Geotechnical engineering investigation analysis method and system based on multi-source data fusion

Abstract

The invention discloses a geotechnical engineering investigation and analysis method and system based on multi-source data fusion, and belongs to the technical field of geological prospecting, and the method comprises the steps: collecting multi-source heterogeneous data of drilling holes in a karst area, and generating a comprehensive feature vector of the drilling holes; according to the drilling comprehensive feature vector, establishing a graph attention network, and generating an update feature; according to the updated features, generating an exploration prediction result of geotechnical engineering; establishing a dynamic adjustment model, and generating a model weight set value of the graph attention network; adjusting the model weight of the graph attention network according to the model weight set value of the graph attention network; according to the method, the multi-source heterogeneous data of the drill holes are converted into the comprehensive feature vectors of the drill holes, so that the problems of data isomerism and information islands are solved; in combination with a dynamic graph attention network, the model weight is optimized in real time, and the accuracy and adaptivity of karst cave distribution prediction under the complex karst geological condition are remarkably improved; and meanwhile, continuous iterative optimization can be carried out on the model, and the geotechnical engineering investigation accuracy of the karst area is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological exploration, and in particular relates to a geotechnical engineering investigation and analysis method and system based on multi-source data fusion. Background Art

[0002] Geotechnical engineering investigation refers to the exploration and analysis of natural conditions such as overburden, rock, and groundwater at the construction site of a geotechnical engineering project. In karst areas, geological exploration is even more complex and challenging.

[0003] Exploration in karst areas usually uses a variety of means and methods to ensure the accuracy and comprehensiveness of the survey, including geological drilling, borehole television, seismic exploration, resistivity method and ground radar.

[0004] However, in the field of geotechnical engineering investigation, especially geological exploration in karst areas, traditional methods mainly rely on a single data source or simple data superposition for analysis, which has significant technical limitations. Existing technologies generally fail to effectively integrate multi-source heterogeneous data (such as text, image, geospatial, and geophysical data), resulting in low data utilization and prominent information island phenomena. For example, the extraction of structured geological features in core text data is insufficient, the noise interference of image data is not systematically processed, the coordinate system of geospatial data is not standardized, and the fusion mechanism of geophysical data and multi-source information is imperfect. These problems all restrict the comprehensiveness and accuracy of exploration and analysis. In addition, when modeling the correlation between boreholes, traditional graph network models have fixed attention weights or rigid adjustment strategies, and are unable to dynamically optimize model parameters according to actual exploration results, resulting in poor adaptability of prediction results, especially in complex karst geological conditions, where errors are significant. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the present invention provides a geotechnical engineering investigation and analysis method and system based on multi-source data fusion, which solves the above problems.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A geotechnical engineering investigation and analysis method based on multi-source data fusion, comprising the following steps:

[0007] Collect multi-source heterogeneous data from boreholes in karst areas; the multi-source heterogeneous data includes core text data, image data, geospatial data, and geophysical data;

[0008] Generate comprehensive feature vectors of boreholes based on multi-source heterogeneous data of boreholes in karst areas;

[0009] According to the comprehensive feature vector of drilling holes, a graph attention network is established to generate updated features;

[0010] Generate geotechnical engineering exploration prediction results based on the updated features;

[0011] Establish a dynamic adjustment model to generate model weight settings for the graph attention network;

[0012] Adjust the model weights of the graph attention network according to the model weight setting values ​​of the graph attention network.

[0013] On the basis of the above technical solutions, the present invention also provides the following optional technical solutions:

[0014] Further technical solution: The generation method of the update feature specifically includes the following steps:

[0015] Standardize and preprocess multi-source heterogeneous data in karst areas;

[0016] Feature-level data fusion is performed on the standardized multi-source heterogeneous data in the karst area to generate the comprehensive feature vector of the borehole.

[0017] Further technical solution: The feature-level data fusion method is specifically as follows:

[0018] By formula:

[0019] F fused =W text *F text +W image *F image +W space *F space +W geo *F geo ;

[0020] Generate drilling comprehensive feature vector F fused ;

[0021] In the formula, F text It represents the standard value of core text data, F image It represents the standard value of image data, F space It represents the standard value of spatial data, F geo It represents the standard value of geophysical data, W text 、W image 、W space 、W geo are all weight coefficients.

