A method, medium, equipment and product for identifying geological anomalies in tunnels.

Abstract

This invention discloses a method, medium, equipment, and product for identifying tunnel geological anomalies, relating to the field of tunnel geological anomaly identification. The method includes: mapping multimodal geological data to a unified vector space, extracting entities, relationships, and entity attributes, and constructing a multimodal knowledge graph; acquiring the current engineering geological background and exploration data, performing matching calculations with the multimodal knowledge graph to obtain the correlation value between the current data and the multimodal knowledge graph; selecting the N knowledge graph nodes and their relationships with the highest correlation to the current data, converting them into structured text descriptions in an entity-relationship-answer format, and combining this with the current exploration data and the tunnel geological anomaly identification question to be solved to form a structured prompt for this query; inputting this prompt into a large language model to obtain the answer to the tunnel geological anomaly identification question. The tunnel geological anomaly identification method of this invention has strong adaptive optimization capabilities.

Description

Technical Field

[0001] This invention relates to the field of tunnel geological anomaly identification technology, and in particular to a method, medium, equipment and product for identifying tunnel geological anomalies. Background Technology

[0002] Tunnels and underground engineering projects are critical infrastructure in transportation, water conservancy, mining, and other fields. However, during construction, geological disasters such as sudden water inrush and mudslides seriously threaten construction safety, affect project progress, and may cause significant economic losses and casualties. These disasters often originate from unfavorable geological bodies such as water-rich fractured zones and karst channels hidden in front of the tunnel face. Therefore, accurate assessment of geological anomalies in front of the tunnel face (especially their water-richness and degree of fracture) is a prerequisite for effective disaster forecasting and prevention.

[0003] Currently, advanced geological prediction technologies for tunnel construction have developed into various methods, mainly divided into direct detection methods (such as advanced horizontal drilling) and indirect geophysical detection methods (such as tunnel seismic prediction (TSP), transient electromagnetic method (TEM), and ground-penetrating radar method (GPR)). Furthermore, the concept of combining multiple methods for joint detection is gradually becoming more widespread. Despite the continuous enrichment of the technological system, existing technologies still face a series of severe challenges in practical applications, mainly manifested in the following three limitations: First, insufficient handling of data heterogeneity and difficulty in multi-source information fusion. Different geophysical methods such as TSP, TEM, and GPR have significant differences in their physical basis, observation methods, output parameters (such as seismic wave velocity, apparent resistivity, and electromagnetic wave reflection characteristics), and detection scales (from tens to hundreds of meters). This results in highly heterogeneous data from different devices in terms of format, dimensions, physical meaning, and spatial resolution. Although the concept of "integrated forecasting" has been proposed for many years, in practice, the lack of a unified and efficient preprocessing process and data alignment standards (such as precise spatiotemporal registration and normalization methods) makes it difficult to effectively integrate heterogeneous data into a unified interpretation model. This prevents the full utilization of the complementary advantages of multi-source data and may even increase interpretation uncertainty due to information conflicts. Secondly, the interpretation process is highly dependent on human experience, making standardization and quantification difficult. Currently, the determination of the properties of geophysical anomalies (such as whether they are water-rich bodies) and their parameters (such as the degree of water richness and fragmentation) largely depends on the professional knowledge and practical experience of interpreters. This experience-based judgment is highly subjective and lacks a structured knowledge system (such as an expert rule base and a typical geological model library) and quantitative identification model support. This leads to different conclusions drawn by different personnel on the same set of data, making it difficult to guarantee the reliability and consistency of interpretation results, and even more difficult to achieve efficient knowledge transfer and large-scale application. Although data-driven methods such as machine learning have been attempted, they often neglect the integration of prior knowledge such as geological genesis mechanisms, resulting in insufficient interpretability. Finally, the lack of dynamic feedback and adaptive optimization mechanisms leads to poor model adaptability. Advanced geological forecasting is a dynamic and progressive process. Drilling, as a direct means of exposure, can provide the most reliable verification data. However, in existing technological processes, the "true ground value" data obtained through drilling is usually only used for post-exposure verification and fails to be systematically and in real-time fed back to the earlier geophysical data interpretation stage. Once a geophysical interpretation model (such as anomaly threshold setting) is established, it often lacks the ability to automatically correct and optimize based on subsequent excavation exposure or drilling verification results. This lack of a closed-loop feedback mechanism of "detection-verification-model update" makes it difficult for the forecasting system to adapt to the specific geological conditions of different work areas, restricting its accuracy and robustness in long-term application. In addition, existing knowledge management relies heavily on unstructured document storage (such as reports and rule texts), lacking knowledge graph-driven semantic modeling, making it difficult for knowledge to be automatically reasoned and reused by machines, further exacerbating the subjectivity of the interpretation process.

