Stratum engineering property analysis and evaluation method and system based on semantic and spatial information fusion

Abstract

The application discloses a stratum engineering property analysis and evaluation method and system based on semantic and spatial information fusion, comprising the following steps: obtaining multiple groups of different types of sampling data and depth coordinate information for a selected borehole, and constructing a feature similarity graph and a spatial adjacency graph; extracting attribute features and spatial features from the feature similarity graph and the spatial adjacency graph; adaptively fusing the attribute features and the spatial features to obtain a plurality of fusion features; performing alignment learning on the fusion features to obtain an alignment total loss; performing fusion processing on the fusion features to obtain a total fusion feature; reconstructing the multiple groups of different types of sampling data based on the total fusion feature, calculating a reconstruction total loss, and performing network training with the minimum alignment total loss and reconstruction total loss as the target, finding the optimal total fusion feature, and performing clustering; and completing stratum engineering property analysis and evaluation based on the clustering result. Through the attribute features and the spatial features, the application can automatically divide strata and quantitatively evaluate the engineering properties and uniformity of the strata.

Description

Technical Field

[0001] This invention belongs to the field of interdisciplinary technology of geotechnical engineering information technology and artificial intelligence, specifically involving a method and system for analyzing and evaluating the geological properties based on the fusion of semantic and spatial information. Background Technology

[0002] Currently, the analysis and evaluation of stratigraphic properties in geotechnical engineering surveys mainly rely on manual methods. Engineers manually divide the stratigraphic units based on data from core drilling, in-situ testing (such as standard penetration tests and static cone penetration tests), and laboratory geotechnical tests, combined with regional geological experience. They then describe the physical and mechanical properties of each soil layer according to specifications (such as "plastic silty clay") and conduct a qualitative evaluation of the homogeneity of the strata.

[0003] In recent years, with the development of machine learning technology, some studies have begun to explore the use of algorithms to assist in stratigraphic division and evaluation. For example, existing studies have used the K-Means clustering algorithm to characterize and divide multi-attribute soil and rock data; used fuzzy classification methods to handle the uncertainty of multi-source information; and used convolutional neural networks to extract local patterns from well logging curves for lithological identification.

[0004] Although existing technologies have made some progress in automating geotechnical engineering stratigraphic division and evaluation, the following shortcomings still exist:

[0005] First, existing methods neglect the prior constraint of the spatial continuity of the stratigraphy. When processing borehole data, current methods often treat test data from each depth point as independent samples, clustering solely based on the similarity of attribute characteristics. However, strata possess natural continuity and topological relationships in the vertical direction—the physical and mechanical properties of adjacent depth points should exhibit gradual changes rather than random jumps. Existing methods fail to explicitly incorporate this spatial structure prior into the modeling process, resulting in unreasonable "jagged" boundaries in the clustering results, which are severely inconsistent with geological realities and consequently affect the accuracy of subsequent engineering performance evaluations.

[0006] Second, the integration of attribute features and spatial features is insufficient. Existing methods typically perform simple stitching or shallow fusion of multi-view data, failing to fully consider the dynamic changes in the relative importance of attribute and spatial information under different geological conditions (such as stratigraphic boundaries vs. stratigraphic interiors). At stratigraphic boundaries, abrupt changes in attribute features should dominate; within stratigraphic interiors, spatial continuity should dominate. Existing methods lack the ability to adaptively adjust these dynamic weights.

[0007] Third, the evaluation of geological engineering properties still relies on human experience and lacks objective quantitative standards. Existing clustering algorithms stop at outputting stratigraphic division labels and fail to further transform the clustering results into evaluation conclusions that can be directly used in engineering. Engineers still need to manually read the test data of each layer and perform property descriptions and homogeneity judgments based on the clustered stratigraphic divisions. The evaluation process is highly subjective and has poor repeatability.

[0008] Fourth, there is a lack of quantitative evaluation mechanisms for intra- and inter-layer homogeneity. Existing methods mostly provide qualitative descriptions of formation homogeneity (such as "homogeneous," "relatively homogeneous," and "non-homogeneous"), lacking quantitative calculation indicators based on clustering results. Existing methods fail to provide quantifiable evaluation criteria for the dispersion of sample points within a layer to the cluster center and the similarity between adjacent strata.