[0022] Further technical solution: The generation method of the update feature specifically includes the following steps:

[0023] The graph structure is constructed with boreholes as nodes and geographical distance or lithologic similarity as edges;

[0024] Establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network to generate updated features;

[0025] Dynamically adjust the attention weights of graph attention networks.

[0026] Further technical solution: The expression of the graph attention network is specifically:

[0027]

[0028] In the expression, h i ′ represents the updated feature, h i Represents node i, h j It represents the node j adjacent to node i, a i，j represents the attention weight of node j to node i, W represents the weight matrix, Ν(i) represents the neighbor set of node i, and a i，i It represents the self-attention weight of node i, γ1 and γ2 are both model weights, and γ1+γ2=1.

[0029] Further technical solution: The method of dynamically adjusting the attention weight of the graph attention network is specifically as follows:

[0030] By formula:

[0031]

[0032] Generate node j’s attention weight a for node i i，j ;

[0033] In the formula, Wh i Wh j Wh k They represent the transformed features of node i and node j respectively, ║ represents the feature concatenation operation, LeakyReLU represents the activation function, and a represents the attention parameter vector; T refers to the transposed form of the attention parameter vector, and k refers to all neighbors of node i;

[0034] According to the attention weight a of node j to node i i，j , dynamically adjust the attention weights of the graph attention network.

[0035] Further technical solution: The method for generating the model weight setting value of the graph attention network specifically includes the following steps:

[0036] Obtain the geotechnical engineering exploration prediction results and actual geotechnical engineering exploration results from historical data to generate accurate exploration times;

[0037] According to the accurate number of explorations, a loss assessment model is established to generate the loss rate of exploration results;

[0038] According to the loss rate of exploration results, a dynamic adjustment model is established to generate the model weight setting value of the graph attention network.

[0039] Further technical solution: The expression of the loss assessment model is specifically:

[0040]

[0041] In the expression, η represents the loss rate of exploration results, M right It represents the exact number of explorations, and M0 represents the total number of explorations.

[0042] Further technical solution: The expression of the dynamic adjustment model is specifically:

[0043]

[0044] In the expression, γ′2 represents the model weight setting value of the graph attention network, γ2 represents the initial setting value of the model weight of the graph attention network, η represents the exploration result loss rate, and η0 represents the exploration result loss rate warning value.

[0045] A geotechnical engineering investigation and analysis system based on multi-source data fusion, the system includes

[0046] The data acquisition module is used to collect multi-source heterogeneous data from boreholes in karst areas; the multi-source heterogeneous data includes core text data, image data, geospatial data, and geophysical data;

[0047] The borehole feature fusion unit is used to generate a comprehensive borehole feature vector based on multi-source heterogeneous data of boreholes in karst areas;

[0048] The feature update unit is used to establish a graph attention network based on the comprehensive feature vector of the drilling hole and generate updated features;

[0049] A prediction module is used to generate geotechnical engineering exploration prediction results based on the updated features;

[0050] An adjustment analysis unit is used to establish a dynamic adjustment model and generate model weight setting values ​​for the graph attention network;

[0051] A model weight adjustment module is used to adjust the model weight of the graph attention network according to the model weight setting value of the graph attention network;

[0052] The drilling feature fusion unit specifically includes:

[0053] Preprocessing module, used for standardized preprocessing of multi-source heterogeneous data in karst areas;

[0054] The fusion module is used to perform feature-level data fusion on the standardized multi-source heterogeneous data in the karst area to generate a comprehensive feature vector of the borehole;

[0055] The updating feature unit specifically includes:

[0056] A graph structure building module is used to construct a graph structure using boreholes as nodes and geographical distance or lithologic similarity as edges;

[0057] Update feature generation module, which is used to establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network, and generate updated features;

[0058] Adjustment module, used to dynamically adjust the attention weights of the graph attention network;

[0059] The adjustment analysis unit specifically includes:

[0060] The historical data acquisition module is used to obtain the exploration prediction results and actual exploration results of geotechnical engineering in the historical data and generate the accurate exploration times;

[0061] The loss analysis module is used to establish a loss assessment model based on the accurate number of explorations and generate the loss rate of exploration results;

[0062] The setting analysis module is used to establish a dynamic adjustment model based on the loss rate of exploration results and generate the model weight setting value of the graph attention network.