[0004] In summary, while current advanced geological prediction technologies for tunnel construction are becoming increasingly diverse in terms of methodology, they still exhibit significant shortcomings in the depth of data fusion, the objectivity and standardization of the interpretation process, and the system's self-learning and evolutionary capabilities. These limitations have become key bottlenecks restricting further improvements in the accuracy of water inrush and mudslide disaster prediction. Therefore, there is an urgent need in this field for a new collaborative interpretation method that can deeply integrate multidimensional heterogeneous data, embed structured domain knowledge, and possess dynamic feedback and adaptive optimization capabilities. This method would overcome the limitations of existing technologies and provide more reliable technical support for achieving safe tunnel construction. Summary of the Invention

[0005] The purpose of this invention is to address the lack of adaptive optimization capabilities in current advanced geological prediction methods for tunnel construction, and to propose a method for identifying geological anomalies in tunnels, comprising the following steps:

[0006] S1. Map multimodal geological data to a unified vector space, extract entities, relationships, and entity attributes, and construct a multimodal knowledge graph;

[0007] S2. Obtain the current engineering geological background and exploration data, and perform matching calculations with the multimodal knowledge graph to obtain the correlation value between the current data and the multimodal knowledge graph;

[0008] S3. Select the N knowledge graph nodes and their relationships that are most relevant to the current data, and transform them into structured text descriptions. Together with the current detection data and the tunnel geological anomaly identification problem to be solved, these text descriptions constitute the structured prompts for this query. Input them into the large language model to obtain the answer to the tunnel geological anomaly identification problem to be solved.

[0009] Furthermore, the entity types include: detection methods, geological anomalies, and decision rules; the attributes of detection methods include: detection method name, effective detection distance, and detection parameters; the attributes of geological anomalies include: location, size, water abundance, and degree of fragmentation; the attributes of decision rules include: rule name, preconditions, execution actions, rule description, and associated flowchart steps; detection methods include TSP, TEM, GPR, and borehole.

[0010] Furthermore, the decision-making rules include detection method rules, which are as follows:

[0011] First, execute the TSP test. If the TSP test result is normal, proceed with construction as normal; if it is abnormal, execute the TEM test.

[0012] If the TEM detection results are normal, proceed with construction as normal; if abnormal, execute GPR.

[0013] If the GPR detection results are normal, proceed with construction as normal; if abnormal, proceed with drilling.

[0014] Geological anomalies were identified based on borehole imaging, and advanced support was implemented.

[0015] Furthermore, the method for calculating the correlation value between the current data and the multimodal knowledge graph is as follows:

[0016]

[0017] in, This represents the correlation score between the current data and the multimodal knowledge graph. and express , and Weighting coefficients Indicates the similarity of entity attributes. Indicates the connectivity of relational paths. Indicates the matching degree of historical cases;

[0018] The formula for calculating entity attribute similarity is as follows:

[0019]

[0020] in, express and cosine similarity, A vector representing the current probe data; Represents the entity attribute vector in a knowledge graph;

[0021] Select entity attributes from the knowledge graph and match them with the current probe data vector. The entity with the largest value is taken as the associated entity of the current probe data vector. The formula for calculating the connectivity of the relationship path is as follows:

[0022]

[0023] in, pw represents the number of hops in the shortest path from the associated entity of the current probe data to the target entity in the knowledge graph, and pw represents the weight product of all relation edges on the shortest path.

