Methods, devices and programs for analyzing the geological structure of shield tunnels

Abstract

This application discloses a method, apparatus, and program product for analyzing the geological structure of shield tunnels, belonging to the field of shield tunnel technology. The method includes: acquiring a tag database and a test database. The tag database includes multiple tag datasets with geological type tags and containing different fusion characteristics of soil columns from exploration holes. The test database includes at least one test dataset without geological type tags and containing different fusion characteristics of a target segment ring. For each test dataset, a large language model is used for type recognition processing, and the tag database is used for type matching processing. Candidate geological types are obtained based on the type recognition and matching results. The candidate geological types of each test dataset are integrated to obtain the geological type of the target segment ring, and geological structure analysis is performed accordingly to obtain the geological structure analysis information of the target segment ring. This application can effectively improve the accuracy of geological structure analysis of shield tunnels.

Description

Technical Field

[0001] This application relates to the field of shield tunnel technology, and in particular to a method, apparatus and program product for analyzing the geological structure of shield tunnels. Background Technology

[0002] Earth pressure balance (EPB) shield tunneling has become a primary construction method for urban underground tunnels due to its high efficiency and safety. Tunnels constructed using EPB shield tunneling are called shield tunnels. Shield tunnels typically exhibit different mechanical responses under complex geological conditions. The uncertainty of geological conditions poses a significant challenge to the construction process, directly impacting the safety and efficiency of shield tunnel construction. Clearly defining the geological structure of the shield tunnel can effectively reduce construction risks and ensure the smooth progress of the project.

[0003] To address this, related technologies utilize pre-trained deep learning models to identify the stratigraphic types of unknown areas within shield tunnels and analyze the geological structure of these areas based on the identified stratigraphic types. The deep learning models are trained using multiple training samples and their corresponding stratigraphic type labels. However, these technologies rely excessively on a fixed mapping between training samples and their corresponding stratigraphic type labels. This makes them prone to misjudgments when facing complex geological scenarios (such as mixed strata and transition zones), resulting in low accuracy in analyzing the geological structure of shield tunnels. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, and program product for analyzing the geological structure of shield tunnels, which aims to effectively improve the accuracy of geological structure analysis of shield tunnels.

[0005] To achieve the above objectives, one aspect of this application proposes a method for analyzing the geological structure of a shield tunnel, the method comprising:

[0006] Obtain a label database and a test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without the geological type labels and containing different fusion characteristics of target segment rings;

[0007] The large language model is used to perform type recognition processing on each of the test datasets, and the label database is used to perform type matching processing on each of the test datasets. Based on the type recognition results and type matching results of each of the test datasets, the candidate geological types of each of the test datasets are obtained.

[0008] The candidate geological types of each of the datasets to be tested are integrated to obtain the geological type of the target segment ring;

[0009] Geological structure analysis is performed based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring.

[0010] To achieve the above objectives, another aspect of this application proposes a geological structure analysis device for shield tunnels, the device comprising:

[0011] The acquisition module is used to acquire a label database and a test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without the geological type labels and containing different fusion characteristics of target segment rings;

[0012] The first processing module is used to perform type recognition processing on each of the test datasets using a large language model, perform type matching processing on each of the test datasets using the label database, and obtain candidate geological types for each of the test datasets based on the type recognition results and type matching results of each of the test datasets.

[0013] The second processing module is used to integrate the candidate geological types of each of the datasets to be tested to obtain the geological type of the target segment ring.

[0014] The third processing module is used to perform geological structure analysis processing based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring.

[0015] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for analyzing the geological structure of a shield tunnel.

[0016] According to the shield tunnel geological structure analysis method, apparatus, and program product provided in this application, a label database and a test database are first obtained. The label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes. The test database includes at least one test dataset without geological type labels and containing different fusion characteristics of the target segment ring. Then, a large language model is used to perform type recognition processing on each test dataset, and the label database is used to perform type matching processing on each test dataset. Based on the type recognition results and type matching results of each test dataset, candidate geological types of each test dataset are obtained. Then, the candidate geological types of each test dataset are integrated to obtain the geological type of the target segment ring. Finally, geological structure analysis processing is performed based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring. According to the technical solution of this application embodiment, a label database and a test database are set up. The former includes multiple label datasets with geological type labels and different fusion characteristics of soil columns from exploration holes, while the latter includes at least one test dataset without geological type labels and different fusion characteristics of the target tunnel segment ring. A dual geological type identification mechanism of abstract reasoning and concrete matching is introduced for each test dataset to identify candidate geological types, thus effectively improving the accuracy of candidate geological types. Subsequently, the final geological type of the target tunnel segment ring is determined by integrating the actual geological type of the target tunnel segment ring under the multi-source dataset, thus effectively reducing the local errors of some test datasets and improving the identification accuracy of the geological type of the target tunnel segment ring. Finally, geological structure analysis is performed based on the geological type of the target tunnel segment ring to obtain the geological structure analysis information of the target tunnel segment ring, thus effectively improving the analysis accuracy of the geological structure of the shield tunnel, thereby providing reliable geological structure analysis information for the construction process of the shield tunnel and helping to improve the construction efficiency and safety of the shield tunnel. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for analyzing the geological structure of a shield tunnel provided in this application;

[0018] Figure 2 This is an example diagram showing the position of the accelerometer sensor provided in this application;

[0019] Figure 3 yes Figure 1 Flowchart of step S102;

[0020] Figure 4 yes Figure 1 Flowchart of step S103;

[0021] Figure 5 yes Figure 4 Flowchart of step S501;

[0022] Figure 6 yes Figure 1 Flowchart of step S104;

[0023] Figure 7 This is a diagram illustrating the specific implementation process of a shield tunnel geological structure analysis method provided in this application;

[0024] Figure 8 This is an example diagram of the geological structure of the segment rings provided in this application;

[0025] Figure 9 This is a structural diagram of a shield tunnel geological structure analysis device provided in this application. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0027] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0028] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0030] To address the shortcomings of related technologies, this application provides a method, apparatus, and program product for analyzing the geological structure of shield tunnels, aiming to effectively improve the accuracy of geological structure analysis of shield tunnels.

[0031] First, the implementation steps of a shield tunnel geological structure analysis method provided in this application embodiment will be described in detail below with reference to the accompanying drawings.

[0032] This application provides a method for analyzing the geological structure of a shield tunnel, which can be applied to a terminal, a server, or software running on either a terminal or a server. The terminal can be a tablet, laptop, desktop computer, etc., but is not limited to these. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. Furthermore, the server can be a node server in a blockchain network, but is not limited to these. Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0033] Reference Figure 1 , Figure 1 This is a flowchart of a method for analyzing the geological structure of a shield tunnel provided in this application. The method for analyzing the geological structure of a shield tunnel may include, but is not limited to, the following steps S101-S104.

[0034] S101, Obtain the label database and the test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without geological type labels and containing different fusion characteristics of target segment rings.

[0035] It should be noted that a shield tunnel has multiple segment rings and multiple exploration boreholes with soil columns. The geological type of the exploration boreholes and soil columns is known, and any segment ring with an unknown geological type is called the target segment ring. Both the exploration boreholes and the target segment rings have multiple attribute data, such as shield parameters, vibration acceleration signals, and images of excavated soil. Fusion features refer to features obtained by fusing at least two attribute data.

[0036] In this step, a label database and a test database are acquired to facilitate the analysis and processing of the geological structure of the target tunnel segment ring in subsequent steps. The label database includes multiple label datasets, each with a geological type label and containing different fusion characteristics of the soil columns from the exploration wells. The test database includes at least one test dataset, none of which has a geological type label, and each dataset contains different fusion characteristics of the target tunnel segment ring.

[0037] S102, use a large language model to perform type recognition processing on each dataset to be tested, use a label database to perform type matching processing on each dataset to be tested, and obtain candidate geological types for each dataset to be tested based on the type recognition results and type matching results.

[0038] It should be noted that the type identification result of the dataset to be tested refers to the geological type obtained by the large language model through type identification processing of the dataset to be tested, while the type matching result of the dataset to be tested refers to the geological type obtained by the label database through type matching processing of the dataset to be tested.

[0039] In this step, for each dataset to be tested, a large language model is called to perform type recognition processing to obtain the type recognition result of the dataset to be tested. At the same time, a label database is used to perform type matching processing to obtain the type matching result of the dataset to be tested. Then, based on the type recognition result and the type matching result of the dataset to be tested, a candidate geological type of the dataset to be tested is generated. This candidate geological type reflects the actual geological type of the target segment ring under a single dataset to be tested.