[0009] In summary, existing technologies in geotechnical engineering have not yet solved key problems such as spatial structural information fusion, attribute-spatial adaptive weighting, automated output of evaluation results, and quantitative calculation of homogeneity in the analysis and evaluation of the engineering properties of stratigraphic units. There is an urgent need for a method and system that can integrate semantic and spatial information, automatically divide stratigraphic units, and quantitatively evaluate their engineering properties and homogeneity. Summary of the Invention

[0010] To address the aforementioned issues, this invention proposes a method and system for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information. By integrating the attribute characteristics and spatial structure information of geotechnical engineering survey data, the method automatically divides stratigraphic units and quantitatively evaluates their engineering properties and homogeneity.

[0011] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0012] In a first aspect, the present invention provides a method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information, comprising:

[0013] For the selected borehole, acquire multiple sets of sampling data of different types, as well as the depth coordinate information of each sampling point;

[0014] Based on multiple sets of sampling data of different types, construct corresponding feature similarity maps;

[0015] Based on the depth coordinate information of each sampling point, a spatial adjacency graph is constructed;

[0016] For each feature similarity map and spatial adjacency map, a graph convolutional network is used to extract attribute features and spatial features respectively;

[0017] A first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data.

[0018] Alignment learning is performed on the fusion features corresponding to each group of sampled data to obtain the total alignment loss;

[0019] A second attention mechanism network is used to fuse the fusion features corresponding to each group of sampled data to obtain the total fusion feature.

[0020] Based on the total fusion feature, multiple sets of sampled data of different types are reconstructed, and the total reconstruction loss before and after reconstruction of each sampled data is calculated. With the goal of minimizing the alignment total loss and the total reconstruction loss, the second attention mechanism network is repeatedly trained until the optimal total fusion feature is obtained.

[0021] The optimal total fusion features are clustered to obtain clusters, and each cluster corresponds to a type of stratum;

[0022] Based on the clustering results and predefined analysis strategies, the analysis and evaluation of the engineering properties of the formation are completed.

[0023] Secondly, the present invention provides a stratigraphic engineering property analysis and evaluation system based on semantic and spatial information fusion, characterized in that it includes a storage medium and a processor;

[0024] The storage medium is used to store instructions;

[0025] The processor is configured to operate according to the instructions to perform the method according to any one of the first aspects.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] First, it improves the geological rationality of stratigraphic division. By simultaneously constructing a feature similarity map and a spatial adjacency map, and injecting spatial continuity as an explicit constraint into the feature learning process, the clustering results exhibit a continuous blocky distribution in the depth direction, avoiding the problem of unreasonable boundaries in traditional methods. The clustering results are more consistent with the actual stratigraphic distribution.

[0028] Second, it achieves adaptive fusion of attribute and spatial information. By introducing an attention mechanism to dynamically adjust the fusion weights of feature similarity and spatial continuity, attribute features are prioritized at stratigraphic boundaries to capture abrupt changes, while spatial continuity is prioritized within stratigraphic layers to ensure smooth results, thus flexibly adapting to the information fusion needs under different geological conditions.

[0029] Third, enhance the robustness and consistency of multi-view fusion. By using a cross-view dual-alignment learning module, different views are forced to maintain consistency in feature similarity and spatial adjacency. Reliable spatial prior knowledge is injected into the learning process, effectively suppressing single-view noise interference, and making the model more resistant to interference and better at generalization in practical applications.

[0030] Fourth, it automates and quantifies the evaluation of geological engineering properties. Based on clustering results, it automatically performs geological property discrimination, intra-layer homogeneity evaluation, and inter-layer homogeneity evaluation, transforming unsupervised clustering output into evaluation conclusions that conform to engineering specifications. This solves the problems of traditional evaluation methods relying on human experience, strong subjectivity, and poor repeatability, significantly improving the efficiency and objectivity of geological evaluation in geotechnical engineering surveys.

[0031] Fifth, it has broad applicability. This invention is not limited to specific types of survey data, but can be widely applied to the evaluation of geological properties in various geotechnical engineering survey tasks such as overhead lines, cable lines, and site engineering, and has good transferability and promotion value. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0033] Figure 1 This is a schematic flowchart of a method for analyzing and evaluating the engineering properties of geological formations according to an embodiment of the present invention.