[0024] The formula for calculating the matching degree of historical cases is as follows:

[0025]

[0026] in, and These are weighting coefficients. This represents the feature vector of historical cases in a knowledge graph. This represents the current engineering geological background vector. This represents the geological background vector of historical cases in the knowledge graph.

[0027] Furthermore, , and The dynamic weighting coefficients are calculated as follows:

[0028]

[0029]

[0030]

[0031]

[0032] in, , and yes , and The interaction score with query q, where query vector q represents the features of the current probe data. , , and For learnable parameters, , and These are entity vectors, relation vectors, and historical case context vectors obtained from the current engineering geological background and exploration data based on the knowledge graph. , and They are , and Attention weights , and They represent and In the initial dynamic weights at the current time step, λ is the fusion coefficient. , and They represent and Historical values ​​from the previous time step.

[0033] Furthermore, and The historical value update method is as follows:

[0034]

[0035]

[0036]

[0037] in, , and They represent and The historical values ​​of the first two time steps, Indicates the learning rate. , and They represent , and Feedback signals.

[0038] Furthermore, , , and These are aligned feature vectors extracted from ground-penetrating radar data, transient electromagnetic method data, and borehole descriptions, respectively. This indicates vector concatenation.

[0039] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for identifying tunnel geological anomalies.

[0040] The present invention also proposes an electronic device, including a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to execute the above-described method for identifying tunnel geological anomalies.

[0041] The present invention also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-described method for identifying tunnel geological anomalies.

[0042] The beneficial effects of the technical solution provided by this invention are:

[0043] This invention constructs a multimodal knowledge graph, integrating data from four detection methods—TSP, TEM, GPR, and borehole drilling—along with historical data from each. Detection is performed progressively according to the detection distance of each method, from farthest to closest. Prior knowledge and reasoning rules are provided. The correlation between the current data and the multimodal knowledge graph is measured through three aspects: entity attribute similarity, relational path connectivity, and historical case matching degree. Finally, by fusing multimodal features with the current detection data and the unresolved tunnel geological anomaly identification problem, a more accurate structured prompt is constructed. A dynamic feedback mechanism adjusts the weights of entity attribute similarity, relational path connectivity, and historical case matching degree, achieving accurate identification of geological anomalies. The tunnel geological anomaly identification method of this invention has strong adaptive optimization capabilities, significantly improving the automation, reliability, and adaptability of advanced geological prediction. Attached Figure Description

[0044] Figure 1 This is a flowchart of the tunnel geological anomaly identification method according to an embodiment of the present invention;

[0045] Figure 2 This is a block diagram of an electronic device according to an exemplary embodiment of Embodiment 1 of the present invention;

[0046] Figure 3 This is a map showing the results of GPR radar detection in a geological forecast report for a certain tunnel project. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0048] The flowchart of the tunnel geological anomaly identification method according to an embodiment of the present invention is as follows: Figure 1 Specifically, it includes the following steps:

[0049] S1. Map multimodal geological data to a unified vector space, extract entities, relationships, and entity attributes, and construct a multimodal knowledge graph.

[0050] The data includes: multimodal geological data in front of the tunnel face, including images, waveforms, numerical values, and text, from project archives, geological forecast reports of historical tunnel projects, construction logs, as-built geological summaries, and accident analysis reports.

[0051] Entities include: detection methods (TSP, TEM, GPR, and boreholes), geological anomalies (such as water-rich faults and karst caves), detection data (such as TSP wave velocity and transient electromagnetic apparent resistivity), prevention and control measures (such as grouting parameters and support schemes), and decision-making rules. The knowledge graph uses geological anomalies as its core nodes.

[0052] The attributes of the detection method include: detection method name, effective detection distance, and detection parameters; the attributes of the geological anomaly include: location, size, water abundance, and degree of fragmentation; the attributes of the decision rule include: rule name, preconditions, execution actions, rule description, and associated flowchart steps.