[0040] Optionally, the type of large language model can be set according to the actual situation, and this application embodiment does not limit it. For example, the large language model can be DeepSeek, ChatGPT, Claude, Gemini, etc., but is not limited to this.

[0041] S103, integrate the candidate geological types of each dataset to be tested to obtain the geological type of the target segment ring.

[0042] In this step, considering that the geological type of the target segment ring may be the same or different under different test datasets, in order to reduce the local error of a single test dataset, after obtaining the candidate geological types of each test dataset, the candidate geological types of each test dataset are integrated into the geological type of the target segment ring. This geological type reflects the actual geological type of the target segment ring under all test datasets, that is, the final geological type of the target segment ring.

[0043] S104. Based on the geological type of the target segment ring, perform geological structure analysis to obtain the geological structure analysis information of the target segment ring.

[0044] In this step, the geological type of the target segment ring is used as the data basis for geological structure analysis. Geological structure analysis processing is performed to obtain the geological structure analysis information of the target segment ring, thereby realizing the analysis and processing of the geological structure of the target segment ring.

[0045] As can be seen, the embodiments of this application set up a label database and a test database. The former includes multiple label datasets with geological type labels and different fusion characteristics of soil columns from exploration holes, while the latter includes at least one test dataset without geological type labels and different fusion characteristics of target segment rings. A dual geological type identification mechanism of abstract reasoning and concrete matching is introduced for each test dataset to identify candidate geological types of each test dataset.

[0046] Specifically, in this mechanism, on the one hand, the large language model has cross-modal abstract reasoning capabilities. By calling the large language model to perform deep semantic analysis on the fusion features in the test dataset and identifying the geological type of the test dataset accordingly, it can reduce the dependence on traditional fixed mappings and reduce the risk of misjudgment in complex geological scenarios. On the other hand, the label database has a large number of labeled fusion features. By matching the fusion features in the test dataset with these labeled fusion features to identify the geological type of the test dataset, it can provide a highly deterministic and concrete reference for the identification of the geological type of the test dataset. Considering that the large language model may output semantically related but actually mismatched geological types due to the ambiguity of geological features, and that database matching may return low similarity results due to the lack of similar samples, generating candidate geological types for the test dataset based on the type identification results and type matching results can promote the formation of a complementary relationship between the abstract reasoning process and the concrete matching process in the identification of geological types, reduce the risk of misjudgment caused by each other, and thus effectively improve the identification accuracy of the geological type of the test dataset, providing an accurate data foundation for subsequent geological result analysis and processing.

[0047] After identifying candidate geological types for each test dataset, considering that individual test datasets may have local errors due to factors such as sensor noise leading to misclassification of candidate geological types, this embodiment integrates the candidate geological types of each test dataset into the geological type of the target tunnel segment ring. By integrating the actual geological type of the target tunnel segment ring from multiple source datasets, the final geological type of the target tunnel segment ring can be determined, effectively reducing local errors in some test datasets and improving the accuracy of geological type identification. Subsequently, this embodiment performs geological structure analysis based on the geological type of the target tunnel segment ring to obtain geological structure analysis information, thereby enabling the determination of the geological structure of the target tunnel segment ring. In summary, this embodiment can effectively improve the analysis accuracy of the geological structure of shield tunnels, providing reliable geological structure analysis information for the construction process of shield tunnels, and contributing to improved construction efficiency and safety.

[0048] The steps described above will be explained in further detail below.

[0049] In some implementations, the step S101 above, which involves obtaining the tag database and the database to be tested, may include the following steps S201-S206.

[0050] S201, acquire shield tunneling parameters, vibration acceleration signals and spoil images of soil columns in each exploration hole.

[0051] It should be noted that shield parameters refer to the parameters of the tunnel boring machine (TBM) during operation, which may include, but are not limited to, shield thrust (F), cutterhead torque (T), cutterhead rotation speed (CRS), advance speed (AR), upper soil chamber pressure (USP), lower soil chamber pressure (LSP), grouting pressure (GP), and grouting volume (GV). Vibration acceleration signal refers to the vibration acceleration signal generated when the cutterhead at the front of the TBM rotates and cuts the ground. This may include, but is not limited to, triaxial acceleration signals and uniaxial acceleration signals along the tunnel's excavation axis (X-axis). Excavated soil image refers to the image of the excavated soil transmitted to the excavated soil conveyor belt.

[0052] In this step, the shield tunneling parameters, vibration acceleration signals, and spoil images of the soil columns in each exploration hole are obtained in order to construct a tag database.

[0053] Optionally, the methods for acquiring shield parameters, vibration acceleration signals, and slag images can be set according to actual conditions, and this embodiment does not limit this.

[0054] For example, the shield parameters of the soil column of each exploration hole are collected through a shield parameter acquisition system. The shield parameter acquisition system refers to the sensor system that comes with the shield machine. It is usually installed by the shield machine manufacturer and does not require additional configuration.

[0055] As another example, a vibration signal acquisition system is used to collect the vibration acceleration signals of the soil columns in each exploration hole. This system may include, but is not limited to, acceleration sensors, signal acquisition cards, and acquisition terminals. Figure 2 As shown, the acceleration sensors may include, but are not limited to, one triaxial accelerometer and three unilateral accelerometers. The triaxial accelerometer is installed in the soil chamber wall between the upper part of the main drive motor and the lower part of the personnel compartment inside the tunnel boring machine. The first unilateral accelerometer (i.e., X1) is installed in the soil chamber wall between the main drive motor at an upper right 45° angle and the hydraulic cylinder. The second unilateral accelerometer (i.e., X2) is installed in the soil chamber wall between the main drive motor at a lower right 45° angle and the hydraulic cylinder. The third unilateral accelerometer (i.e., X3) is installed in the soil chamber wall between the main drive motor at a lower left 45° angle and the hydraulic cylinder. The triaxial accelerometer can collect acceleration signals in three directions: along the tunnel boring axis (X-axis), perpendicular to the tunnel boring axis (Y-axis), and horizontally along the tunnel boring axis (Z-axis). The unilateral accelerometers can collect acceleration signals along the tunnel boring axis (X-axis). The accelerometer sensor can be connected to the signal acquisition card via a data cable such as a Bayonet Nut Connector (BNC) RF interface. The signal acquisition card can be connected to the acquisition terminal located in the tunnel boring machine control room via a data cable such as a Universal Serial Bus (USB). There is no specific limitation on the acquisition frequency of the acceleration signal; for example, the acquisition frequency can be 500Hz, but it is not limited to this.

[0056] For example, a soil image acquisition system is used to acquire soil images of the soil columns in each exploration hole. This system consists of a camera tripod and a high-speed image acquisition camera. The camera tripod is welded to any position on the side of the soil conveyor belt. The height of the camera tripod is adjusted to be 30 to 50 cm higher than the soil conveyor belt. The high-speed image acquisition camera is then fixed to the camera tripod, and its acquisition mode is adjusted to autofocus acquisition mode, thereby automatically acquiring image data of the soil on the soil conveyor belt. There is no specific limitation on the acquisition frequency of the high-speed image acquisition camera; for example, the acquisition frequency could be 1 frame / min, but it is not limited to this.

[0057] S202, feature extraction processing is performed on the shield parameters, vibration acceleration signals and slag images of the soil columns of each exploration hole to obtain the shield parameter features, vibration features and geological image features of the soil columns of each exploration hole.

[0058] In this step, for each exploration borehole soil column, the shield tunneling parameters of the exploration borehole soil column are feature-extracted to obtain the shield tunneling parameter features of the exploration borehole soil column; the vibration acceleration signal of the exploration borehole soil column is feature-extracted to obtain the vibration features of the exploration borehole soil column; and the slag and soil image of the exploration borehole soil column is feature-extracted to obtain the geological image features of the exploration borehole soil column. In this way, key features highly correlated with geological types can be captured from three dimensions: equipment operation, structural response, and stratum characteristics, thus providing a reliable data foundation for subsequent type identification and type matching processing.

[0059] Optionally, the shield tunneling parameter features, vibration features, and geological image features are regularized by rewriting the shield tunneling parameter features into the form of {parameter name: value (level)}, and the vibration features and geological image features into the form of {feature name: value / feature description}, thereby obtaining shield tunneling parameter features, vibration features, and geological image features in JSON format, so that various features match the input format of the subsequent large language model.

