling method based on the map deep learning geological prediction model. The electronic device 101 may be a computer, a notebook computer, a tablet computer, a workstation, a personal digital assistant, and the like. The components shown herein, their connections and relationships, and their functions, are meant to be exemplary only, and are not meant to limit implementations of the application described and/or claimed herein. The processor 103 may be comprised of integrated circuits in some embodiments, for example, a single packaged integrated circuit, or may be comprised of multiple integrated circuits packaged with the same or different functions, including one or more central processing units (Central Processing unit, CPU), microprocessors, digital processing chips, graphics processors, a combination of various control chips, and the like. The processor 103 is a Control Unit (Control Unit) of the electronic device 101, connects the respective components of the entire electronic device 101 using various interfaces and lines, and executes various functions of the electronic device 101 and processes data by running or executing programs or modules stored in the memory 102 and calling data stored in the memory 102.
It should be noted that, the execution of the computer program by the processor of the electronic device is a step in the three-dimensional geological modeling method based on the map-deep learning geological prediction model, so that the implementation of the three-dimensional geological modeling method based on the map-deep learning geological prediction model is applicable to the electronic device, and the same or similar beneficial effects can be achieved.
Example IV
The invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program which realizes the steps in the three-dimensional geological modeling method based on the map deep learning geological prediction model when being executed by a processor. Wherein the computer readable medium may comprise: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, randomAccess Memory), an electrical carrier signal, a telecommunication signal, a software distribution medium, and so forth.
The above embodiments are merely illustrative of the principles of the present invention and its effectiveness, and are not intended to limit the invention. Modifications and variations may be made to the above-described embodiments by those skilled in the art without departing from the spirit and scope of the invention. Accordingly, it is intended that all equivalent modifications and variations of the invention be covered by the claims, which are within the ordinary skill of the art, be within the spirit and scope of the present disclosure.
Claims (10)
1. The construction method of the map deep learning geological prediction model is characterized by comprising the following steps of:
s1, constructing a sample data set, and carrying out graph structuring processing on the sample data set to form a geological node graph structural grid as a training data set, wherein the training data set comprises training data, verification data and test data;
s2, constructing a graph deep learning geological prediction model based on a GNN-transform model, wherein the graph deep learning geological prediction model comprises a GNN module, a transform geological element graph global feature extraction module and a full-connection classification mapping module which are sequentially connected, the GNN module is used for generating embedded vectors for all graph nodes, and the transform geological element graph global feature extraction module is used for further encoding the embedded vectors output by the GNN module;
s3, setting a loss function for the map deep learning geological prediction model, and training the map deep learning geological prediction model based on the training data set and the loss function, wherein the training is performed based on the training data to achieve parameter updating, and when the loss function on the verification data is not lowered any more, the training is stopped, so that the trained map deep learning geological prediction model is obtained.
2. The method for constructing a map deep learning geological prediction model according to claim 1, wherein in step S1, performing a map structuring process on the sample data set to form a geological node map structural grid comprises the following steps:
s11, regularly sampling the sample data set, wherein the step of sampling comprises the following steps: dispersing a geological modeling sample into space scattered points, carrying out category coding according to the prior knowledge of the existing geology and the stratum age from new to old and the natural number from small to large, wherein each stratum category corresponds to a natural number i, and taking a stratum category corresponding value i of a graph node as a graph node characteristic;
s12, constructing a preliminary geological node grid: matching the scattered points in the S11 to the positions of grid nodes on the regular grid units with preset sizes, and endowing the stratum category corresponding values i of the scattered points to corresponding grid nodes so as to construct a preliminary geological node grid;
s13, setting a virtual geological structure in the preliminary geological node grid, reserving original attributes of grid nodes related to the virtual geological structure in the preliminary geological node grid to serve as known data, performing mask shielding processing on the attributes of the other grid nodes to serve as test data, and regarding most of the grid nodes as training data and the rest as verification data according to a preset proportion for the known data;
s14, performing triangulation processing on the regular grid nodes, and constructing edge connection among the grid nodes, so that each grid node is positioned on the graph structure to obtain a final geological node graph structure grid.