[0053] Among the four advanced geological prediction and detection methods—TSP, TEM, GPR, and borehole—TSP has a detection range of 100m and is used for preliminary groundwater detection; TEM has a detection range of 30-80m and is used to verify the existence of groundwater and geological anomalies; GPR has a detection range of 20-30m and is used to determine the specific water-bearing area and water content; and ultra-short distance (≤20m) detection is carried out through boreholes, using borehole cores and in-hole imaging for direct observation to identify unfavorable geological bodies ahead.

[0054] The decision rules include the detection method rules, which are as follows:

[0055] First, execute the TSP test. If the TSP test result is normal, proceed with construction as normal; if it is abnormal, execute the TEM test.

[0056] If the TEM detection results are normal, proceed with construction as normal; if abnormal, execute GPR.

[0057] If the GPR detection results are normal, proceed with construction as normal; if abnormal, proceed with drilling.

[0058] Geological anomalies were identified based on borehole imaging, and advanced support was implemented.

[0059] It is also necessary to establish rules for interpreting data into attributes. These rules are expressed in the form of "IF-THEN". For example, the following rules:

[0060] IF TSP analysis showed that the longitudinal wave velocity Vp of the rock mass decreased by >15% AND Poisson's ratio σ>0.3, thus inferring that the rock mass was highly fragmented and may contain fluid.

[0061] The IF TEM inversion results show that the apparent resistivity ρs < 200 Ω·m and the decay curve is flat, and THEN infers that the water-bearing capacity of this region is high.

[0062] The IF GPR image shows a strong amplitude, negative phase reflection interface, and THEN infers the presence of a distinct water-bearing interface, the outline of which can be delineated from the image.

[0063] S2. Obtain the current engineering geological background and exploration data, and perform matching calculations with the multimodal knowledge graph to obtain the correlation value between the current data and the multimodal knowledge graph. The engineering geological background includes the surrounding rock grade, surrounding rock integrity, surrounding rock hardness, spatial distribution of structural planes, and groundwater enrichment. Exploration data includes TSP, TEM, GPR, and borehole exploration data.

[0064] Acquire the current engineering TEM, GPR, and borehole detection data. Under aligned time and space coordinates, convert different types of detection data into a unified feature vector to obtain the data vector V. input =[V GPR ;V TEM ;V Drill ]. Where V GPR It is a feature vector extracted from ground-penetrating radar (GPR) image slices using a pre-trained CNN (such as ResNet); V TEM It is the characteristic vector of the apparent resistivity curve of the transient electromagnetic method (TEM); V Drill It is the feature vector obtained by a text encoder (such as BERT) from the borehole description text.

[0065] The correlation between the current data and the multimodal knowledge graph is determined by the similarity of entity attributes. Relationship path connectivity and historical case matching degree The three aspects are measured, and the specific calculation method is as follows.

[0066] Will The cosine similarity is calculated between the attribute vectors of entity nodes in the knowledge graph and the attribute vectors of the entity nodes. The specific calculation formula is as follows:

[0067]

[0068] in, express and cosine similarity, A vector representing the current probe data; This represents the entity attribute vector in the knowledge graph.

[0069] Select entity attributes from the knowledge graph and match them with the current probe data vector. The entity with the largest value is taken as the associated entity of the current probe data vector. The formula for calculating the connectivity of the relationship path is as follows:

[0070]

[0071] in, This represents the number of hops along the shortest path from the associated entity in the current probe data to the target entity in the knowledge graph, where pw represents the weight product of all relation edges along that shortest path. The shorter the path, the higher the weight, and the greater the correlation.

[0072] The formula for calculating the matching degree of historical cases is as follows:

[0073]

[0074] in, and These are weighting coefficients. This represents the feature vector of historical cases in a knowledge graph. This represents the current engineering geological background vector. This represents the geological background vector of historical cases in the knowledge graph.