[0060] Optionally, the feature extraction processing method can be set according to the actual situation, and this embodiment does not limit it.

[0061] For example, interquartile range (IQR) statistical analysis is performed on the shield tunneling parameters of the soil column in the exploration borehole. Based on the specific values ​​of various parameters, three levels (high, medium, and low) are assigned, thus obtaining the levels of various parameters in the shield tunneling parameters of the soil column as the shield tunneling parameter characteristics of the soil column. The root mean square value, peak value, kurtosis, dominant frequency, and peak time of the vibration acceleration signal of the soil column in the exploration borehole are calculated as the vibration characteristics of the soil column. Different image recognition models are used to perform image recognition and analysis on the slag and soil images of the soil column in the exploration borehole, obtaining features such as the soil-rock ratio, soil-rock color, rock texture, and rock strength as the geological image features of the soil column in the exploration borehole. There are no restrictions on the number and type of image recognition models. For example, there can be four image recognition models, corresponding to the four features of soil-rock ratio, color, rock texture, and rock strength. The type of image recognition model can be Convolutional Neural Network (CNN), YOLOv3, Faster R-CNN, etc., but is not limited to these.

[0062] S203, the shield tunneling parameter characteristics, vibration characteristics and geological image characteristics of each exploration hole soil column are fused to obtain the first fused characteristics, second fused characteristics and third fused characteristics of each exploration hole soil column.

[0063] It should be noted that the first and second fusion features are obtained by fusing any two of the shield tunneling parameter features, vibration features, and geological image features, while the third fusion feature is obtained by fusing the shield tunneling parameter features, vibration features, and geological image features.

[0064] In this step, for each exploration hole soil column, any two of the shield tunneling parameter features, vibration features, and geological image features are fused to obtain the first fused feature and the second fused feature. This achieves feature fusion and interaction between any two of the equipment operation dimension, structural response dimension, and stratum characteristic dimension. At the same time, the shield tunneling parameter features, vibration features, and geological image features of the exploration hole soil column are fused into the third fused feature. This achieves feature fusion and interaction between the equipment operation dimension, structural response dimension, and stratum characteristic dimension, thus effectively improving feature diversity.

[0065] Optionally, the feature objects selected during the fusion of the first fusion feature and the second fusion feature can be set according to the actual situation, and this embodiment does not limit this.

[0066] For example, the first fusion feature is obtained by fusing shield tunneling parameter features and vibration features, while the second fusion feature is obtained by fusing shield tunneling parameter features and geological image features. Here, the essence of shield tunneling parameter features is the real-time output of the interaction between stratum mechanical properties and shield machine operation, which is a dynamic feedback of stratum type to the shield tunnel construction process. The essence of vibration features and geological image features is the real-time output of stratum mechanical properties themselves, which is a static feedback of stratum type to the shield tunnel construction process. Accordingly, during feature fusion and interaction, shield tunneling parameter features are selected as the common feature object for the fusion of the first and second fusion features, and vibration features and geological image features are selected as the non-common feature objects for the fusion of the first and second fusion features, respectively. In this way, dynamic features are used to compensate for the missing contextual information of static features, which can accurately uncover more hidden features related to stratum type, thereby effectively improving the accuracy of feature fusion and interaction.

[0067] Optionally, the feature fusion processing method can be set according to the actual situation, and this embodiment does not limit it.

[0068] For example, shield tunneling parameter features and vibration features are sequentially stitched together to form the first fusion feature, shield tunneling parameter features and geological image features are sequentially stitched together to form the second fusion feature, and shield tunneling parameter features, vibration features and geological image features are sequentially stitched together to form the third fusion feature. In this way, feature fusion and interaction of different dimensions can be quickly achieved through stitching and fusion.

[0069] For example, using shield tunneling parameter features as the query and vibration features as the key and value, a self-attention mechanism is used to fuse the shield tunneling parameter features and vibration features into a first fused feature; using shield tunneling parameter features as the query and geological image features as the key and value, a self-attention mechanism is used to fuse the shield tunneling parameter features and geological image features into a second fused feature; using the first fused feature as the query and geological image features as the key and value, a self-attention mechanism is used to fuse the first fused feature and geological image features into a third fused feature. In this way, the self-attention mechanism can effectively improve the effect of feature fusion and interaction in different dimensions, and achieve more accurate feature fusion.

[0070] S204. Based on the first fusion characteristics and geological type labels of the soil columns in each exploration well, construct the first label dataset.

[0071] In this step, for each soil column from the exploration well, the geological type obtained from the exploration report is assigned a value, resulting in a geological type label for the soil column. This geological type label is then added to the end of the first fusion feature of the soil column, ensuring that the first fusion feature carries the corresponding label information. By traversing each soil column, the first fusion feature and corresponding geological type label for each soil column can be obtained, thus forming the first label dataset.

[0072] Optionally, the method for assigning geological types can be set according to the actual situation, and this implementation method does not limit this. For example, the geological type includes any one of soft soil strata, hard rock strata, mixed soft and hard strata, or boulder strata. The type is assigned a value according to the degree of softness or hardness, that is, soft soil strata are assigned a value of 1, boulder strata are assigned a value of 2, mixed soft and hard strata are assigned a value of 3, and hard rock strata are assigned a value of 4, but it is not limited to this.

[0073] S205. Based on the second fusion characteristics and geological type labels of the soil columns in each exploration well, a second label dataset is constructed.

[0074] It should be understood that the description of constructing the second-label dataset here can follow the description of constructing the first-label dataset in step A04 above. The only difference is that the data objects here are the second fusion feature and the geological type label, and the resulting dataset is the second-label dataset. This will not be elaborated further.

[0075] S206. Based on the third fusion characteristics and geological type labels of the soil columns in each exploration well, a third-label dataset is constructed.

[0076] It should be understood that the description of constructing the third-label dataset here can follow the description of constructing the first-label dataset in step A04 above. The only difference is that the data objects here are the third fusion feature and geological type label, and the resulting dataset is the third-label dataset. This will not be elaborated further.

[0077] Optionally, after forming the first-label dataset, the second-label dataset, and the third-label dataset, text segmentation methods are used to segment these three datasets, and text embedding models are used to transform the data in these three datasets into vector form and store them. The text segmentation methods and text embedding models are not limited; for example, text segmentation methods may include, but are not limited to, fixed-size segmentation, semantic segmentation, document structure-based slicing, recursive segmentation, sentence window retrieval, automatic merging retrieval, etc., and text embedding models may include, but are not limited to, Transformer-based bidirectional encoders (i.e., BERT), Sentence-BERT, FastText, OpenAIAda, etc.

[0078] In some implementations, the step S101 above, which involves obtaining the tag database and the database to be tested, may include the following steps S301-S306.

[0079] S301, acquire shield parameters, vibration acceleration signals and slag images of the target segment ring;

[0080] S302, feature extraction processing is performed on the shield parameters, vibration acceleration signals and slag images of the target segment ring to obtain the shield parameter features, vibration features and geological image features of the target segment ring;

[0081] S303, the shield tunneling parameter features, vibration features and geological image features of the target segment ring are fused to obtain the first fused feature, the second fused feature and the third fused feature of the target segment ring; wherein, the first fused feature and the second fused feature are each obtained by fusing any two of the shield tunneling parameter features, vibration features and geological image features, and the third fused feature is obtained by fusing the shield tunneling parameter features, vibration features and geological image features;

[0082] S304, Construct the first test dataset based on the first fusion feature of the target segment ring;

[0083] S305, construct the second test dataset based on the second fusion feature of the target segment ring;

[0084] S306. Based on the third fusion feature of the target segment ring, construct the third test dataset.

[0085] It should be noted that the description of obtaining the test database here can follow the description of obtaining the label database in the above embodiments. The only difference is that the constructed test dataset is the first test dataset, the second test dataset, and the third test dataset, and geological type labels are not required when constructing the test dataset. This will not be elaborated further.

[0086] In some implementations, refer to Figure 3 In step S102 above, a large language model is used to perform type recognition processing on each dataset to be tested, and a label database is used to perform type matching processing on each dataset to be tested. Based on the type recognition results and type matching results of each dataset to be tested, candidate geological types of each dataset to be tested are obtained, including the following steps S401-S403.

[0087] S401, input the target dataset and the first prompt word into the large language model to obtain the type recognition result of the target dataset.