[0034] Figure 2 This is a schematic diagram illustrating the principle of the geological engineering property analysis and evaluation method in this embodiment of the invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0036] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0037] Example 1

[0038] This invention provides a method for analyzing and evaluating the engineering properties of geological formations based on the fusion of semantic and spatial information, such as... Figures 1-2 As shown, it includes the following steps:

[0039] (1) For the selected borehole, obtain multiple sets of sampling data of different types, as well as the depth coordinate information of each sampling point;

[0040] (2) Based on multiple sets of sampling data of different types, construct corresponding feature similarity maps;

[0041] (3) Construct a spatial adjacency graph based on the depth coordinate information of each sampling point;

[0042] (4) For each feature similarity map and spatial adjacency map, a graph convolutional network is used to extract attribute features and spatial features respectively;

[0043] (5) The first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data;

[0044] (6) Alignment learning is performed on the fusion features corresponding to each group of sampled data to obtain the total alignment loss;

[0045] (7) The second attention mechanism network is used to fuse the fusion features corresponding to each group of sampled data to obtain the total fusion features;

[0046] (8) Based on the total fusion feature, reconstruct multiple sets of different types of sampled data, and calculate the total reconstruction loss before and after reconstruction of each sampled data. With the goal of minimizing the total alignment loss and the total reconstruction loss, repeatedly train the second attention mechanism network until the optimal total fusion feature is obtained.

[0047] (9) Cluster the optimal total fusion features to obtain clusters, and each cluster corresponds to a type of stratum;

[0048] (10) Based on the clustering results and the predefined analysis strategy, complete the analysis and evaluation of the geological engineering properties.

[0049] In one specific embodiment of the present invention, the types of sampling indicators included in each group of sampling data are different.

[0050] In one specific embodiment of the present invention, for the first The group of sampled data, and the corresponding data matrix is , ,in, The total number of sampling points. For the first The feature dimension (i.e., data dimension) of the group sampled data; the method for constructing the feature similarity map includes:

[0051] Constructing the first based on the KNN method Feature adjacency matrix corresponding to group sampled data As a feature similarity map, Feature adjacency matrix The formula for calculating the elements is:

[0052] ,

[0053] In the formula, Represents the characteristic adjacency matrix elements, and They represent the first Sampling points in group sampling data and sampling points The feature vector formed by the corresponding sample data, Indicates sampling point of The set of feature vectors of the nearest neighbor sampling points.

[0054] In one specific embodiment of the present invention, sampling points are defined. The depth coordinates are The method for constructing the spatial adjacency graph includes:

[0055] Constructing a spatial adjacency matrix based on the KNN method As a spatial adjacency graph Spatial adjacency matrix The formula for calculating the elements is:

[0056] ,

[0057] In the formula, Representing the spatial adjacency matrix elements, Indicates sampling point of The set of depth coordinates of the nearest neighbor sampling points. Sampling points The depth coordinates.

[0058] In one specific embodiment of the present invention, the calculation formulas for the attribute features and spatial features are as follows:

[0059] ,

[0060] in,

[0061] ,

[0062] ,

[0063] In the formula, For the first The attribute features of the sampled data represent the embedded representations that reflect the feature patterns obtained from the feature similarity map; For the first The spatial features of the sampled data represent the embedded representation of the encoded spatial structure information obtained from the spatial adjacency graph; Indicates the first The corresponding normalized feature adjacency matrix of the sampled data groups has been self-connected. This represents a normalized spatial adjacency matrix with added self-connections; and All represent degree matrices, Indicates the first The learning weights corresponding to the group of sampled data, This represents the activation function.

[0064] In one specific embodiment of the present invention, for the first Sampling points in group sampling data The corresponding sample data is defined by the following attribute features: Its spatial characteristics are defined as follows: The first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data, including:

[0065] Based on the aforementioned attribute features and spatial features Construct splicing features ;

[0066] The splicing features are obtained through nonlinear transformation. Mapping to a common attention latent space yields intermediate features. intermediate features The calculation formula is:

[0067] ,in, This is the weight matrix. It is the bias vector;

[0068] Through affine transformation and normalization operations, first attention weight coefficients corresponding to attribute features and spatial features are generated, respectively. Second attention weight coefficient , In the formula, To map the implicit representation to a weight vector of scalar scores, and for the two weight coefficients to satisfy... , This is a normalization operation;

[0069] The fusion features corresponding to each group of sampled data are calculated using the following formula:

[0070] ,

[0071] ,

[0072] In the formula, Indicates the first The fusion features corresponding to the group of sampled data Indicates the first Sampling points in group sampling data The fusion features corresponding to the sample data.