[0063] The present invention provides a geotechnical engineering investigation and analysis method and system based on multi-source data fusion, which has the following beneficial effects compared with the existing technology:

[0064] The present invention effectively solves the problems of data heterogeneity and information islands by converting the multi-source heterogeneous data of boreholes into a unified comprehensive feature vector of boreholes. In combination with the dynamic graph attention network, the attention weight is dynamically allocated based on the geographic distance associated with the borehole nodes and the lithological similarity. A loss assessment model is established through feedback from historical exploration data, and the model weights are optimized in real time, which significantly improves the accuracy and adaptability of cave distribution prediction under complex karst geological conditions. At the same time, based on the weighted fusion of neighbor node information and comparative analysis of standard ranges, the risk of local geological anomalies is accurately identified, supplemented by a dynamic adjustment mechanism driven by the loss rate of exploration results, to achieve continuous iterative optimization of the model and improve the accuracy of geotechnical engineering surveys in karst areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 A flowchart of a geotechnical engineering investigation and analysis method based on multi-source data fusion provided by an embodiment of the present invention.

[0066] Figure 2This is a flowchart of step S20 provided in an embodiment of the present invention.

[0067] Figure 3 This is a flowchart of step S30 provided in an embodiment of the present invention.

[0068] Figure 4 This is a flowchart of step S50 provided in an embodiment of the present invention.

[0069] Figure 5 A structural diagram of a geotechnical engineering investigation and analysis system based on multi-source data fusion provided by an embodiment of the present invention.

[0070] Figure 6 This is a module block diagram of the drilling feature fusion unit provided in an embodiment of the present invention.

[0071] Figure 7 This is a module block diagram of the update feature unit provided in an embodiment of the present invention.

[0072] Figure 8 This is a module block diagram of the adjustment analysis unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0074] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0075] See also Figure 1 , a geotechnical engineering investigation and analysis method based on multi-source data fusion provided by an embodiment of the present invention, comprising the following steps:

[0076] Step S10: Collecting multi-source heterogeneous data of boreholes in karst areas; wherein the multi-source heterogeneous data includes core text data, image data, geospatial data and geophysical data;

[0077] In this embodiment, core text data is text data that records structured information such as lithology type and cave development status, such as the number of holes drilled in karst areas and hole labels; image data refers to the image information in images of holes drilled in karst areas that reflects rock structure and fracture distribution through visual features, such as image information; geospatial data refers to data on spatial distribution, such as the coordinates of holes drilled and the distance between holes formed by drilling; geophysical data refers to data that indirectly reflects the physical properties of underground rock masses, such as seismic waves and resistivity;

[0078] Step S20: generating a comprehensive feature vector of the borehole based on the multi-source heterogeneous data of the borehole in the karst area;

[0079] Step S30: Building a graph attention network based on the drilling comprehensive feature vector to generate updated features;

[0080] Step S40: generating geotechnical engineering exploration prediction results based on the updated features;

[0081] Step S50: establishing a dynamic adjustment model to generate model weight setting values ​​for the graph attention network;

[0082] Step S60: Adjust the model weight of the graph attention network according to the model weight setting value of the graph attention network.

[0083] See also Figure 2 As a preferred embodiment of the present invention, the generation method of the update feature specifically includes the following steps:

[0084] S21: Standardization preprocessing of multi-source heterogeneous data in karst areas;

[0085] S22: Perform feature-level data fusion on the standardized multi-source heterogeneous data in the karst area to generate a comprehensive borehole feature vector.

[0086] As a preferred embodiment of the present invention, the method of standardizing multi-source heterogeneous data in karst areas includes:

[0087] Perform standardized preprocessing on image data, including resizing, normalization and noise removal;

[0088] Perform structured processing on core text data to extract key geological features;

[0089] It should be explained that the key geological characteristics include lithology type, cave development status, etc.

[0090] Conduct coordinate system 1 on geospatial data to generate continuous geographic information;

[0091] For example, by establishing a coordinate system, continuous geographic coordinate information is generated.