[0075] The weighted sum of the three factors above yields the correlation value between the current data and the multimodal knowledge graph. The calculation method is as follows:

[0076]

[0077] in, This represents the correlation score between the current data and the multimodal knowledge graph. and express , and Weighting coefficients Indicates the similarity of entity attributes. It represents the connectivity of relational paths, used to measure the similarity between the current data and the typical physical properties of a specific geological entity in the knowledge graph. This indicates the matching degree of historical cases, measuring the similarity between the current working conditions and historical cases in the knowledge graph.

[0078] and The weighting coefficients are dynamic and change with the number of inference iterations, satisfying the following conditions: . , and The specific calculation method is as follows:

[0079] Query the entity vectors corresponding to the current engineering geological background and exploration data in the knowledge graph. relation vectors and historical case context vectors For each vector, calculate the interaction score between query q and the vector:

[0080]

[0081] in, , and yes , and The interaction score with query q, where query vector q represents the features of the current probe data. , , and These are learnable parameters.

[0082] Will , and Normalization yields the attention weights:

[0083]

[0084] in, , and They are , and Attention weights, satisfying .

[0085] Load the weights from the previous time step from the history record. , and To reflect long-term learning performance, the initial values ​​are all 1 / 3. After each inference, the historical weights are updated based on the feedback signal. Combining short-term attention and long-term historical components, the initial dynamic weights are calculated.

[0086]

[0087] in, , and They represent and In the initial dynamic weights at the current time step, λ is the fusion coefficient (set to 0.5) to control the balance between short-term and long-term. , and They represent and Historical values ​​from the previous time step.

[0088] Normalize the initial dynamic weights at the current time step to obtain and Value:

[0089]

[0090] Update and optimize historical weights based on feedback:

[0091]

[0092]

[0093]

[0094] in, , and They represent and The historical values ​​of the first two time steps, This represents the learning rate, which is set to 0.05 in this invention. , and They represent , and The actual feedback signal.

[0095] when , , , , , The comparison entity node in the knowledge graph is: water-rich body. The correlation value between the current data and the multimodal knowledge graph is 0.795. If the threshold θ=0.75, then the water-rich body related rule is successfully filtered.

[0096] S3. Select the N knowledge graph nodes and their relationships with the highest relevance to the current data, and convert them into structured text descriptions in the format of entity-relationship-answer. These descriptions, along with the current detection data and the unresolved tunnel geological anomaly identification problem, constitute the structured hints for this query. Input these hints into the large language model to obtain the answer to the unresolved tunnel geological anomaly identification problem. Use the generated structured hints to perform supervised fine-tuning of the multimodal large language model, enabling the model's internal weights to learn and memorize professional knowledge in the field of geological prediction.

[0097] For example, the generated structured hints might be:

[0098] Given the following geological rules and cases: [Rule 1] When the GPR image shows strong amplitude and negative phase reflection at the tunnel face arch [Feature A], and the corresponding TEM data shows low resistivity [Feature B], historical cases (correlation degree 0.92) indicate that it is a high-pressure, water-rich area. [Rule 2] Current data: The GPR image shows [Feature A] at location X, and the TEM shows [Feature B] at the corresponding location. Please analyze the current geological risk based on the above rules.

[0099] In one exemplary embodiment, a computer-readable storage medium is included, which stores a computer program that, when executed by a processor, implements the above-described method for identifying tunnel geological anomalies.

[0100] Please see Figure 2 In one exemplary embodiment, the device further includes an electronic device including at least one processor, at least one memory, and at least one communication bus.

[0101] The memory stores a computer program, which includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the aforementioned method for identifying tunnel geological anomalies.

[0102] In one exemplary embodiment, a computer program product is proposed, including a computer program / instructions that, when executed by a processor, implement the steps of the tunnel geological anomaly identification method described above.

[0103] Figure 3 This is a GPR radar detection result map from a geological forecast report for a tunnel project. The horizontal axis represents the amplitude along the tunnel axis, and the vertical axis represents the detection depth (unit: meters). The density waveform in the map indicates the geological conditions ahead of the tunnel face. Based on manual assessment, the rock mass within the red box is fractured to relatively fractured, with poor self-stabilizing ability, containing fissure water, and exhibiting well-developed weak interlayers and slip zones. Significant collapses are likely to occur at the arch and sidewalls during excavation; therefore, timely reinforcement of initial support is recommended.