[0073] In one specific embodiment of the present invention, the first Group sampling data and the first The fusion features corresponding to the group sampled data are as follows: and ,in The total number of sampling points. The dimension of the latent space; the alignment learning of the fusion features corresponding to each group of sampled data to obtain the total alignment loss includes:

[0074] Computational fusion features and fusion features Cross cosine similarity matrix between , , its first Line 1 Column elements Defined as:

[0075] ,

[0076] in, Indicates the first The first group of sampled data The latent representation vector of each sample data; Indicates the first The first group of sampled data The latent representation vector of each sample data, Representing vectors Norm;

[0077] Define structural alignment loss The calculation formula is:

[0078] ,

[0079] In the formula, Represents the given target adjacency matrix; and All are indicator functions, when hour, Otherwise, it is 0; when hour, Otherwise, it is 0. For the target adjacency matrix No. Line 1 The value of the column;

[0080] The adjacency matrices of each feature are concatenated to obtain the attribute feature map. , , attribute feature map Substituting the formula for calculating the structural alignment loss, we obtain the feature alignment loss. ;

[0081] Spatial adjacency matrix Substituting into the formula for calculating structural alignment loss, we obtain the spatial alignment loss as follows: ;

[0082] Calculate the total alignment loss, the formula for which is:

[0083] ,

[0084] In the formula, and All are hyperparameters, and satisfy the following: , , This represents the total number of data sets of different types.

[0085] In one specific embodiment of the present invention, the method for calculating the optimal total fusion feature includes:

[0086] For the second attention mechanism network, the first The first point of attention will be the first Fusion features corresponding to group sampled data Each vector is mapped to a query vector through a linear transformation. Key vector Sum value vector ;

[0087] With the first Using the sampled data as the query center, calculate the first... The first set of sampled data An update of attention head The calculation formula used is:

[0088] ,

[0089] In the formula, Scaling factor The total number of data sets of different types;

[0090] The outputs of all attention heads are concatenated and then passed through a learnable linear projection matrix. By integrating, we obtain the first Augmented representation of group sampled data :

[0091] ,

[0092] in, This indicates a splicing operation. Indicates the number of attention heads;

[0093] Calculate total fusion features The calculation formula used is:

[0094] ,

[0095] In the formula, This represents a multilayer perceptron. Indicates a splicing operation;

[0096] For the The reconstruction process of group sampled data is defined as follows:

[0097] ,

[0098] in, Represents the reconstructed first... The data matrix of the group sampled data, This represents a normalized feature adjacency matrix with added self-connections. , Represents the identity matrix. Represents the characteristic adjacency matrix. It is a degree matrix. It is the first The weight matrix of the group sampled data, It is a non-linear activation function;

[0099] Calculate reconstruction loss The calculation formula is:

[0100] ,

[0101] In the formula, Represents the norm, Represents the first before reconstruction The data matrix of the group sampled data;

[0102] Calculate the overall loss The calculation formula is:

[0103] ;

[0104] Based on comprehensive losses With the goal of minimizing the total features, the second attention mechanism network is continuously trained until the optimal total fusion feature is obtained.

[0105] In one specific embodiment of the present invention, the step of completing the analysis and evaluation of geological engineering properties based on clustering results and a predefined analysis strategy includes:

[0106] Based on each cluster center, a predefined engineering property template is matched to determine the stratigraphic properties;

[0107] Calculate the first All sampled data within each cluster are routed to its cluster center. Average distance:

[0108] ,

[0109] in, For belonging to the first A collection of sampled data from each cluster, With preset threshold In comparison, if If the result is positive, the strata are considered homogeneous; otherwise, they are considered heterogeneous, suggesting the possible existence of sub-layers or local interlayers.

[0110] Calculate the distance between the cluster centers of adjacent clusters , and preset threshold Comparison: If If the two results are positive, it indicates that the physical and mechanical properties of the corresponding strata are similar, and they can be merged into the same engineering geological unit in engineering; otherwise, the original clustering result is maintained.

[0111] The method for analyzing and evaluating the engineering properties of geological formations in this invention will be described below with reference to a specific implementation method. This method specifically includes:

[0112] Step S10: For the selected borehole, acquire multiple sets of sampling data of different types, as well as the depth coordinate information of each sampling point.

[0113] For the selected borehole, obtain Groups of different types of sampling data, for the first The group of sampled data, and the corresponding data matrix is , ,in, This represents the total number of sampling points (specifically, the total number of depth sampling points). For the first The feature dimension of the sampled data (i.e., the number of features involved in the sampled data). Each sampling point Associate a depth coordinate , forming a depth coordinate vector In this embodiment, the survey parameters involved in the sampling data include, but are not limited to: standard penetration test blow count, static cone tip resistance and sidewall friction, natural density, water content, liquid limit, plastic limit, compressive modulus, cohesion, and internal friction angle. Different types of sampled data should be complementary. For example: the first set of sampled data (also known as visual data) Figure 1 The first set contains basic physical quantities; the second set of sampled data (also known as visual data) contains basic physical quantities. Figure 2 The first set of sampled data contains derived / structural features; the third set (also known as View 3) contains gradients / rates of change. The purpose of this design is to decouple the physical dimensions of geotechnical information: the view... Figure 1 Preserve the original physical response for lithological baseline identification; Figure 2 View 3 reveals pore structure and cementation characteristics through rock physics transformation, which is used for diagenetic analysis; View 4 highlights stratigraphic abrupt interfaces through gradient calculation, which is used for high-precision stratigraphic positioning. The three views work together to reduce lithological misjudgments and boundary drift caused by a single view, and improve the physical consistency and geometric accuracy of stratigraphic division.