[0088] It should be noted that the target dataset refers to any dataset to be tested, and the first prompt word is used to guide the large language model to identify the geological type of the target dataset based on the fusion features of each labeled dataset and the mapping relationship between the geological type labels.

[0089] In this step, a large language model is pre-trained. During training, each labeled dataset and a pre-set second prompt word are input into the large language model, allowing it to gradually learn and store the mapping relationship between the fusion features of each labeled dataset and the geological type labels. Accordingly, during type recognition processing, the target dataset and the first prompt word are input into the trained large language model. The large language model will then identify the geological type of the target dataset based on its learned mapping relationship, thus achieving type recognition processing. Here, the large language model possesses cross-modal abstract reasoning capabilities. By calling the large language model to perform deep semantic analysis of the fusion features in the target dataset and identifying the geological type accordingly, it can reduce the reliance on traditional fixed mappings and lower the risk of misjudgment in complex geological scenarios.

[0090] Optionally, the first prompt word can be set according to the actual situation, and this embodiment does not limit it. For example, the first prompt word could be "Based on the stored mapping relationship, determine the geological type corresponding to the data contained in the currently input dataset," etc., but it is not limited to this. In addition, the second prompt word can also be set according to the actual situation, and this embodiment does not limit it. For example, the second prompt word could be "Analyze the relationship between the fusion features and geological type labels in the currently input dataset, remember the connection between them, and store it," etc.

[0091] S402, use the text matching degree calculation method to calculate the similarity between each label dataset and the target dataset, and determine the geological type label of the label dataset with the highest similarity as the type matching result of the target dataset.

[0092] In this step, for each labeled dataset, the text matching degree calculation method is used to calculate the similarity between the fusion features in the labeled dataset and the fusion features in the target dataset, i.e., the similarity of the labeled dataset relative to the target dataset. By traversing multiple labeled datasets, the similarity between multiple labeled datasets and the target dataset can be obtained. Subsequently, the labeled dataset with the highest similarity is selected from the multiple labeled datasets, and the geological type label of the selected labeled dataset is determined as the type matching result for the target dataset. Here, each labeled dataset has a large number of labeled fusion features. By calculating the similarity between the fusion features in the target dataset and the labeled fusion features in each labeled dataset, and selecting the geological type label of the labeled dataset with the highest similarity as the type matching result for the target dataset, the geological type corresponding to the target dataset can be quickly and accurately determined, thus providing a highly deterministic and concrete reference for the identification of the geological type of the dataset to be tested.

[0093] Optionally, the text matching degree calculation method can be set according to the actual situation, and this embodiment does not limit it. For example, the text matching degree calculation method may include, but is not limited to, cosine similarity method, term frequency-inverse document frequency method, semantic similarity method, latent semantic analysis method, etc.

[0094] Optionally, the method for determining similarity can be set according to the actual situation, and this embodiment does not limit it.

[0095] For example, only one text matching calculation method is selected to calculate the similarity between the label dataset and the target dataset. For instance, the cosine similarity between the label dataset and the target dataset can be calculated, which can quickly determine the similarity between the label dataset and the target dataset, thereby effectively improving the determination efficiency of type matching processing.

[0096] For example, at least two text matching calculation methods are selected to calculate the initial similarity between the label dataset and the target dataset, and the calculated initial similarities are averaged to obtain the final similarity. For instance, the cosine similarity between the label dataset and the target dataset is calculated, while the term frequency-inverse document frequency (TF-IF) method is used to calculate the similarity between the label dataset and the target dataset. Then, the average of the cosine similarity between the label dataset and the target dataset and the similarity calculated by the TF-IF method is determined as the similarity between the label dataset and the target dataset. Here, cosine similarity captures the semantic similarity between the label dataset and the target dataset, while the TF-IF method captures the term frequency similarity between the label dataset and the target dataset. By integrating the similarities from different dimensions into the final similarity, the accuracy of type matching processing can be effectively improved.

[0097] It is worth noting that label datasets with the same type of fusion features can be selected to calculate similarity. For example, if the target dataset contains the first fusion feature, then a label dataset containing the first fusion feature can be selected to calculate its similarity with the target dataset. Of course, label datasets with different types of fusion features can also be selected to calculate similarity; this implementation method does not limit this.

[0098] S403, when the type identification result and type matching result of the target dataset are different, update the first prompt word and text matching degree calculation method, and return to the step of inputting the target dataset and the first prompt word into the large language model until the type identification result and type matching result of the target dataset are the same, and determine the type identification result of the target dataset as the candidate geological type of the target dataset.

[0099] In this step, after obtaining the type identification result and type matching result of the target dataset, it is determined whether the type identification result and type matching result are the same. If they are the same, it means that the geological types identified from the two perspectives are consistent and the obtained geological types are accurate. At this time, the type identification result of the target dataset is determined as the candidate geological type of the target dataset. If not, it means that the geological types identified from the two perspectives are inconsistent and the geological types obtained from the two perspectives are questionable. At this time, it is necessary to redesign the first prompt word and change the text matching degree calculation method. Then, return to the step of inputting the target dataset and the first prompt word into the large language model, i.e., step S401, to achieve iterative processing.

[0100] Here, considering that large language models may output semantically related but actually mismatched geological types due to fuzzy geological features, and that database matching may return low similarity results due to the lack of similar samples, the process returns to the corresponding step to achieve iterative processing when the type identification result and type matching result of the test dataset are inconsistent, until the type identification result and type matching result of the test dataset are consistent. The type identification result of the target dataset is then determined as the candidate geological type of the target dataset. This enables the abstract reasoning process and the concrete matching process to form a complementary and mutually verifying relationship in the identification of geological types, reducing the risk of misjudgment caused by each other, thereby effectively improving the identification accuracy of geological types in the test dataset and providing an accurate data foundation for subsequent geological result analysis and processing.

[0101] Optionally, the method for updating the first prompt word can be set according to the actual situation, and this embodiment does not limit it.

[0102] For example, a new prompt word template is randomly selected from a pre-defined first database as the new first prompt word. The first database pre-stores several prompt word templates, which are manually designed and have the same properties as the first prompt word. Thus, by traversing the first database, the first prompt word can be updated quickly, improving the update efficiency and consequently enhancing the efficiency of type recognition processing.

[0103] For example, both the type identification result and the type matching result indicate the corresponding geological type, which is a specific numerical value. The difference between the type identification result and the type matching result is calculated as the first type deviation information. The first type deviation information indicates the degree of difference between the type identification result and the type matching result. Then, any prompt word template corresponding to the first type deviation information is selected from a preset second database, and the selected prompt word template is used as the new first prompt word. The second database pre-stores several preset first type deviation information sets and multiple prompt word templates corresponding to each first type deviation information set. These prompt word templates are artificially designed and have the same properties as the first prompt words. Here, the first type deviation information, which characterizes the degree of difference between the type identification result and the type matching result, is used as the traversal benchmark to traverse the second database, searching for the prompt word template corresponding to the traversal benchmark as the new first prompt word. This makes the first prompt word more closely match the actual geological type, effectively improving the update accuracy of the first prompt word, and thus helping to improve the accuracy of type identification processing.

[0104] Optionally, the update method of the text matching degree calculation method can be set according to the actual situation, and this embodiment does not limit it.

[0105] For example, one or more new text matching degree calculation methods are randomly selected as the new text matching degree calculation methods. In this way, the text matching degree calculation methods can be updated quickly, improving the update efficiency of the text matching degree calculation methods, and thus effectively improving the efficiency of type matching processing.

[0106] For example, both the type identification result and the type matching result indicate the corresponding geological type, which is a specific numerical value. The absolute value of the difference between the type identification result and the type matching result is calculated as the second type deviation information. The second type deviation information indicates the degree of difference between the type identification result and the type matching result. It is determined whether the second type deviation information is less than a preset deviation threshold. If so, it means that the difference between the type identification result and the type matching result is not significant, and the type matching result has a certain degree of accuracy. In this case, there is no need to correct the type matching result; it is only necessary to change the text matching degree calculation method. Accordingly, a text matching degree calculation method corresponding to the second type deviation information is selected from a preset third database as the new text matching degree calculation method. The third database is preset... The database stores several preset second-type deviation information and multiple text matching degree calculation methods corresponding to each second-type deviation information. The number of methods selected from the third database is the same as the number of current methods. If not, it indicates that the difference between the type recognition result and the type matching result is large, and the type matching result may have a high error. In this case, not only should the text matching degree calculation method be changed, but the accuracy of type matching processing should also be further improved. Accordingly, the text matching degree calculation method corresponding to the second-type deviation information is selected from the third database as the new text matching degree calculation method. The number of methods selected from the third database must be greater than the number of current methods. For example, the number of methods selected from the third database is equal to the sum of the number of current methods and one. Here, the second type deviation information, which characterizes the degree of difference between the type recognition result and the type matching result, is used as the traversal benchmark to traverse the third database. The text matching degree calculation method corresponding to the traversal benchmark is then selected as the new text matching degree calculation method. When the difference between the type recognition result and the type matching result is large, the number of text matching degree calculation methods is increased. When the difference between the type recognition result and the type matching result is small, the current number of text matching degree calculation methods is maintained. This can make the text matching degree calculation method more consistent with the actual geological type, improve the accuracy of the text matching degree calculation method, and thus help improve the accuracy of type matching processing.