3. The method for constructing a map deep learning geological prediction model according to claim 1, wherein the map deep learning geological prediction model construction comprises the following steps:
s21, constructing a GNN module, and based on graph neighborhood information of the training data set, generating embedded vectors for each graph node, wherein the embedded vectors are used for representing the spatial relationship among geological elements in a modeling space;
s22, constructing a global feature extraction module of a transducer geological element map, and further encoding an embedded vector output by the GNN module based on the global feature extraction module of the transducer geological element map to obtain high-dimensional features representing the spatial features of the geological element;
s23, constructing a stratum category-oriented fully-connected classification mapping module to map the high-dimensional characteristic representation to stratum categories contained in the geological sample.
4. The method for constructing the map deep learning geological prediction model according to claim 3, wherein the GNN module comprises a sub-sampling unit and a space map convolution unit, the sub-sampling unit is used for randomly sampling an oversized map of a geological node map structural grid into sub-maps with fixed and smaller scale, and the number of neighbor sampling points is set for each map convolution layer so as to aggregate the l-order neighborhood information of each node into an embedded vector of a map node with a specified dimension; the space map convolution unit is used for aggregating characteristic information of neighborhood geological nodes and comprises a plurality of space map convolution layers, space coordinates of each map node are reserved by the space map convolution as map node neighborhood characteristic aggregation weight factors, residual connection is used between layers, and larger receptive fields are obtained layer by layer to obtain global information.
5. The method for constructing a map deep learning geological prediction model according to claim 3, wherein the global feature extraction module of the map geological element map comprises a stacked normalization layer, a multi-head attention layer and a feedforward neural network layer, and the dimension output by the GNN module is equal to the embedded dimension of the map nodes processed and input by the global feature extraction module of the map geological element map.
6. The method for constructing a map deep learning geological prediction model according to claim 3, wherein the fully connected classification mapping module comprises a plurality of fully connected layers, the input dimension of which is the output dimension of the transform geological factor map global feature extraction module, and the output dimension of which is adjusted according to the number of stratum categories of the training task so as to map the output result to the designated category of the prediction task.
7. The method for constructing a map deep learning geological prediction model according to claim 1, wherein the loss function is a model mixture loss function based on stratum categories and stratum distribution of geological element maps, and the map node category cross entropy loss function is directly constructed for each node prediction stratum category 0 Constructing graph similarity loss l for batch sampling graph nodes 1 To minimize the difference between the predicted formation distribution of each batch of randomly sampled graph nodes and the formation distribution in the real modeling space, wherein the model mixing loss function is:
8. a three-dimensional geological modeling method based on a map deep learning geological prediction model is characterized by comprising the following steps:
(A) Acquiring real geological data, and carrying out graph structuring processing on the real geological data to convert the real geological data into geological node graph structural grids;
(B) Constructing a map deep learning geological prediction model pre-training workflow, taking the map deep learning geological prediction model as a main body as a pre-training model, storing corresponding pre-training model parameters after training through samples with large data volume in the pre-training stage, loading parameters of corresponding pre-training layers as a multiplexing model on a prediction application task, and carrying out fine adjustment on model parameters so as to predict unknown node stratum categories on a geological node map structural grid.
9. An electronic device, comprising: at least one processor; and a memory communicatively coupled to at least one of the processors; wherein the memory stores instructions executable by the processor for execution by the processor to implement the three-dimensional geologic modeling method based on a graph-deep learning geologic prediction model of claim 8.
10. A computer-readable storage medium storing computer instructions for execution by the computer to implement the map deep learning geological prediction model-based three-dimensional geological modeling method of claim 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211710956.1A CN116245013A (en) | 2022-12-29 | 2022-12-29 | Geological prediction model construction method, modeling method, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211710956.1A CN116245013A (en) | 2022-12-29 | 2022-12-29 | Geological prediction model construction method, modeling method, equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116245013A true CN116245013A (en) | 2023-06-09 |
