Large model agent system and method for shield tunnel structure state analysis

Abstract

The invention provides a large model agent system and method for shield tunnel structure state analysis, and the large model agent system employs a layered decoupling system architecture. The system sequentially comprises a natural language interaction layer, a large model agent cognition layer, a task planning and tool scheduling layer, a structure state understanding and reasoning layer and a shield tunnel structure state spatio-temporal data and knowledge supporting layer from top to bottom. By means of the layered architecture design, efficient, systematic and intelligent analysis of the shield tunnel structure state can be achieved under the conditions that the number of components is huge and structure indexes are complex, and the structure state cognition level and the decision support capacity of the shield tunnel in the operation and maintenance stage are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of tunnel and underground engineering structure health monitoring technology, specifically involving a shield tunnel structure status analysis and intelligent decision support system that integrates three-dimensional laser scanning data, spatiotemporal database and artificial intelligence large model, and in particular a shield tunnel structure status comprehensive analysis system and method based on large model agent. Background Technology

[0002] With the rapid development of urban rail transit and municipal underground space, shield tunneling has become one of the main forms of underground engineering construction. Shield tunnels are usually assembled from a large number of precast concrete segments. During the operation period, their structural condition is affected by a variety of factors such as stratum deformation, groundwater changes, disturbance from nearby construction, train loads, and long-term material deterioration, which can easily lead to structural defects such as convergence deformation, segment deflection, joint misalignment, and corners.

[0003] To understand the service status of shield tunnel structures, high-precision detection technologies such as 3D laser scanning have been introduced in engineering practice to periodically scan the inner surface of the tunnel, thereby obtaining 3D geometric deformation data detailed down to the segment and joint levels. By comparing the results of multiple scans, a data set of structural deformation with distinct temporal and spatial characteristics can be formed. However, with the increase in the detection cycle and the extension of tunnel mileage, the accumulated data volume grows exponentially, and the data dimensions are complex with diverse index types. Traditional methods relying on manual or simple statistical analysis are no longer sufficient to meet the actual needs of engineering. Existing shield tunnel structural condition analysis technologies mainly have the following shortcomings: (1) Low efficiency in data management and utilization: Most projects only analyze data from a single period or a few key sections, lacking a unified spatiotemporal database system, making it difficult to systematically explore the temporal evolution patterns and spatial distribution characteristics of deformation data.

[0004] (2) Fragmented analysis methods: Different structural indicators (such as convergence, deflection, and misalignment) are usually analyzed independently by different algorithms or empirical formulas, lacking a comprehensive understanding of the relationship between multiple indicators and multiple components.

[0005] (3) High professional threshold and decision-making depends on human experience: structural status judgment and operation and maintenance decision-making rely heavily on a few experienced engineering experts. Newly hired technical personnel have difficulty quickly understanding a large amount of complex data, resulting in low analysis efficiency and strong subjectivity.

[0006] (4) Existing intelligent methods lack engineering understanding: Although some studies have attempted to introduce machine learning or deep learning methods for structural state recognition, most models only remain at the level of numerical prediction and are difficult to understand engineering semantics, specification requirements and actual engineering scenarios, and cannot directly provide interpretable suggestions for engineering decisions.

[0007] In recent years, large language models have demonstrated powerful capabilities in natural language understanding, complex reasoning, and cross-domain knowledge fusion, providing new technical pathways for complex data analysis and intelligent decision-making in the engineering field. However, how to deeply integrate large language models with spatiotemporal data on the structural state of shield tunnels to construct an intelligent agent system capable of understanding the tunnel's structural state still requires further research and engineering implementation. Summary of the Invention

[0008] The purpose of this invention is to provide a large model proxy system and method for analyzing the structural state of shield tunnels. By constructing a spatiotemporal database of the structural state of shield tunnels and combining it with a large model proxy that has professional knowledge and understanding capabilities, the invention enables automatic analysis of a large amount of segment and joint deformation data, identification of temporal and spatial correlations, and generation of structural state assessment and decision recommendations. This solves the problems of low analysis efficiency, high professional threshold, and insufficient comprehensive judgment capabilities in the existing technology.

[0009] The specific technical solution of the present invention is as follows: A large-scale proxy system for structural state analysis of shield tunnels, employing a layered and decoupled system architecture, includes: (1) Spatiotemporal data and knowledge support layer of shield tunnel structure status As the underlying infrastructure of the system, it is used to uniformly support the multi-period spatiotemporal data resources and engineering professional knowledge resources required for the structural state analysis of shield tunnels.