[0092] As a preferred embodiment of the present invention, the feature-level data fusion method is specifically as follows:

[0093] By formula:

[0094] F fused =W text *F text +W image *F image +W space *Fspace +W geo *F geo ;

[0095] Generate drilling comprehensive feature vector F fused ;

[0096] In the formula, F text It represents the standard value of core text data, F image It represents the standard value of image data, F space It represents the standard value of spatial data, F geo It represents the standard value of geophysical data, W text 、W image 、W space 、W geo All are weight coefficients;

[0097] It needs to be explained that W text 、W image 、W space 、W geo The value of is determined by relevant personnel in this field, and the method of determining the value includes but is not limited to expert consultation method, etc.

[0098] See also Figure 3 As a preferred embodiment of the present invention, the generation method of the update feature specifically includes the following steps:

[0099] S31: Construct a graph structure with boreholes as nodes and geographical distance or lithologic similarity as edges;

[0100] S32: Establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network, and generate updated features;

[0101] S33: Dynamically adjust the attention weights of the graph attention network.

[0102] As a preferred embodiment of the present invention, the expression of the graph attention network is specifically:

[0103]

[0104] In the expression, h i ′ represents the updated feature, h i Represents node i, h j It represents the node j adjacent to node i, a i，j represents the attention weight of node j to node i, W represents the weight matrix, Ν(i) represents the neighbor set of node i, and a i，i It represents the self-attention weight of node i, γ1 and γ2 are both model weights, and γ1+γ2=1;

[0105] It should be explained that the node node Both can be understood as the comprehensive feature vector F of the drilling fused ; For example, the node node It can be understood as drilling 1 and drilling 2, then the node Refers to the borehole comprehensive feature vector F generated by the multi-source heterogeneous data of borehole 1 fused Similarly, node Refers to the borehole comprehensive feature vector F generated by the multi-source heterogeneous data of borehole 2 fused ;

[0106] Specifically, the neighbor set N(i) of node i includes node i itself;

[0107] Specifically, update the feature h i ′ refers to the updated features of node i, which integrates its own features and the weighted information of neighboring nodes;

[0108] In this embodiment, interactive fusion of multiple boreholes is achieved by fusing data from multiple boreholes; for example, borehole 1 and borehole 2 both have caves and are close to each other. The model infers that the area between borehole 1 and borehole 2 has caves, and the closer the distance, the higher the possibility that the area between borehole 1 and borehole 2 has caves.

[0109] As a preferred embodiment of the present invention, the method of dynamically adjusting the attention weight of the graph attention network is specifically as follows:

[0110] By formula:

[0111]

[0112] Generate node j’s attention weight a for node i i，j ;

[0113] In the formula, Wh i Wh j They represent the transformed features of node i and node j respectively, ║ represents the feature concatenation operation, LeakyReLU represents the activation function, and a represents the attention parameter vector; T refers to the transposed form of the attention parameter vector, and k refers to all neighbors of node i;

[0114] It needs to be explained that Wh i Wh j Wh k It is transformed by the weight matrix W.

[0115] As a preferred embodiment of the present invention, the geotechnical engineering exploration prediction results are generated in the following manner:

[0116] The feature h will be updated i 'Compare with the standard range;

[0117] If we update the feature h i If ′ is within the standard range, it is predicted that there are no karst caves in the rock and soil of the area;

[0118] If we update the feature h i If ′ is within the non-standard range, it is predicted that there are caves in the rock and soil in this area.

[0119] See also Figure 4 As a preferred embodiment of the present invention, the method for generating the model weight setting value of the graph attention network specifically includes the following steps:

[0120] S51: Obtaining the exploration prediction results and actual exploration results of geotechnical engineering in historical data, and generating accurate exploration times;

[0121] It should be explained that the number of accurate explorations refers to the number of times that the predicted results of geotechnical exploration are consistent with the actual results of geotechnical exploration. In addition, the predicted results of geotechnical exploration and the actual results of geotechnical exploration are both historical data.

[0122] S52: Based on the accurate number of explorations, a loss assessment model is established to generate the loss rate of the exploration results;

[0123] S53: Based on the loss rate of the exploration results, a dynamic adjustment model is established to generate the model weight setting value of the graph attention network.

[0124] As a preferred embodiment of the present invention, the expression of the loss assessment model is specifically:

[0125]

[0126] In the expression, η represents the loss rate of exploration results, M right It represents the exact number of explorations, and M0 represents the total number of explorations;

[0127] It should be explained that the total number of explorations refers to the total number of geotechnical engineering exploration prediction results in historical data.