Family
ID=86635609
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211710956.1A Pending CN116245013A (en) | 2022-12-29 | 2022-12-29 | Geological prediction model construction method, modeling method, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116245013A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116821452A (en) * | 2023-08-28 | 2023-09-29 | 南京邮电大学 | Graph node classification model training method and graph node classification method |
| CN116842416A (en) * | 2023-07-18 | 2023-10-03 | 中国电建集团华东勘测设计研究院有限公司 | Calculation method suitable for coral reef sand foundation pipe pile side friction resistance |
| CN117393050A (en) * | 2023-10-17 | 2024-01-12 | 哈尔滨工业大学（威海） | Protein function recognition method, device and readable storage medium |
| CN117633140A (en) * | 2024-01-25 | 2024-03-01 | 中化地质矿山总局山东地质勘查院 | Urban geological investigation method based on big data cloud computing technology |
-
2022
- 2022-12-29 CN CN202211710956.1A patent/CN116245013A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116842416A (en) * | 2023-07-18 | 2023-10-03 | 中国电建集团华东勘测设计研究院有限公司 | Calculation method suitable for coral reef sand foundation pipe pile side friction resistance |
| CN116842416B (en) * | 2023-07-18 | 2024-03-15 | 中国电建集团华东勘测设计研究院有限公司 | Calculation method suitable for coral reef sand foundation pipe pile side friction resistance |
| CN116821452A (en) * | 2023-08-28 | 2023-09-29 | 南京邮电大学 | Graph node classification model training method and graph node classification method |
| CN116821452B (en) * | 2023-08-28 | 2023-11-14 | 南京邮电大学 | Graph node classification model training method and graph node classification method |
| CN117393050A (en) * | 2023-10-17 | 2024-01-12 | 哈尔滨工业大学（威海） | Protein function recognition method, device and readable storage medium |
| CN117633140A (en) * | 2024-01-25 | 2024-03-01 | 中化地质矿山总局山东地质勘查院 | Urban geological investigation method based on big data cloud computing technology |
| CN117633140B (en) * | 2024-01-25 | 2024-04-16 | 中化地质矿山总局山东地质勘查院 | Urban geological investigation method based on big data cloud computing technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116245013A (en) | Geological prediction model construction method, modeling method, equipment and storage medium | |
| US20180247227A1 (en) | Machine learning systems and methods for data augmentation | |
| RU2282242C2 (en) | Method for evening out a grid of nodes in accordance to characteristic features in digital image | |
| CN111832227A (en) | Shale gas saturation determination method, device and equipment based on deep learning | |
| Xiao et al. | Conditioning of deep-learning surrogate models to image data with application to reservoir characterization | |
| Shi et al. | Gnn-surrogate: A hierarchical and adaptive graph neural network for parameter space exploration of unstructured-mesh ocean simulations | |
| Kumar et al. | Ensemble-based assimilation of nonlinearly related dynamic data in reservoir models exhibiting non-Gaussian characteristics | |
| Zhang et al. | SaltISCG: Interactive salt segmentation method based on CNN and graph cut | |
| CN113420593B (en) | Small sample SAR automatic target recognition method based on hybrid inference network | |
| Razak et al. | Embedding physical flow functions into deep learning predictive models for improved production forecasting | |
| CN114692529B (en) | CFD high-dimensional response uncertainty quantification method and device, and computer equipment | |
| GB2584447A (en) | Apparatus method and computer-program product for processing geological data | |
| GB2584449A (en) | Apparatus method and computer-program product for calculating a measurable geological metric | |
| CN113419278B (en) | Well-seismic joint multi-target simultaneous inversion method based on state space model and support vector regression | |
| CN115375867A (en) | Method, system, equipment and medium for calculating geothermal resource quantity by using grid model | |
| Howe | Gridfields: model-driven data transformation in the physical sciences | |
| CN115685314A (en) | Seismic reservoir physical property parameter prediction method and device | |
| Da Silva et al. | Selection of a dimensionality reduction method: An application to deal with high-dimensional geostatistical realizations in oil reservoirs | |
| CN114297929A (en) | Machine learning-fused radial basis function curved surface complex ore body modeling method and device | |
| Cirac et al. | Deep hierarchical distillation proxy-oil modeling for heterogeneous carbonate reservoirs | |
| Sire et al. | Quantizing rare random maps: application to flooding visualization | |
| Li et al. | Time series clustering based on relationship network and community detection | |
| Li et al. | Single-well lithofacies identification based on logging response and convolutional neural network | |
| Lyu et al. | Rough discrete fracture network multi-parameter joint modeling based on improved neural spline flow | |
| Soriano-Vargas et al. | A synthetic case study of measuring the misfit between 4D seismic data and numerical reservoir simulation models through the Momenta Tree |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240402 Address after: 430000 Lu Mill Road, Hongshan District, Wuhan, Hubei Province, No. 388 Applicant after: CHINA University OF GEOSCIENCES (WUHAN CITY) Country or region after: China Applicant after: CHINA RAILWAY DESIGN Corp. Applicant after: China National Railway Group Co.,Ltd. Address before: 430000 Lu Mill Road, Hongshan District, Wuhan, Hubei Province, No. 388 Applicant before: CHINA University OF GEOSCIENCES (WUHAN CITY) Country or region before: China |