[0010] (2) Structural state understanding and reasoning layer It is used for engineering semantic understanding, state discrimination and reasoning analysis of spatiotemporal data of shield tunnel structural status, and is the core functional layer for realizing intelligent analysis and evaluation of shield tunnel structural status.

[0011] (3) Task planning and tool scheduling layer It is used to address the needs of structural condition analysis and evaluation of shield tunnels, and to perform engineering decomposition of analysis tasks, execution path planning, and automatic scheduling of analysis tools.

[0012] (4) Large model agent cognitive layer It is used to build intelligent agents with semantic understanding capabilities for shield tunnel engineering, structural state analysis and cognitive capabilities, and autonomous task planning capabilities. It is the core cognitive hub connecting user analysis needs and system analysis capabilities.

[0013] (5) Natural Language Interaction Layer It is used to enable engineers and the system to perform structured state analysis interaction and express evaluation results based on natural language.

[0014] Each layer interacts with the others through standardized data and semantic interfaces, and works collaboratively under the unified scheduling of the large model agent system, thereby enabling multi-source data understanding of the shield tunnel structure status, spatiotemporal correlation analysis, engineering semantic assessment, and decision support.

[0015] Furthermore, the spatiotemporal data and knowledge support layer for the shield tunnel structure status includes a shield tunnel structure status spatiotemporal database submodule and a shield tunnel structure status engineering knowledge base submodule.

[0016] The shield tunnel structural state spatiotemporal database submodule takes the shield tunnel lining structure as the object, performs refined modeling of the tunnel structure according to the spatial hierarchy of lining rings, segments and joints, assigns a unique number to each lining ring, segment and its circumferential and longitudinal joints, and uses it as the spatial index unit for structural state data.

[0017] The shield tunnel structure state spatiotemporal database submodule is used to store shield tunnel structure deformation data acquired at different times based on three-dimensional laser scanning technology. The shield tunnel structure deformation data includes at least whole-ring convergence deformation data, segment deflection deformation data, circumferential joint misalignment angle data, and longitudinal joint misalignment angle data.

[0018] The shield tunnel structural deformation data includes various types of data that are associated with and stored with corresponding time and spatial number information, thereby constructing a spatiotemporal data system for the shield tunnel structural state, supporting multi-condition queries and calls based on time, space, and structural index dimensions.

[0019] The shield tunnel engineering knowledge base submodule is used to store engineering knowledge data related to the structural state analysis of shield tunnels, including but not limited to shield tunnel design specifications, structural verification standards, operation and maintenance technical procedures, typical disease cases, and expert experience texts, which are used to provide professional knowledge constraints for understanding and reasoning about the structural state during the large model analysis process.

[0020] Furthermore, the structural state understanding and reasoning layer is built based on a large language model and uses retrieval-enhanced generation technology to jointly input the relevant knowledge content in the shield tunnel structural state engineering knowledge base submodule with the shield tunnel structural deformation data to be analyzed, so that the large language model has both data-driven capability and engineering knowledge constraint capability during the analysis process.

[0021] Furthermore, the structural state understanding and reasoning layer is capable of comprehensive analysis and reasoning about the following: 1) The evolution trend of structural deformation data at different time scales; 2) Spatial distribution characteristics and relationships among different lining rings, segments, and joints; 3) The synergistic change characteristics among multiple structural deformation indices; 4) The matching relationship between structural deformation state and engineering specifications and empirical thresholds; In this way, the large language model can not only describe numerical changes, but also make comprehensive judgments on the structural state of shield tunnels with engineering semantics.

[0022] Furthermore, the task planning and tool scheduling layer includes: 1) Task parsing unit, used to receive the analysis target from the large model agent cognition layer and parse it into a structural state analysis task; 2) Task decomposition unit, used to decompose a comprehensive analysis task into multiple sub-tasks, wherein the sub-tasks include at least data screening task, indicator extraction task, time series analysis task and spatial correlation analysis task; 3) Execution order planning unit, used to determine the execution order of subtasks based on the logical dependencies between them; 4) Data and tool scheduling unit, used to retrieve relevant data from the shield tunnel structure state spatiotemporal database according to task requirements, and transmit the processing results to the structure state understanding and reasoning layer; By setting up this layer, the system can automatically complete the process of analyzing the structural status of complex shield tunnels without human intervention.