[0114] Step S20: Construct a feature similarity graph and a spatial adjacency graph, and use a graph convolutional network to extract attribute features and spatial features respectively.

[0115] This step specifically includes:

[0116] Step S21: Construct a feature similarity map.

[0117] For the Group sampling data, let its corresponding data matrix be . (The rows of this data matrix correspond to sampling points, and the columns correspond to the sampled data values ​​of those sampling points.) The total number of sampling points. For the first The feature dimensions of the group sampled data; the method for constructing the feature similarity map includes:

[0118] Constructing the first based on the KNN method Feature adjacency matrix corresponding to group sampled data As a feature similarity map, Feature adjacency matrix The formula for calculating the elements is:

[0119] ,

[0120] In the formula, Represents the characteristic adjacency matrix elements, and They represent the first Sampling points in group sampling data and sampling points The feature vector formed by the corresponding sample data, This represents the sampling points calculated based on the Euclidean distance (or other applicable distance metric) in the feature space. of The set of feature vectors of the nearest neighbor sampling points.

[0121] Step S22: Construct a spatial adjacency graph.

[0122] The spatial adjacency graph is shared across all groups of sampled data (i.e., the spatial adjacency graphs corresponding to all groups of sampled data are identical), and is constructed solely based on depth coordinate information. Let's define the sampling points. The depth coordinates are The method for constructing the spatial adjacency graph includes:

[0123] Constructing a spatial adjacency matrix based on the KNN method As a spatial adjacency graph Spatial adjacency matrix The formula for calculating the elements is:

[0124] ,

[0125] In the formula, Representing the spatial adjacency matrix elements, Indicates sampling point of The set of depth coordinates of the nearest neighbor sampling points. Sampling points The depth coordinates.

[0126] The feature similarity map and spatial adjacency map are then input into a shared-parameter neural network GCN. The formulas for calculating the attribute features and spatial features of the sampled data are as follows:

[0127] ,

[0128] in,

[0129] ,

[0130] ,

[0131] In the formula, For the first The attribute features of the sampled data represent the embedded representations that reflect the feature patterns obtained from the feature similarity map; For the first The spatial features of the sampled data represent the embedded representation of the encoded spatial structure information obtained from the spatial adjacency graph; Indicates the first The normalized feature adjacency matrix corresponding to the group sampled data has been self-connected. This represents a normalized spatial adjacency matrix with added self-connections; and All represent degree matrices, Indicates the first The learning weights corresponding to the group of sampled data, This represents the activation function.

[0132] Step S30: Adaptively fuse attribute features and spatial features using a first attention mechanism network to obtain fused features corresponding to each group of sampled data.

[0133] For the Sampling points in group sampling data The corresponding sample data is defined by the following attribute features: Its spatial characteristics are defined as follows: The first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data, including:

[0134] Based on the aforementioned attribute features and spatial features Construct splicing features ;

[0135] The splicing features are obtained through nonlinear transformation. Mapping to a common attention latent space yields intermediate features. intermediate features The calculation formula is:

[0136] ,in, This is the weight matrix. It is the bias vector;

[0137] Through affine transformation and normalization operations, first attention weight coefficients corresponding to attribute features and spatial features are generated, respectively. Second attention weight coefficient , In the formula, To map the implicit representation to a weight vector of scalar scores, and for the two weight coefficients to satisfy... , This is a normalization operation;

[0138] The fusion features corresponding to each group of sampled data are calculated using the following formula:

[0139] ,

[0140] ,

[0141] In the formula, Indicates the first The fusion features corresponding to the group of sampled data Indicates the first Sampling points in group sampling data The fusion features corresponding to the sample data.

[0142] Attention mechanisms enable models to automatically adjust the importance of information sources based on the local content of the input data. In regions with abrupt changes in features, such as lithological boundaries, the model tends to assign more attention to these sources. Higher values ​​lead to a greater reliance on feature similarity to capture abrupt changes; in homogeneous regions within strata, there is a tendency to assign... Higher values ​​allow for greater utilization of spatial continuity to ensure smoothness and consistency in the interpretation results.