[0104] Will Figure 3 The results of GPR radar detection are input into a large language model for analysis, and the following questions are posed: "This is a GPR radar image. What conclusions can be drawn, and what are the next steps?"

[0105] The conclusions drawn from the analysis of the large language model obtained by the method of this invention are as follows:

[0106] Yes, this GPR radar chart needs further analysis and interpretation.

[0107] Although the image itself presents a typical ground-penetrating radar profile (horizontal axis represents scanning distance, and vertical axis represents depth), geological conclusions cannot be drawn directly from this image alone. Based on the provided knowledge base, this type of data requires a systematic "cascaded interpretation process," in which GPR is a key step for "short-range" fine characterization.

[0108] What should be done next?

[0109] 01. Preliminary identification of abnormal features:

[0110] Observe the reflection pattern and amplitude: Check for typical water-bearing signals such as "strong amplitude," "negative phase," or "low frequency" (as described in the knowledge base). Analyze structural disorder: Dense, disordered reflection lines in the image may indicate fractured zones in the rock mass, which is consistent with the description of "disordered and discontinuous in-phase axes."

[0111] Find interface features: Locate the locations where reflected waves are continuous or discontinuous to determine whether there are faults or structural surfaces.

[0112] 02. Collaborative interpretation using other data:

[0113] Correlate TSP and transient electromagnetic data: If TSP (decreased P-wave velocity, increased Poisson's ratio) or transient electromagnetic (low-resistivity anomaly) results for the area are known, compare the reflection characteristics of the ground-penetrating radar (GPR) with these data. For example, if the TSP identifies a "suspected fluid-bearing fractured zone," the presence of strong amplitude and negative phase signals in the GPR further supports this judgment. Combine ground-penetrating radar interpretation criteria: Based on the criteria in the knowledge base (such as in step three), analyze whether it conforms to the typical characteristics of a "high water-rich risk area" or a "water-rich fractured zone."

[0114] 03. Conduct drilling verification if necessary:

[0115] If the anomaly identified by GPR is high-risk (e.g., strong amplitude, negative phase, chaotic structure, and located in a critical construction area), advanced horizontal drilling must be carried out to obtain the "true ground value".

[0116] Drilling will provide quantitative parameters such as RQD and water inflow to verify the interpretation results of GPR and provide precise targets for subsequent engineering measures such as grouting.

[0117] In summary, this map serves as a starting point for the analysis, but its geological significance can only be ultimately determined through further feature identification, multi-source data collaboration, and, if necessary, field verification. Therefore, the next step is crucial and cannot be omitted.

[0118] The output of a typical large language model is:

[0119] Yes, this GPR (Ground-penetrating radar) image shows reflected signals from underground structures, but it currently only presents raw data without any processing or interpretation. Whether a "next step" is needed depends on your objectives.

[0120] 01. If the purpose is preliminary observation: the current image is sufficient to identify some obvious layered structures and anomalous bodies.

[0121] 02. If the purpose is to conduct a professional analysis, then proceeding to the next step is strongly recommended:

[0122] Data processing: such as denoising, filtering (e.g., frequency filtering or wavelet transform), gain correction, etc., to enhance signal-to-noise ratio and target resolution. Image interpretation: using geological knowledge to determine the meaning of different reflective layers (e.g., surface, soil layer, bedrock, buried objects, etc.). Quantitative analysis: measuring depth, thickness, or calculating volume.

[0123] Therefore, further data processing and interpretation are necessary to obtain clearer and more valuable information.

[0124] I suggest you decide which specific steps are needed based on your research objectives.

[0125] In summary, the output of typical large-scale models is completely inadequate to provide judgments comparable to those made by humans. They merely indicate shortcomings and suggest further data processing and interpretation, without explaining how to proceed or offering specific advice. In contrast, the output of this invention can analyze image conditions, provide qualitative judgments, explain interpretation criteria, and offer suggestions for further exploration.