[0023] Furthermore, the large model agent cognitive layer is built based on a large language model, and its functions include: 1) Perform semantic understanding on the user's natural language input and identify the implicit intention to analyze the structural state of the shield tunnel; 2) Mapping engineering problems expressed in natural language into structural state analysis objectives, including structural safety assessment, deformation trend analysis, abnormal component identification, and risk assessment; 3) Generate corresponding analysis strategies based on the analysis objectives, and pass the analysis strategies to the task planning and tool scheduling layer; 4) Through prompt word constraints and context control, the large model agent always participates in the analysis as a shield tunnel structural engineering and operation and maintenance expert, ensuring that the analysis results conform to engineering logic and professional common sense.

[0024] Furthermore, the natural language interaction layer enables users to obtain professional analysis conclusions on the structural status of shield tunnels through natural language interaction, without having to directly operate complex data systems or analysis software.

[0025] Furthermore, this large model agent system also includes a system coordination and constraint mechanism that runs through all the above layers. This mechanism is used to impose engineering rationality constraints on the analysis process and output results, ensuring that the system is used only for structural state analysis and decision support, and does not directly generate construction or control instructions.

[0026] A method for analyzing the structural state of shield tunnels, employing the large model proxy system for shield tunnel structural state analysis described above, includes the following steps: S1: Spatiotemporal data construction of shield tunnel structural status Acquire 3D laser scanning data of the shield tunnel at different time points; Using the lining ring, segments and joints as basic spatial units, the scanned data were registered, fitted and deformation indexes were extracted. Construct a spatiotemporal database of shield tunnel structural status that includes time and spatial identifiers.

[0027] S2: Construction of a Knowledge Base for Shield Tunnel Structural Status Engineering Collect design specifications, operation and maintenance procedures, disease cases and expert experience related to the structural condition analysis of shield tunnels; The engineering knowledge is structured and a searchable index is established to form an engineering knowledge base for the structural status of shield tunnels.

[0028] S3: Understanding the Requirements for Shield Tunnel Structural Status Analysis Receive user requests for structural state analysis or evaluation in natural language form; By using a large-scale model for analyzing the structural state of shield tunnels as a proxy cognitive layer, semantic parsing of user input is performed to identify the analysis intent and generate structural state analysis targets.

[0029] S4: Task Planning for Shield Tunnel Structural Status Analysis The analysis task is decomposed according to the structural state analysis objectives; The system automatically plans the execution flow for data retrieval, indicator analysis, time series analysis, and spatial correlation analysis.

[0030] S5: Intelligent Understanding and Assessment of Shield Tunnel Structural Status Retrieve spatiotemporal data of structural status and related engineering knowledge; The input to the large language model is generated using a retrieval-enhanced approach; To comprehensively understand the structural state, determine trends, and conduct engineering semantic assessments.

[0031] S6: Generation and Expression of Shield Tunnel Structural Status Assessment Results Generate structural condition assessment conclusions and decision recommendations; The structural state analysis results are output to the user in natural language.

[0032] The present invention has the following beneficial effects: (1) It can systematically integrate multi-phase, high-precision structural deformation data of shield tunnels to achieve standardized and scalable spatiotemporal database management; (2) It can simultaneously analyze a large number of segments and joints with multiple deformation indices, and automatically identify the temporal evolution law and spatial correlation characteristics; (3) By introducing a large model agent with professional understanding capabilities, the engineering semantic expression of the structural state analysis results is realized; (4) Supports natural language interaction, which greatly reduces the threshold for engineers to use complex data and analysis tools; (5) Provide efficient and explainable intelligent decision support for the operation and maintenance of shield tunnels.

[0033] Through the aforementioned hierarchical architecture design, this invention enables efficient, systematic, and intelligent analysis of the structural status of shield tunnels under conditions of a large number of components and complex structural indicators, significantly improving the level of structural status awareness and decision support capabilities during the operation and maintenance phase of shield tunnels. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a framework diagram of the large model proxy system for shield tunnel structural state analysis according to the present invention; Figure 2 This is a flowchart of the method for analyzing the structural state of a shield tunnel according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0037] Reference Figure 1 A large model agent system for shield tunnel structural state analysis is proposed. It adopts a layered and decoupled system architecture, which includes, from top to bottom: natural language interaction layer, large model agent cognition layer, task planning and tool scheduling layer, structural state understanding and reasoning layer, and shield tunnel structural state spatiotemporal data and knowledge support layer.