[0143] Step S40: Perform alignment learning on the fusion features corresponding to each group of sampled data to obtain the total alignment loss.

[0144] Specifically, the Group sampling data and the first The fusion features corresponding to the group sampled data are as follows: and ,in The total number of sampling points. The dimension of the latent space; the alignment learning of the fusion features corresponding to each group of sampled data to obtain the total alignment loss includes:

[0145] Computational fusion features and fusion features Cross cosine similarity matrix between , , its first Line 1 Column elements Defined as:

[0146] ,

[0147] in, Indicates the first The first group of sampled data The latent representation vector of each sample data; Indicates the first The first group of sampled data The latent representation vector of each sample data, Representing vectors Norm.

[0148] Define the target adjacency matrix Define the structural alignment loss The calculation formula is:

[0149]

[0150] In the formula, Represents the given target adjacency matrix; and All are indicator functions, when hour, Otherwise, it is 0; when hour, Otherwise, it is 0. For the target adjacency matrix No. Line 1 The values ​​of the columns allow the model to focus more on the region indicated by the target adjacency matrix.

[0151] Next, the adjacency matrices of each feature are concatenated to obtain the attribute feature map. , , attribute feature map Substituting the formula for calculating the structural alignment loss, we obtain the feature alignment loss. ;

[0152] Spatial adjacency matrix Substituting into the formula for calculating structural alignment loss, we obtain the spatial alignment loss as follows: ;

[0153] Calculate the total alignment loss, the formula for which is:

[0154] ,

[0155] In the formula, and These are all hyperparameters, used to balance the contribution strength of feature similarity constraints and spatial continuity constraints, respectively, and satisfying the following: , , This represents the total number of data sets of different types.

[0156] Step S50: The second attention mechanism network is used to fuse the fusion features corresponding to each group of sampled data to obtain the total fusion feature. Based on the total fusion feature, multiple groups of sampled data of different types are reconstructed, and the total reconstruction loss before and after reconstruction of each sampled data is calculated. With the goal of minimizing the alignment total loss and the total reconstruction loss, the second attention mechanism network is repeatedly trained until the optimal total fusion feature is obtained. The optimal total fusion feature is clustered to obtain clusters, and each cluster corresponds to a stratum.

[0157] A multi-head attention mechanism is used to fuse the features corresponding to each group of sampled data. Perform dynamic integration. Specifically, for the first... One point of attention, Each vector is mapped to a query vector through a linear transformation. Key vector Sum value vector With the first Using the sampled data as the query center, calculate the first... The first set of sampled data An update of attention head The calculation formula used is:

[0158] ,

[0159] In the formula, This is a scaling factor used to adjust the range of dot product values.

[0160] The outputs of all attention heads are concatenated and then passed through a learnable linear projection matrix. By integrating, we obtain the first Augmented representation of group sampled data :

[0161] ,

[0162] in, This indicates a splicing operation. This indicates the number of attention points.

[0163] To form a globally unified representation, The data are then stitched together and fed into a multilayer perceptron (MLP) for nonlinear fusion and dimensionality reduction to obtain the total fused features. :

[0164] ,

[0165] In the formula, This represents a multilayer perceptron. Indicates a splicing operation;

[0166] For the The reconstruction process of group sampled data is defined as follows:

[0167] ,

[0168] in, Represents the reconstructed first... The data matrix of the group sampled data, This represents a normalized feature adjacency matrix with added self-connections. , Represents the identity matrix. Represents the characteristic adjacency matrix. It is a degree matrix. It is the first The weight matrix of the group sampled data, It is a non-linear activation function.

[0169] Calculate reconstruction loss The calculation formula is:

[0170] ,

[0171] Calculate the overall loss The calculation formula is:

[0172] ;

[0173] Based on comprehensive losses With the goal of minimizing the total features, the second attention mechanism network is continuously trained until the optimal total fusion feature is obtained.

[0174] After training converges, the K-means clustering algorithm is applied to the optimal total fused features to obtain the final cluster labels and corresponding cluster centers.

[0175] Step S60: Analyze and evaluate the engineering properties of the formation based on the clustering results.

[0176] This step specifically includes:

[0177] Step S61: Determination of stratigraphic characteristics.

[0178] Using the cluster centers of each stratum cluster To match a predefined engineering property template. For example, based on the standard penetration test (SPT) number corresponding to the cluster centers. Determining the consistency state of cohesive soil: when When, it is determined to be plastic flow; when When, it is determined to be soft plastic; when When, it is judged to be malleable; when When, it is determined to be hard plastic; when When the soil is hard, it is judged as such; for silt or sand, the standard penetration test blow count is used. Determine density: when At that time, it was judged to be loose. Sometimes slightly denser, when When, it is judged as medium-density; when At that time, it was determined to be dense.