[0107] In some implementations, refer to Figure 4 In step S103 above, the candidate geological types of each dataset to be tested are integrated to obtain the geological type of the target segment ring, including any one of the following steps S501-S502:

[0108] S501, If ​​there are at least two datasets to be tested, the candidate geological types of the multiple datasets to be tested are integrated based on the number of times each type of candidate geological type appears in the multiple datasets to be tested, and the geological types are obtained.

[0109] S502 If there is one and only one dataset to be tested, then the candidate geological type is determined as the geological type.

[0110] In this embodiment, in practical applications, one or more datasets can be selected to analyze the geological structure of the target segment ring. If only one dataset is selected, the candidate geological type of the selected dataset is directly determined as the geological type of the target segment ring. If two or more datasets are selected, the frequency of occurrence of each candidate geological type in the multiple datasets is first determined, and then the frequency of occurrence of each candidate geological type is used as a benchmark to integrate the candidate geological types of these datasets into the geological type of the target segment ring. This can further reduce the local errors of some datasets and thus more accurately determine the actual geological type of the target segment ring.

[0111] In some implementations, refer to Figure 5 In step S501 above, the candidate geological types of multiple datasets are integrated based on the frequency of occurrence of each candidate geological type in multiple datasets to obtain the geological type. This may include any one of the following steps A01-A02:

[0112] A01. If there is only one candidate geological type with the highest frequency of occurrence, then the candidate geological type with the highest frequency of occurrence shall be determined as the geological type.

[0113] A02. If there are at least two candidate geological types that appear most frequently, then obtain the number of iterations for each dataset in the type identification and type matching processes, and determine the candidate geological type of the dataset with the fewest iterations as the geological type.

[0114] In this embodiment, if there is only one candidate geological type with the highest frequency of occurrence, then the candidate geological type with the highest frequency of occurrence is directly determined as the geological type of the target segment ring. If there are at least two candidate geological types with the highest frequency of occurrence, then the number of iterations for each test dataset in the type identification and type matching processes is obtained. Here, the number of iterations refers to the number of times steps S401-S403 above are executed, and the candidate geological type of the test dataset with the fewest iterations is selected as the geological type of the target segment ring. It should be noted that the case where there are at least two test datasets with the fewest iterations is not considered here.

[0115] Here, when there is only one candidate geological type with the highest frequency of occurrence, this type can fully reflect the actual geological type of the target segment ring. Determining the geological type of the target segment ring through a majority vote effectively reduces the impact of local errors caused by biases in some datasets, thereby significantly improving the accuracy of geological type identification. When there are at least two candidate geological types with the highest frequency of occurrence, it indicates that multiple candidate geological types can fully reflect the actual geological type of the target segment ring, and the majority vote approach is no longer applicable. To address this, this implementation introduces the number of iterations in the type identification and type matching processes of the test dataset. This number of iterations indirectly reflects the data quality of the candidate geological types: fewer iterations indicate a faster convergence speed in determining the candidate geological types (i.e., no need for multiple adjustments), higher consistency between type matching and type identification processes, and more reliable candidate geological types. Conversely, more iterations indicate a slower convergence speed (i.e., requiring multiple adjustments), lower consistency between type matching and type identification processes, and less reliable candidate geological types. Based on this, the candidate geological type with the fewest iterations is selected, representing the most reliable result, and is used as the geological type of the target segment ring. This allows for the selection of the highest-quality candidate geological type from multiple most frequent candidate geological types as the actual geological type of the target segment ring, effectively reducing the impact of local errors caused by biases in some test datasets and thus significantly improving the accuracy of geological type identification for the target segment ring.

[0116] In some implementations, refer to Figure 6 In step S104 above, geological structure analysis is performed based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring. This may include any one of the following steps S601-S602:

[0117] S601, If ​​the geological type is not a mixed soft and hard strata type, then the geological type will be determined as geological structure analysis information;

[0118] S602, if the geological type is a mixed soft and hard strata type, then the target segment ring is processed by rock-soil ratio calculation to obtain the rock-soil ratio value of the target segment ring, and the rock-soil ratio value and geological type are determined as geological structure analysis information; wherein, the rock-soil ratio value is the ratio between hard rock strata and soft soil strata in the mixed soft and hard strata.

[0119] It should be noted that the rock-soil ratio value refers to the ratio between hard rock strata and soft soil strata in the target segment ring belonging to the mixed hard and soft strata type. It can accurately reflect the distribution of soft soil strata and hard rock strata in the cross section.

[0120] In this embodiment, considering that the impact of mixed soft and hard strata on the construction process of shield tunnels is much greater than that of other types of strata, for target segment rings that do not belong to the mixed soft and hard strata type, the geological type of the target segment ring is directly used as the geological structure analysis information of the target segment ring and output. For target segment rings that belong to the mixed soft and hard strata type, further analysis and processing are performed, that is, the soil-rock ratio of the target segment ring is calculated to obtain the soil-rock ratio value of the target segment ring. The soil-rock ratio value and the geological type of the target segment ring are used as the geological structure analysis information of the target segment ring and output. In this way, reliable geological structure analysis information can be provided for the construction process of shield tunnels (especially in mixed soft and hard strata), which helps to improve the construction efficiency and safety of shield tunnels.

[0121] In some embodiments, step S602 above, which involves performing a soil-rock ratio calculation on the target segment ring to obtain the soil-rock ratio value of the target segment ring, may include the following steps B01-B04:

[0122] B01, obtain the average value of soft soil type characteristics and the average value of hard rock type characteristics, and obtain the soil and rock type characteristic value of the target segment ring; where the average value of soft soil type characteristics refers to the average value of soil and rock type characteristic values ​​of soft soil strata in the tunnel excavation area, and the average value of hard rock type characteristics refers to the average value of soil and rock type characteristic values ​​of hard rock strata in the tunnel excavation area.

[0123] B02, obtain the difference between the soil and rock type characteristic value of the target segment ring and the average value of the soft soil type characteristic value as the first type difference;

[0124] B03, obtain the difference between the average value of soft soil type characteristics and the average value of hard rock type characteristics as the second type difference;

[0125] B04, obtain the ratio of the first type of difference to the second type of difference as the soil-rock ratio value.

[0126] In related technologies, due to the significant difficulty in analyzing strata containing mixed soft soil and hard rock, geological structure analysis is often performed only on the soft soil and hard rock strata of the tunnel segment rings, rarely on the mixed soft soil and hard rock strata. To address this, this embodiment performs a soil-rock ratio calculation on target tunnel segment rings belonging to the mixed soft-hard strata type. The aim is to capture the ratio between hard rock and soft soil strata within these target tunnel segment rings, and based on this, to perform geological structure analysis on the target tunnel segment rings belonging to the mixed soft-hard strata type.

[0127] In practical implementation, the excavated area of ​​the tunnel refers to the area in the shield tunnel where all or part of the tunnel excavation has been completed, and the geological type of the segment rings located within the excavated area is known. Based on this, firstly, the sum of the soil and rock type characteristic values ​​of all segment rings belonging to the soft soil stratum type is obtained, and the ratio of this sum to the number of segment rings belonging to the soft soil stratum type is taken as the average value of the soft soil type characteristic. This average value of the soft soil type characteristic can characterize the actual geological condition of the soft soil stratum in the excavated area of ​​the tunnel, as shown in the following formula (1):

[0128] (1);

[0129] In equation (1), This represents the average value of soft soil type characteristics; This indicates the number of segment rings belonging to the soft soil stratum type; Indicates the first The characteristic values ​​of the soil and rock type of a segment ring belonging to the soft soil stratum type.