[0038] (1) Spatiotemporal data and knowledge support layer of shield tunnel structure status As the underlying infrastructure of the system, it is used to uniformly support the multi-period spatiotemporal data resources and engineering professional knowledge resources required for the structural state analysis of shield tunnels.

[0039] The spatiotemporal data and knowledge support layer for shield tunnel structural status includes a shield tunnel structural status spatiotemporal database submodule and a shield tunnel structural status engineering knowledge base submodule.

[0040] The shield tunnel structural state spatiotemporal database submodule takes the shield tunnel lining structure as the object, performs refined modeling of the tunnel structure according to the spatial hierarchy of lining rings, segments and joints, assigns a unique number to each lining ring, segment and its circumferential and longitudinal joints, and uses it as the spatial index unit for structural state data.

[0041] The shield tunnel structure state spatiotemporal database submodule is used to store shield tunnel structure deformation data acquired at different times based on three-dimensional laser scanning technology. The shield tunnel structure deformation data includes at least whole-ring convergence deformation data, segment deflection deformation data, circumferential joint misalignment angle data, and longitudinal joint misalignment angle data.

[0042] The shield tunnel structural deformation data includes various types of data that are associated with and stored with corresponding time and spatial number information, thereby constructing a spatiotemporal data system for the shield tunnel structural state, supporting multi-condition queries and calls based on time, space, and structural index dimensions.

[0043] The shield tunnel engineering knowledge base submodule is used to store engineering knowledge data related to the structural state analysis of shield tunnels, including but not limited to shield tunnel design specifications, structural verification standards, operation and maintenance technical procedures, typical disease cases, and expert experience texts, which are used to provide professional knowledge constraints for understanding and reasoning about the structural state during the large model analysis process.

[0044] (2) Structural state understanding and reasoning layer It is used for engineering semantic understanding, state discrimination and reasoning analysis of spatiotemporal data of shield tunnel structural status, and is the core functional layer for realizing intelligent analysis and evaluation of shield tunnel structural status.

[0045] The structural state understanding and reasoning layer is built on a large language model and uses retrieval-enhanced generation technology. It combines the relevant knowledge content in the shield tunnel structural state engineering knowledge base sub-module with the shield tunnel structural deformation data to be analyzed, so that the large language model has both data-driven capability and engineering knowledge constraint capability during the analysis process.

[0046] The structural state understanding and reasoning layer is capable of comprehensive analysis and reasoning about the following: 1) The evolution trend of structural deformation data at different time scales; 2) Spatial distribution characteristics and relationships among different lining rings, segments, and joints; 3) The synergistic change characteristics among multiple structural deformation indices; 4) The matching relationship between structural deformation state and engineering specifications and empirical thresholds; In this way, the large language model can not only describe numerical changes, but also make comprehensive judgments on the structural state of shield tunnels with engineering semantics.

[0047] (3) Task planning and tool scheduling layer It is used to address the needs of structural condition analysis and evaluation of shield tunnels, and to perform engineering decomposition of analysis tasks, execution path planning, and automatic scheduling of analysis tools.

[0048] The task planning and tool scheduling layer includes: 1) Task parsing unit, used to receive the analysis target from the large model agent cognition layer and parse it into a structural state analysis task; 2) Task decomposition unit, used to decompose a comprehensive analysis task into multiple sub-tasks, wherein the sub-tasks include at least data screening task, indicator extraction task, time series analysis task and spatial correlation analysis task; 3) Execution order planning unit, used to determine the execution order of subtasks based on the logical dependencies between them; 4) Data and tool scheduling unit, used to retrieve relevant data from the shield tunnel structure state spatiotemporal database according to task requirements, and transmit the processing results to the structure state understanding and reasoning layer; By setting up this layer, the system can automatically complete the process of analyzing the structural status of complex shield tunnels without human intervention.

[0049] (4) Large model agent cognitive layer It is used to build intelligent agents with semantic understanding capabilities for shield tunnel engineering, structural state analysis and cognitive capabilities, and autonomous task planning capabilities. It is the core cognitive hub connecting user analysis needs and system analysis capabilities.