[0128] As a preferred embodiment of the present invention, the expression of the dynamic adjustment model is specifically:

[0129]

[0130] In the expression, γ′2 represents the model weight setting value of the graph attention network, γ2 represents the initial setting value of the model weight of the graph attention network, η represents the exploration result loss rate, and η0 represents the exploration result loss rate warning value;

[0131] It should be explained that the exploration result loss rate warning value η0 refers to the maximum value of the exploration result loss rate η under standard conditions; the exploration result loss rate warning value η0 can be understood as the error of the exploration prediction results under standard conditions, that is, the maximum value of the exploration result loss rate allowed under standard conditions.

[0132] See also Figure 5 The present invention also provides a geotechnical engineering investigation and analysis system based on multi-source data fusion, the system comprising:

[0133] The data acquisition module 10 is used to collect multi-source heterogeneous data of boreholes in karst areas; wherein the multi-source heterogeneous data includes core text data, image data, geospatial data and geophysical data;

[0134] The borehole feature fusion unit 20 is used to generate a borehole comprehensive feature vector based on multi-source heterogeneous data of the boreholes in the karst area;

[0135] An updating feature unit 30 is used to establish a graph attention network based on the drilling comprehensive feature vector and generate an updated feature;

[0136] A prediction module 40 is used to generate geotechnical engineering exploration prediction results based on the updated features;

[0137] An adjustment analysis unit 50 is used to establish a dynamic adjustment model and generate model weight setting values ​​of the graph attention network;

[0138] The model weight adjustment module 60 is used to adjust the model weight of the graph attention network according to the model weight setting value of the graph attention network.

[0139] See also Figure 6 As a preferred embodiment of the present invention, the drilling feature fusion unit specifically includes:

[0140] The preprocessing module 21 is used to perform standardized preprocessing on multi-source heterogeneous data in karst areas;

[0141] The fusion module 22 is used to perform feature-level data fusion on the standardized multi-source heterogeneous data in the karst area to generate a comprehensive borehole feature vector.

[0142] See also Figure 7 As a preferred embodiment of the present invention, the updating feature unit specifically includes:

[0143] A graph structure building module 31 is used to build a graph structure using boreholes as nodes and geographical distances or lithologic similarities as edges;

[0144] An update feature generation module 32 is used to establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network, and generate an update feature;

[0145] The adjustment module 33 is used to dynamically adjust the attention weight of the graph attention network.

[0146] See also Figure 8 As a preferred embodiment of the present invention, the adjustment analysis unit specifically includes:

[0147] The historical data acquisition module 51 is used to obtain the exploration prediction results and actual exploration results of geotechnical engineering in the historical data and generate the accurate exploration times;

[0148] The loss analysis module 52 is used to establish a loss assessment model based on the accurate number of explorations and generate a loss rate of exploration results;

[0149] The setting analysis module 53 is used to establish a dynamic adjustment model based on the loss rate of the exploration results and generate the model weight setting value of the graph attention network.

[0150] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A geotechnical engineering investigation and analysis method based on multi-source data fusion, characterized in that: The following steps are involved: Collect multi-source heterogeneous data from boreholes in karst areas; the multi-source heterogeneous data includes core text data, image data, geospatial data, and geophysical data; Generate comprehensive feature vectors of boreholes based on multi-source heterogeneous data of boreholes in karst areas; According to the comprehensive feature vector of drilling holes, a graph attention network is established to generate updated features; Generate geotechnical engineering exploration prediction results based on the updated features; Establish a dynamic adjustment model to generate model weight settings for the graph attention network; Adjust the model weights of the graph attention network according to the model weight setting values ​​of the graph attention network.

2. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 1, characterized in that: The generation method of the update feature specifically includes the following steps: Standardize and preprocess multi-source heterogeneous data in karst areas; Feature-level data fusion is performed on the standardized multi-source heterogeneous data in the karst area to generate the comprehensive feature vector of the borehole.

3. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 2 is characterized in that: The method of feature-level data fusion is specifically as follows: By formula: F fused =W text *F text +W image *F image +W space *F space +W geo *F geo ; Generate drilling comprehensive feature vector F fused ; In the formula, F text It represents the standard value of core text data, F image It represents the standard value of image data, F space It represents the standard value of spatial data, F geo It represents the standard value of geophysical data, W text 、W image 、W space 、W geo are all weight coefficients.

4. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 1 is characterized in that: The generation method of the update feature specifically includes the following steps: The graph structure is constructed with boreholes as nodes and geographical distance or lithologic similarity as edges; Establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network to generate updated features; Dynamically adjust the attention weights of graph attention networks.

5. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 4 is characterized in that: The expression of the graph attention network is specifically: In the expression, h i ′ represents the updated feature, h i Represents node i, h j It represents the node j adjacent to node i, a i，j represents the attention weight of node j to node i, W represents the weight matrix, Ν(i) represents the neighbor set of node i, and a i，i It represents the self-attention weight of node i, γ1 and γ2 are both model weights, and γ1+γ2=1.

6. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 4 is characterized in that: The method of dynamically adjusting the attention weight of the graph attention network is specifically as follows: By formula: Generate node j’s attention weight a for node i i，j ; In the formula, Wh i Wh j Wh k They represent the transformed features of node i and node j respectively, ║ represents the feature concatenation operation, LeakyReLU represents the activation function, and a represents the attention parameter vector; T refers to the transposed form of the attention parameter vector, and k refers to all neighbors of node i; According to the attention weight a of node j to node i i，j , dynamically adjust the attention weights of the graph attention network.

7. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 1, characterized in that: The method for generating the model weight setting value of the graph attention network specifically includes the following steps: Obtain the geotechnical engineering exploration prediction results and actual geotechnical engineering exploration results from historical data to generate accurate exploration times; According to the accurate number of explorations, a loss assessment model is established to generate the loss rate of exploration results; According to the loss rate of exploration results, a dynamic adjustment model is established to generate the model weight setting value of the graph attention network.

8. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 7 is characterized in that: The expression of the loss assessment model is specifically: In the expression, η represents the loss rate of exploration results, M right It represents the exact number of explorations, and M0 represents the total number of explorations.

9. The geotechnical engineering investigation and analysis method based on multi-source data fusion according to claim 1, characterized in that: The expression of the dynamic adjustment model is specifically: In the expression, γ′2 represents the model weight setting value of the graph attention network, γ2 represents the initial setting value of the model weight of the graph attention network, η represents the exploration result loss rate, and η0 represents the exploration result loss rate warning value.

10. A geotechnical engineering investigation and analysis system based on multi-source data fusion, characterized in that: The system is used to execute a geotechnical engineering investigation and analysis method based on multi-source data fusion as described in any one of claims 1 to 9; the system comprises The data acquisition module is used to collect multi-source heterogeneous data from boreholes in karst areas; the multi-source heterogeneous data includes core text data, image data, geospatial data, and geophysical data; The borehole feature fusion unit is used to generate a comprehensive borehole feature vector based on multi-source heterogeneous data of boreholes in karst areas; The feature update unit is used to establish a graph attention network based on the comprehensive feature vector of the drilling hole and generate updated features; A prediction module is used to generate geotechnical engineering exploration prediction results based on the updated features; An adjustment analysis unit is used to establish a dynamic adjustment model and generate model weight setting values ​​for the graph attention network; A model weight adjustment module is used to adjust the model weight of the graph attention network according to the model weight setting value of the graph attention network; The drilling feature fusion unit specifically includes: Preprocessing module, used for standardized preprocessing of multi-source heterogeneous data in karst areas; The fusion module is used to perform feature-level data fusion on the standardized multi-source heterogeneous data in the karst area to generate a comprehensive feature vector of the borehole; The updating feature unit specifically includes: A graph structure building module is used to construct a graph structure using boreholes as nodes and geographical distance or lithologic similarity as edges; Update feature generation module, which is used to establish a graph attention network, substitute the drilling comprehensive feature vector into the graph attention network, and generate updated features; Adjustment module, used to dynamically adjust the attention weights of the graph attention network; The adjustment analysis unit specifically includes: The historical data acquisition module is used to obtain the exploration prediction results and actual exploration results of geotechnical engineering in the historical data and generate the accurate exploration times; The loss analysis module is used to establish a loss assessment model based on the accurate number of explorations and generate the loss rate of exploration results; The setting analysis module is used to establish a dynamic adjustment model based on the loss rate of exploration results and generate the model weight setting value of the graph attention network.

Images