[0179] Step S62: Evaluation of intralayer homogeneity.

[0180] Calculate the first All sampled data within each cluster are routed to its cluster center. Average distance:

[0181] ,

[0182] in, For belonging to the first A collection of sampled data from each cluster, With preset threshold In comparison, if If the result is positive, the strata are considered homogeneous; otherwise, they are considered heterogeneous, suggesting the possible presence of sub-layers or local interlayers.

[0183] Step S63: Evaluation of interlayer homogeneity.

[0184] Calculate the distance between the cluster centers of adjacent clusters , and preset threshold Comparison: If If the two results are positive, it indicates that the physical and mechanical properties of the corresponding strata are similar, and they can be merged into the same engineering geological unit in engineering; otherwise, the original clustering result is maintained.

[0185] Example 2

[0186] Based on the same inventive concept as in Embodiment 1, this embodiment of the invention provides a stratigraphic engineering property analysis and evaluation system based on semantic and spatial information fusion, including a storage medium and a processor;

[0187] The storage medium is used to store instructions;

[0188] The processor is configured to operate according to the instructions to execute the method according to any one of Embodiment 1.

[0189] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0190] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0191] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0192] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0193] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

[0194] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information, characterized in that, include: For the selected borehole, acquire multiple sets of sampling data of different types, as well as the depth coordinate information of each sampling point; Based on multiple sets of sampling data of different types, construct corresponding feature similarity maps; Based on the depth coordinate information of each sampling point, a spatial adjacency graph is constructed; For each feature similarity map and spatial adjacency map, a graph convolutional network is used to extract attribute features and spatial features respectively; A first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data. Alignment learning is performed on the fusion features corresponding to each group of sampled data to obtain the total alignment loss; A second attention mechanism network is used to fuse the fusion features corresponding to each group of sampled data to obtain the total fusion feature. Based on the total fusion feature, multiple sets of sampled data of different types are reconstructed, and the total reconstruction loss before and after reconstruction of each sampled data is calculated. With the goal of minimizing the alignment total loss and the total reconstruction loss, the second attention mechanism network is repeatedly trained until the optimal total fusion feature is obtained. The optimal total fusion features are clustered to obtain clusters, and each cluster corresponds to a type of stratum; Based on the clustering results and predefined analysis strategies, the analysis and evaluation of the engineering properties of the formation are completed. For the The group of sampled data, and the corresponding data matrix is , ,in, The total number of sampling points. For the first The feature dimensions of the group sampled data; the method for constructing the feature similarity map includes: Constructing the first based on the KNN method Feature adjacency matrix corresponding to group sampled data As a feature similarity map, Feature adjacency matrix The formula for calculating the elements is: ， In the formula, Represents the characteristic adjacency matrix elements, and They represent the first Sampling points in group sampling data and sampling points The feature vector formed by the corresponding sample data, Indicates sampling point of The set of feature vectors of the nearest neighbor sampling points; Define sampling points The depth coordinates are The method for constructing the spatial adjacency graph includes: Constructing a spatial adjacency matrix based on the KNN method As a spatial adjacency graph Spatial adjacency matrix The formula for calculating the elements is: ， In the formula, Representing the spatial adjacency matrix elements, Indicates sampling point of The set of depth coordinates of the nearest neighbor sampling points. Sampling points Depth coordinates Sampling points Depth coordinates; The calculation formulas for the attribute features and spatial features are as follows: ， in, ， ， In the formula, For the first The attribute features of the sampled data are represented by the embedding representation reflecting the feature patterns obtained from the feature similarity map; For the first The spatial features of the sampled data represent the embedded representation of the encoded spatial structure information obtained from the spatial adjacency graph; Indicates the first The corresponding normalized feature adjacency matrix of the sampled data group has been self-connected. This represents a normalized spatial adjacency matrix with added self-connections; and All represent degree matrices, Indicates the first The learning weights corresponding to the group of sampled data, This represents the activation function.

2. The method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information as described in claim 1, characterized in that: The types of sampling indicators included in each set of sampled data are different.