[0126] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying geological anomalies in tunnels, characterized in that, Includes the following steps: S1. Map multimodal geological data to a unified vector space, extract entities, relationships, and entity attributes, and construct a multimodal knowledge graph; S2. Obtain the current engineering geological background and exploration data, and perform matching calculations with the multimodal knowledge graph to obtain the correlation value between the current data and the multimodal knowledge graph; S3. Filter out the N knowledge graph nodes and their relationships that are most relevant to the current data, and transform them into structured text descriptions. Together with the current detection data and the tunnel geological anomaly identification problem to be solved, they form the structured prompts for this query. Input them into the large language model to obtain the answer to the tunnel geological anomaly identification problem to be solved. The method for calculating the correlation between the current data and the multimodal knowledge graph is as follows: in, This represents the correlation score between the current data and the multimodal knowledge graph. and express , and Weighting coefficients Indicates the similarity of entity attributes. Indicates the connectivity of relational paths. Indicates the matching degree of historical cases; The formula for calculating entity attribute similarity is as follows: in, express and cosine similarity, A vector representing the current probe data; Represents the entity attribute vector in a knowledge graph; Select entity attributes from the knowledge graph and match them with the current probe data vector. The entity with the largest value is taken as the associated entity of the current probe data vector. The formula for calculating the connectivity of the relationship path is as follows: in, pw represents the number of hops in the shortest path from the associated entity of the current probe data to the target entity in the knowledge graph, and pw represents the weight product of all relation edges on the shortest path. The formula for calculating the matching degree of historical cases is as follows: in, and These are weighting coefficients. This represents the feature vector of historical cases in a knowledge graph. This represents the current engineering geological background vector. This represents the geological background vector of historical cases in the knowledge graph.

2. The method for identifying tunnel geological anomalies according to claim 1, characterized in that, Entity types include: detection methods, geological anomalies, and decision rules; the attributes of detection methods include: detection method name, effective detection distance, and detection parameters; the attributes of geological anomalies include: location, size, water abundance, and degree of fragmentation; the attributes of decision rules include: rule name, preconditions, execution actions, rule description, and associated flowchart steps; detection methods include TSP, TEM, GPR, and borehole.

3. The method for identifying tunnel geological anomalies according to claim 2, characterized in that, The decision rules include the detection method rules, which are as follows: First, execute the TSP test. If the TSP test result is normal, proceed with construction as normal; if it is abnormal, execute the TEM test. If the TEM detection results are normal, proceed with construction as normal; if abnormal, execute GPR. If the GPR detection results are normal, proceed with construction as normal; if abnormal, proceed with drilling. Geological anomalies were identified based on borehole imaging, and advanced support was implemented.

4. The method for identifying tunnel geological anomalies according to claim 1, characterized in that, , and The dynamic weighting coefficients are calculated as follows: in, , and yes , and The interaction score with query q, where query vector q represents the features of the current probe data. , , and For learnable parameters, , and These are entity vectors, relation vectors, and historical case context vectors obtained from the current engineering geological background and exploration data based on the knowledge graph. , and They are , and Attention weights , and They represent and In the initial dynamic weights at the current time step, λ is the fusion coefficient. , and They represent and Historical values ​​from the previous time step.

5. The method for identifying tunnel geological anomalies according to claim 4, characterized in that, and The historical value update method is as follows: in, , and They represent and The historical values ​​of the first two time steps, Indicates the learning rate. , and They represent , and Feedback signals.

6. The method for identifying tunnel geological anomalies according to claim 3, characterized in that, , , and These are aligned feature vectors extracted from ground-penetrating radar data, transient electromagnetic method data, and borehole descriptions, respectively. This indicates vector concatenation.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method as described in any one of claims 1-6.

8. An electronic device, characterized in that, The device includes a processor and a memory, the processor being interconnected with the memory, wherein the memory is used to store a computer program, the computer program including computer-readable instructions, and the processor is configured to invoke the computer-readable instructions to perform the method as described in any one of claims 1-6.

9. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-6.

Images