[0130] Similarly, the sum of the soil and rock type characteristic values ​​of all segment rings belonging to hard rock strata is obtained, and the ratio of this sum to the number of segment rings belonging to hard rock strata is taken as the average value of hard rock type characteristics. This average value of hard rock type characteristics can characterize the actual geological conditions of hard rock strata in the tunnel excavation area, as shown in the following formula (2):

[0131] (2);

[0132] In equation (2), This represents the average value of hard rock type characteristics; This indicates the number of segment rings belonging to hard rock strata types; Indicates the first The characteristic values ​​of soil and rock type of a segment ring belonging to the hard rock strata type.

[0133] Simultaneously, while obtaining the average values ​​of soft soil type characteristics and hard rock type characteristics, the soil and rock type characteristic values ​​of the target segment ring belonging to the mixed soft and hard strata type are also obtained.

[0134] Furthermore, the characteristic values ​​of the soil and rock types are calculated from the shield tunneling parameters. The specific process is as follows: First, the product of the cutterhead torque value and the cutterhead rotation speed value is obtained as the cutterhead operating characteristic. This operating characteristic is directly related to the cutting resistance of the strata. The cutting resistance is positively correlated with the strata strength; that is, the higher the strata strength, the greater the cutting resistance, and vice versa. Therefore, the cutterhead operating characteristic can be indirectly correlated with the strata strength, thus reflecting the actual geological conditions. Simultaneously, the product of the propulsion speed value and the shield thrust value is obtained as the propulsion operating characteristic. This propulsion operating characteristic is related to the propulsion resistance of the strata. The propulsion resistance is positively correlated with the strata strength; that is, the higher the strata strength, the greater the propulsion resistance, and vice versa. Therefore, the propulsion operating characteristic can be indirectly correlated with the strata strength, thus reflecting the actual geological conditions. Then, the ratio of the cutterhead operation characteristics to the propulsion operation characteristics is obtained as the soil and rock type characteristic value. When the cutting resistance and propulsion resistance increase simultaneously, the soil and rock type characteristic value increases accordingly, which indicates that the formation strength is higher. Conversely, when the cutting resistance and propulsion resistance decrease simultaneously, the soil and rock type characteristic value decreases accordingly, which indicates that the formation strength is lower. This allows the soil and rock type characteristic value to directly quantify the formation strength and thus reflect the actual geological conditions. For ease of understanding, the above soil and rock type characteristic value can be expressed as the following formula (3):

[0135] (3);

[0136] In equation (3), Indicates the characteristic values ​​of soil and rock types; This represents the shield tunneling thrust value, and its unit is _____. ; This indicates the torque value of the cutter head, and its unit is... ; This represents the propulsion speed value, and its unit is _____. ; This indicates the cutter head rotation speed, and its unit is _____. Here, by dimensionlessly fusing the shield thrust, cutterhead torque, propulsion speed, and cutterhead rotation speed of the segment ring, cutterhead operation characteristics and propulsion operation characteristics are obtained. These are then further transformed into soil and rock type characteristic values ​​that can reflect the actual geological conditions. This reduces the negative impact of noise characteristics unrelated to the actual geological conditions, especially the negative impact of the ambiguity of shield parameters on identifying stratum types. As a result, the actual geological conditions of the segment ring can be accurately located, providing a reliable data benchmark for the subsequent calculation of soil and rock ratio values.

[0137] Then, the difference between the characteristic values ​​of the soil and rock type of the target segment ring and the average characteristic value of the soft soil type is calculated, and this difference is used as the first type difference. The first type difference reflects the degree of difference between the actual geological conditions of the target segment ring, which belongs to the mixed soft and hard strata type, and the actual geological conditions of the soft soil strata. At the same time, the difference between the average characteristic value of the soft soil type and the average characteristic value of the hard rock type is calculated, and this difference is used as the second type difference. The second type difference reflects the degree of difference between the actual geological conditions of the soft soil strata and the actual geological conditions of the hard rock strata.

[0138] Finally, the ratio of the first type of difference to the second type of difference is taken as the soil-rock ratio of the target segment ring. The range of the soil-rock ratio value is as follows: When the soil-rock ratio is 0, it indicates that the entire formation is soft soil; when the soil-rock ratio is 1, it indicates that the entire formation is hard rock. This means that the higher the soil-rock ratio, the fewer the soft soil layers and the more the hard rock layers in the target segment ring; conversely, the lower the soil-rock ratio, the more the soft soil layers and the fewer the hard rock layers in the target segment ring. For ease of understanding, the above soil-rock ratio can be expressed as the following formula (4):

[0139] (4);

[0140] In equation (4), Indicates the soil-rock ratio value; This represents the characteristic value of the soil and rock type of the target segment ring.

[0141] Therefore, this embodiment provides a method for calculating the ratio between hard rock strata and soft soil strata in mixed hard-soft strata. This method determines the actual ratio between hard rock strata and soft soil strata in mixed hard-soft strata by comparing the actual geological conditions of the target segment ring belonging to the mixed hard-soft strata type with the actual geological conditions of the soft soil strata, and the actual geological conditions of the soft soil strata with the actual geological conditions of the hard rock strata. In this way, the actual distribution of soft soil strata and hard soil strata in the cross section of the target segment ring belonging to the mixed hard-soft strata type can be comprehensively captured. In subsequent steps, the geological structure analysis of the mixed hard-soft strata will be carried out based on this ratio value, which can further improve the accuracy of the geological structure analysis of mixed hard-soft strata.

[0142] To facilitate understanding of the shield tunnel geological structure analysis method described in this application, a practical application scenario of the method is illustrated below. The shield tunnel project in this example is located in the central area of ​​a southern city. Before construction, borehole sampling was conducted in the tunnel's crossing area, constructing multiple exploration boreholes with soil columns. The depth of each exploration borehole soil column exceeded the tunnel's burial depth, thus covering the entire depth of the tunnel. The shield tunnel employs a dual-line earth pressure balance shield construction method and traverses completely weathered, strongly weathered, and moderately weathered granite strata. Some of the completely weathered granite strata contain isolated boulders. To ensure the smooth progress of the shield tunnel construction, this example uses the aforementioned shield tunnel geological structure analysis method to analyze the geological structure of the 1563rd ring segment in the shield tunnel, thereby providing a basis for construction decisions, such as... Figure 7 As shown, the specific principle is as follows:

[0143] Step 1, Data Collection:

[0144] The shield tunneling parameters for all exploration borehole soil columns and the 1563rd segment ring were collected using a shield parameter acquisition system. These parameters included shield thrust, cutterhead torque, cutterhead rotation speed, advance speed, upper and lower pressure of the soil chamber, grouting pressure, and grouting volume. Vibration acceleration signals for all exploration borehole soil columns and the 1563rd segment ring were collected using a vibration signal acquisition system. These signals included triaxial acceleration signals and uniaxial acceleration signals along the shield tunnel's excavation axis. Excavated soil images for all exploration borehole soil columns and the 1563rd segment ring were collected using a spoil image acquisition system. Relevant data for the exploration borehole soil columns can be obtained from the shield tunnel's exploration report.

[0145] The second step is feature extraction:

[0146] The ChatGPT-4o model was used to analyze the shield parameters, vibration acceleration signals, and slag images of all exploration borehole soil columns and the 1563rd ring segment, obtaining the shield parameter characteristics, vibration characteristics, and geological image characteristics of all exploration borehole soil columns and the 1563rd ring segment.

[0147] Taking the feature extraction operation of the 1563rd ring segment as an example, the specific rules for interquartile range statistical analysis in actual implementation are as follows: 1) Shield thrust value ≤ 23000 The marking is low, and the shield thrust value is ≥53000. The marking is high, and the shield thrust value is between 23,000 and 53,000. The markings between are: 1) Middle; 2) Cutter head torque value ≤ 3000 The marking is low, and the cutter head torque value is ≥8000. The marking is high, and the cutter head torque value is between 3000 and 8000. The mark between them is medium; 3) The propulsion speed value ≤ 10 is marked as low, and the propulsion speed value ≥ 40 is marked as high, and the propulsion speed value is between 10 and 40 The mark between them is medium; 4) The cutter head rotation speed value ≤ 1.5 is marked as low, and the cutter head rotation speed value ≥ 1.8 is marked as high, and the cutter head rotation speed value is between 1.5 and 1.8 The mark between them is medium; The soil pressure value at the top of the soil bin ≤ 115 is marked as low, and the soil pressure value at the top of the soil bin ≥ 360 is marked as high, and the soil pressure value at the top of the soil bin is between 115 and 360 The mark between them is medium; 6) The soil pressure value at the bottom of the soil bin ≤ 150 is marked as low, and the soil pressure value at the bottom of the soil bin ≥ 400<H is marked as high, and the soil pressure value at the bottom of the soil bin is between 150 and 400 The mark between them is medium; 7) The grouting pressure value ≤ 0.24 is marked as low, and the grouting pressure value ≥​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​ Its ≤0.24 Therefore, it is marked as low; the grouting volume is 10. ≤12 Therefore, it is marked as low.