3. The method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information as described in claim 1, characterized in that: For the Sampling points in group sampling data The corresponding sample data is defined by the following attribute features: Its spatial characteristics are defined as follows: ; The first attention mechanism network is used to adaptively fuse attribute features and spatial features to obtain fused features corresponding to each group of sampled data, including: Based on the aforementioned attribute features and spatial features Construct splicing features ; The splicing features are obtained through nonlinear transformation. Mapping to a common attention latent space yields intermediate features. intermediate features The calculation formula is: ,in, This is the weight matrix. It is the bias vector; Through affine transformation and normalization operations, first attention weight coefficients corresponding to attribute features and spatial features are generated, respectively. Second attention weight coefficient , In the formula, To map the implicit representation to a weight vector of scalar scores, and for the two weight coefficients to satisfy... , This is a normalization operation; The fusion features corresponding to each group of sampled data are calculated using the following formula: ， ， In the formula, Indicates the first The fusion features corresponding to the group of sampled data Indicates the first Sampling points in group sampling data The corresponding fusion features of the corresponding sample data This represents the total number of sampling points.

4. The method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information as described in claim 1, characterized in that: No. Group sampling data and the first The fusion features corresponding to the group sampled data are as follows: and ,in, The total number of sampling points. The dimension of the latent space; the alignment learning of the fusion features corresponding to each group of sampled data to obtain the total alignment loss includes: Computational fusion features and fusion features Cross-cosine similarity matrix between , , its first Line number Column elements Defined as: ， in, Indicates the first The first group of sampled data The latent representation vector of each sample data; Indicates the first The first group of sampled data The latent representation vector of each sample data, Representing vectors Norm; Define structural alignment loss The calculation formula is: ， In the formula, Represents the given target adjacency matrix; and All are indicator functions, when hour, Otherwise, it is 0; when hour, Otherwise, it is 0. For the target adjacency matrix No. Line number The value of the column; The adjacency matrices of each feature are concatenated to obtain the attribute feature map. , , attribute feature map Substituting the formula for calculating the structural alignment loss, we obtain the feature alignment loss. ; Spatial adjacency matrix Substituting into the formula for calculating structural alignment loss, we obtain the spatial alignment loss as follows: ; Calculate the total alignment loss, the formula for which is: ， In the formula, and All are hyperparameters, and satisfy the following: , , This represents the total number of data sets of different types.

5. The method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information as described in claim 1, characterized in that: The method for calculating the optimal total fusion feature includes: For the second attention mechanism network, the first The first point of attention will be the first Fusion features corresponding to group sampled data Each vector is mapped to a query vector through a linear transformation. Key vector Sum value vector ; With the first Using the sampled data as the query center, calculate the first... The first set of sampled data An update of attention head The calculation formula used is: ， In the formula, Scaling factor The total number of data sets of different types; The outputs of all attention heads are concatenated and then passed through a learnable linear projection matrix. By integrating, we obtain the first Augmented representation of group sampled data : ， in, This indicates a splicing operation. Indicates the number of attention heads; Calculate total fusion features The calculation formula used is: ， In the formula, This represents a multilayer perceptron. Indicates a splicing operation; For the The reconstruction process of group sampled data is defined as follows: ， in, Represents the reconstructed first... The data matrix of the group sampled data, This represents a normalized feature adjacency matrix with added self-connections. , Represents the identity matrix. Represents the characteristic adjacency matrix. It is a degree matrix. It is the first The weight matrix of the group sampled data, It is a non-linear activation function; Calculate reconstruction loss The calculation formula is: ， In the formula, Represents the norm, Represents the first before reconstruction The data matrix of the group sampled data; Calculate the overall loss The calculation formula is: ； Based on comprehensive losses With the goal of minimizing the total features, the second attention mechanism network is continuously trained until the optimal total fusion feature is obtained.

6. The method for analyzing and evaluating the engineering properties of stratigraphy based on the fusion of semantic and spatial information as described in claim 1, characterized in that: The analysis and evaluation of geological engineering properties based on clustering results and predefined analysis strategies includes: Based on each cluster center, a predefined engineering property template is matched to determine the stratigraphic properties; Calculate the first All sampled data within each cluster are routed to its cluster center. Average distance: ， in, For belonging to the first A collection of sampled data from each cluster, With preset threshold In comparison, if If the result is positive, the strata are considered homogeneous; otherwise, they are considered heterogeneous, suggesting the possible existence of sub-layers or local interlayers. Calculate the distance between the cluster centers of adjacent clusters , and preset threshold Comparison: If If the two clusters have similar physical and mechanical properties, they can be merged into the same engineering geological unit in engineering; otherwise, the original clustering result will be maintained.

7. A stratigraphic engineering property analysis and evaluation system based on semantic and spatial information fusion, characterized in that, Including storage media and processor; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-6.

Images