[0149] The vibration acceleration signal of the 1563rd ring segment was analyzed, and the corresponding root mean square (RMS) value, peak value, kurtosis, dominant frequency, and peak time were obtained as the vibration characteristics of the 1563rd ring segment. The RMS value indicates that the intensity of the vibration acceleration signal within the period is low, i.e., below a preset intensity threshold. The peak value indicates that the vibration acceleration signal exhibits discontinuous multiple peaks within a single period. The kurtosis value is significantly greater than the preset kurtosis threshold, and the dominant frequency is 60 Hz. The peak time provides feedback on the edge position of the vibration acceleration signal within a single cycle.

[0150] The slag and soil images of the 1563rd ring segment were analyzed to obtain the corresponding characteristics such as the soil-rock ratio, soil-rock color, rock texture, and rock strength as geological image features of the 1563rd ring segment. Among them, the soil-rock ratio is 60-70% soft soil, the soil-rock color is a mixture of gray and yellowish-brown, the rock texture shows that the rock structure is clearly visible, and the rock strength is medium.

[0151] After obtaining the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of the 1563rd segment ring, the shield tunneling parameter characteristics of the 1563rd segment ring are rewritten in the form of {parameter name: value (level)}, and the vibration characteristics and geological image characteristics of the 1563rd segment ring are rewritten in the form of {feature name: value / feature description}, thus obtaining the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of the 1563rd segment ring in JSON format.

[0152] It should be understood that the feature extraction operation for all soil columns in the exploration holes is the same as the feature extraction operation for the 1563rd ring segment, and will not be elaborated further.

[0153] The third step is feature fusion:

[0154] For the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of the 1563rd ring segment, the following steps are taken: The shield tunneling parameter characteristics and vibration characteristics of the 1563rd ring segment are sequentially spliced ​​together to form the first fusion characteristic of the 1563rd ring segment; the shield tunneling parameter characteristics and geological image characteristics of the 1563rd ring segment are sequentially spliced ​​together to form the second fusion characteristic of the 1563rd ring segment; the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of the 1563rd ring segment are sequentially spliced ​​together to form the third fusion characteristic of the 1563rd ring segment; the first fusion characteristic of the 1563rd ring segment constitutes the first dataset to be tested; the second fusion characteristic of the 1563rd ring segment constitutes the second dataset to be tested; and the third fusion characteristic of the 1563rd ring segment constitutes the third dataset to be tested. This forms the dataset to be tested.

[0155] For the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of each exploration borehole soil column, the following steps are taken: The geological type of the exploration borehole soil column obtained from the exploration report is assigned a value, resulting in a geological type label for the exploration borehole soil column. The geological type includes any one of soft soil strata, hard rock strata, mixed soft-hard strata, or boulder strata. The soft soil strata type is assigned a value of 1, the boulder strata type a value of 2, the mixed soft-hard strata type a value of 3, and the hard rock strata type a value of 4. Simultaneously, the shield tunneling parameter characteristics and vibration characteristics of the exploration borehole soil column are sequentially concatenated to form the first fusion feature; the shield tunneling parameter characteristics and geological image characteristics of the exploration borehole soil column are sequentially concatenated to form the second fusion feature; and the shield tunneling parameter characteristics, vibration characteristics, and geological image characteristics of the exploration borehole soil column are sequentially concatenated to form the third fusion feature. The geological type label of the exploration borehole soil column is then added to the end of the first, second, and third fusion features, respectively.

[0156] By traversing the soil columns of each exploration borehole, the first fusion feature and corresponding geological type label of each soil column can be obtained, thus forming the first label dataset; by traversing the soil columns of each exploration borehole, the second fusion feature and corresponding geological type label of each soil column can be obtained, thus forming the second label dataset; by traversing the soil columns of each exploration borehole, the third fusion feature and corresponding geological type label of each soil column can be obtained, thus forming the third label dataset; thus, a label database can be formed.

[0157] After obtaining the two databases, the text data in each database is segmented using a semantic segmentation text segmentation method. The segmented text data is then converted into vector form and stored using a Transformer-based bidirectional encoder text embedding model.

[0158] The fourth step is to define any dataset to be tested as the target dataset and perform candidate geological type identification processing:

[0159] Each labeled dataset and a pre-set second prompt are input into the ChatGPT-4o model. This allows the ChatGPT-4o model to progressively learn and store the mapping relationship between the fusion features of each labeled dataset and the geological type labels, resulting in a trained ChatGPT-4o model. The second prompt is "Analyze the relationship between the fusion features and geological type labels in the currently input dataset, remember the connection, and store it." After training the ChatGPT-4o model, the target dataset and the first prompt are input into the trained model. The ChatGPT-4o model will then identify the geological type of the target dataset based on the learned mapping relationship, using this as the type identification result. Simultaneously, for each labeled dataset, the cosine similarity between the fusion features in the labeled dataset and the fusion features in the target dataset is calculated, representing the similarity of the labeled dataset relative to the target dataset. This similarity is obtained by traversing multiple labeled datasets. Finally, the labeled dataset with the highest similarity is selected from these datasets, and the geological type label of the selected labeled dataset is determined as the type matching result for the target dataset. Subsequently, it is determined whether the type recognition result and the type matching result are the same. If they are the same, the type recognition result of the target dataset is determined as the candidate geological type of the target dataset. Otherwise, the first prompt word needs to be redesigned and the text matching degree calculation method needs to be changed. Then, it returns to the step of inputting the target dataset and the first prompt word into the large language model to achieve iterative processing.

[0160] Candidate geological types for each dataset can be obtained by traversing each dataset.

[0161] Step 5: Integration and processing of geological types:

[0162] If there is only one selected test dataset, the candidate geological type of the selected test dataset is directly determined as the geological type of the target segment ring. If there are two or more selected test datasets, the occurrence frequency of each candidate geological type in the multiple test datasets is obtained. If there is only one candidate geological type with the highest occurrence frequency, it is determined as the geological type. If there are at least two candidate geological types with the highest occurrence frequency, the number of iterations in the type identification and type matching processes for each test dataset is obtained, and the candidate geological type of the test dataset with the fewest iterations is determined as the geological type. Here, the number of iterations refers to the number of times the type identification and type matching processes are executed.

[0163] Through the above processing, the geological type of the 1563rd ring segment was identified as a mixed soft and hard stratum.

[0164] Step 6, Geological Structure Analysis:

[0165] Since the geological type of the 1563rd ring segment is a mixed soft and hard stratum, the soil-rock ratio of the 1563rd ring segment is calculated. The calculation process is shown in formulas (1)-(4) above. The soil-rock characteristic value of the 1563rd ring segment is calculated to be 13.75, the average value of the hard rock characteristic is 30, and the average value of the soft soil characteristic is 10. Therefore, the soil-rock ratio of the 1563rd ring segment is 18.75%. Figure 8 As shown, the soil-rock ratio and stratigraphic type of the 1563rd segment ring are determined as the geological structure analysis information for the 1563rd segment ring and output.

[0166] In addition, refer to Figure 9 This application also provides a shield tunnel geological structure analysis device, which can implement the above-mentioned shield tunnel geological structure analysis method. The device includes:

[0167] The acquisition module 701 is used to acquire a label database and a test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without geological type labels and containing different fusion characteristics of target segment rings.

[0168] The first processing module 702 is used to perform type recognition processing on each dataset to be tested using a large language model, perform type matching processing on each dataset to be tested using a label database, and obtain candidate geological types for each dataset to be tested based on the type recognition results and type matching results of each dataset to be tested.

[0169] The second processing module 703 is used to integrate the candidate geological types of each dataset to be tested to obtain the geological type of the target segment ring.

[0170] The third processing module 704 is used to perform geological structure analysis processing based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring.

[0171] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0172] Finally, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method for analyzing the geological structure of a shield tunnel.

[0173] The content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0174] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0175] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for analyzing the geological structure of a shield tunnel, characterized in that, The method includes: Obtain a label database and a test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without the geological type labels and containing different fusion characteristics of target segment rings; The large language model is used to perform type recognition processing on each of the test datasets, and the label database is used to perform type matching processing on each of the test datasets. Based on the type recognition results and type matching results of each of the test datasets, the candidate geological types of each of the test datasets are obtained. The candidate geological types of each of the datasets to be tested are integrated to obtain the geological type of the target segment ring; Geological structure analysis is performed based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring; The step of performing geological structure analysis based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring includes: If the geological type is not a mixed soft and hard strata type, then the geological type is determined as the geological structure analysis information; Alternatively, if the geological type is the mixed hard and soft strata type, then the target segment ring is processed by rock-soil ratio calculation to obtain the rock-soil ratio value of the target segment ring, and the rock-soil ratio value and the geological type are determined as the geological structure analysis information; wherein, the rock-soil ratio value is the ratio between hard rock strata and soft soil strata in the mixed hard and soft strata; The step of performing a soil-rock ratio calculation on the target segment ring to obtain the soil-rock ratio value of the target segment ring includes: The average values ​​of soft soil type characteristics and hard rock type characteristics are obtained, and the soil and rock type characteristic values ​​of the target segment ring are also obtained; wherein, the average value of soft soil type characteristics is the average value of soil and rock type characteristic values ​​of soft soil strata in the tunnel excavation area, and the average value of hard rock type characteristics is the average value of soil and rock type characteristic values ​​of hard rock strata in the tunnel excavation area; The difference between the soil and rock type characteristic value of the target segment ring and the average value of the soft soil type characteristic value is obtained as the first type difference; The difference between the average value of the soft soil type characteristics and the average value of the hard rock type characteristics is obtained as the second type difference; The ratio of the first type of difference to the second type of difference is obtained as the soil-rock ratio value.

2. The method according to claim 1, characterized in that, The acquisition of the tag database and the database to be tested includes: Obtain the shield tunneling parameters, vibration acceleration signals, and spoil images of the soil columns in each of the exploration holes; Feature extraction processing is performed on the shield tunneling parameters, vibration acceleration signals and slag images of the soil columns of each exploration hole to obtain the shield tunneling parameter features, vibration features and geological image features of the soil columns of each exploration hole; The shield tunneling parameters, vibration characteristics, and geological image characteristics of each exploration hole soil column are subjected to feature fusion processing to obtain a first fused feature, a second fused feature, and a third fused feature for each exploration hole soil column; wherein, the first fused feature and the second fused feature are each obtained by fusing any two of the shield tunneling parameters, the vibration characteristics, and the geological image characteristics, and the third fused feature is obtained by fusing the shield tunneling parameters, the vibration characteristics, and the geological image characteristics; Based on the first fusion features and geological type labels of the soil columns from each exploration borehole, a first label dataset is constructed; Based on the second fusion features and geological type labels of the soil columns from each exploration borehole, a second label dataset is constructed; A third-label dataset is constructed based on the third fusion features and geological type labels of the soil columns from each exploration borehole.

3. The method according to claim 1, characterized in that, The acquisition of the tag database and the database to be tested includes: The shield parameters, vibration acceleration signals, and slag images of the target segment ring are obtained. Feature extraction processing is performed on the shield parameters, vibration acceleration signals and slag images of the target segment ring to obtain the shield parameter features, vibration features and geological image features of the target segment ring; The shield tunneling parameter features, vibration features, and geological image features of the target segment ring are subjected to feature fusion processing to obtain a first fused feature, a second fused feature, and a third fused feature of the target segment ring; wherein, the first fused feature and the second fused feature are each obtained by fusing any two of the shield tunneling parameter features, the vibration features, and the geological image features, and the third fused feature is obtained by fusing the shield tunneling parameter features, the vibration features, and the geological image features; Based on the first fusion feature of the target pipe segment ring, a first test dataset is constructed; Based on the second fusion feature of the target pipe segment ring, a second test dataset is constructed; A third test dataset is constructed based on the third fusion feature of the target pipe ring.

4. The method according to claim 1, characterized in that, The process involves using a large language model to perform type identification processing on each of the test datasets, and using the label database to perform type matching processing on each of the test datasets. Based on the type identification results and type matching results of each of the test datasets, candidate geological types for each of the test datasets are obtained, including: The target dataset and the first prompt word are input into the large language model to obtain the type identification result of the target dataset; wherein, the target dataset represents any one of the test datasets, and the first prompt word is used to guide the large language model to identify the geological type of the target dataset based on the mapping relationship between the fusion features of each of the labeled datasets and the geological type label; The similarity between each of the labeled datasets and the target dataset is calculated using a text matching degree calculation method, and the geological type label of the labeled dataset with the highest similarity is determined as the type matching result of the target dataset; When the type identification result and type matching result of the target dataset are different, the first prompt word and the text matching degree calculation method are updated, and the process returns to the step of inputting the target dataset and the first prompt word into the large language model until the type identification result and type matching result of the target dataset are the same, and the type identification result of the target dataset is determined as the candidate geological type of the target dataset.

5. The method according to claim 1, characterized in that, The process of integrating the candidate geological types of each of the datasets to be tested to obtain the geological type of the target segment ring includes: If there are at least two datasets to be tested, the candidate geological types of the multiple datasets to be tested are integrated based on the number of times each type of candidate geological type appears in the multiple datasets to be tested, and the geological type is obtained. Alternatively, if there is one and only one dataset to be tested, then the candidate geological type is determined as the geological type.

6. The method according to claim 5, characterized in that, The step of integrating the candidate geological types from multiple datasets based on the frequency of occurrence of each type of candidate geological type in multiple datasets to obtain the geological type includes: If there is only one candidate geological type with the highest frequency of occurrence, then the candidate geological type with the highest frequency of occurrence shall be determined as the geological type. Alternatively, if there are at least two candidate geological types that appear most frequently, then the number of iterations of each dataset to be tested during the type identification process and the type matching process is obtained, and the candidate geological type of the dataset to be tested with the fewest iterations is determined as the geological type.

7. A geological structure analysis device for shield tunnels, characterized in that, The device includes: The acquisition module is used to acquire a label database and a test database; wherein, the label database includes multiple label datasets with geological type labels and containing different fusion characteristics of soil columns from exploration holes, and the test database includes at least one test dataset without the geological type labels and containing different fusion characteristics of target segment rings; The first processing module is used to perform type recognition processing on each of the test datasets using a large language model, perform type matching processing on each of the test datasets using the label database, and obtain candidate geological types for each of the test datasets based on the type recognition results and type matching results of each of the test datasets. The second processing module is used to integrate the candidate geological types of each of the datasets to be tested to obtain the geological type of the target segment ring. The third processing module is used to perform geological structure analysis processing based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring; The step of performing geological structure analysis based on the geological type of the target segment ring to obtain the geological structure analysis information of the target segment ring includes: If the geological type is not a mixed soft and hard strata type, then the geological type is determined as the geological structure analysis information; Alternatively, if the geological type is the mixed hard and soft strata type, then the target segment ring is processed by rock-soil ratio calculation to obtain the rock-soil ratio value of the target segment ring, and the rock-soil ratio value and the geological type are determined as the geological structure analysis information; wherein, the rock-soil ratio value is the ratio between hard rock strata and soft soil strata in the mixed hard and soft strata; The step of performing a soil-rock ratio calculation on the target segment ring to obtain the soil-rock ratio value of the target segment ring includes: The average values ​​of soft soil type characteristics and hard rock type characteristics are obtained, and the soil and rock type characteristic values ​​of the target segment ring are also obtained; wherein, the average value of soft soil type characteristics is the average value of soil and rock type characteristic values ​​of soft soil strata in the tunnel excavation area, and the average value of hard rock type characteristics is the average value of soil and rock type characteristic values ​​of hard rock strata in the tunnel excavation area; The difference between the soil and rock type characteristic value of the target segment ring and the average value of the soft soil type characteristic value is obtained as the first type difference; The difference between the average value of the soft soil type characteristics and the average value of the hard rock type characteristics is obtained as the second type difference; The ratio of the first type of difference to the second type of difference is obtained as the soil-rock ratio value.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for analyzing the geological structure of a shield tunnel as described in any one of claims 1 to 6.

Images