[0050] The large model agent cognitive layer is built on a large language model, and its functions include: 1) Perform semantic understanding on the user's natural language input and identify the implicit intention to analyze the structural state of the shield tunnel; 2) Mapping engineering problems expressed in natural language into structural state analysis objectives, including structural safety assessment, deformation trend analysis, abnormal component identification, and risk assessment; 3) Generate corresponding analysis strategies based on the analysis objectives, and pass the analysis strategies to the task planning and tool scheduling layer; 4) Through prompt word constraints and context control, the large model agent always participates in the analysis as a shield tunnel structural engineering and operation and maintenance expert, ensuring that the analysis results conform to engineering logic and professional common sense.

[0051] (5) Natural Language Interaction Layer This layer enables engineers and the system to interact and express evaluation results based on natural language for structural state analysis. The natural language interaction layer allows users to obtain professional analysis conclusions about the structural state of shield tunnels without directly operating complex data systems or analysis software.

[0052] Each layer interacts with the others through standardized data and semantic interfaces, and works collaboratively under the unified scheduling of the large model agent system, thereby enabling multi-source data understanding of the shield tunnel structure status, spatiotemporal correlation analysis, engineering semantic assessment, and decision support.

[0053] This large model agent system also includes a system coordination and constraint mechanism that runs through all the above layers. This mechanism is used to impose engineering rationality constraints on the analysis process and output results, ensuring that the system is used only for structural state analysis and decision support, and does not directly generate construction or control instructions.

[0054] Example 2 Reference Figure 2 A method for analyzing the structural state of shield tunnels, employing a large model proxy system for analyzing the structural state of shield tunnels as described in Example 1, includes the following steps: S1: Spatiotemporal data construction of shield tunnel structural status Acquire 3D laser scanning data of the shield tunnel at different time points; Using the lining ring, segments and joints as basic spatial units, the scanned data were registered, fitted and deformation indexes were extracted. Construct a spatiotemporal database of shield tunnel structural status that includes time and spatial identifiers.

[0055] S2: Construction of a Knowledge Base for Shield Tunnel Structural Status Engineering Collect design specifications, operation and maintenance procedures, disease cases and expert experience related to the structural condition analysis of shield tunnels; The engineering knowledge is structured and a searchable index is established to form an engineering knowledge base for the structural status of shield tunnels.

[0056] S3: Understanding the Requirements for Shield Tunnel Structural Status Analysis Receive user requests for structural state analysis or evaluation in natural language form; By using a large-scale model for analyzing the structural state of shield tunnels as a proxy cognitive layer, semantic parsing of user input is performed to identify the analysis intent and generate structural state analysis targets.

[0057] S4: Task Planning for Shield Tunnel Structural Status Analysis The analysis task is decomposed according to the structural state analysis objectives; The system automatically plans the execution flow for data retrieval, indicator analysis, time series analysis, and spatial correlation analysis.

[0058] S5: Intelligent Understanding and Assessment of Shield Tunnel Structural Status Retrieve spatiotemporal data of structural status and related engineering knowledge; The input to the large language model is generated using a retrieval-enhanced approach; To comprehensively understand the structural state, determine trends, and conduct engineering semantic assessments.

[0059] S6: Generation and Expression of Shield Tunnel Structural Status Assessment Results Generate structural condition assessment conclusions and decision recommendations; The structural state analysis results are output to the user in natural language.

[0060] As demonstrated in Examples 1 and 2, through the aforementioned layered architecture design, this invention enables efficient, systematic, and intelligent analysis of the structural status of shield tunnels, even under conditions of a large number of components and complex structural indicators. This significantly improves the understanding of structural status and decision support capabilities during the operation and maintenance phase of shield tunnels. It can systematically integrate multi-phase, high-precision structural deformation data of shield tunnels, achieving standardized and scalable spatiotemporal database management. It can simultaneously analyze multiple deformation indicators of a large number of segments and joints, automatically identifying temporal evolution patterns and spatial correlation characteristics. By introducing a large model agent with professional understanding capabilities, it achieves the engineering semantic expression of structural status analysis results. It supports natural language interaction, significantly reducing the barrier for engineers to use complex data and analysis tools. It provides efficient and interpretable intelligent decision support for the operation and maintenance of shield tunnels.

[0061] Example 3 Reference Figure 2 A method for analyzing the structural state of shield tunnels, employing a large model proxy system for analyzing the structural state of shield tunnels as described in Example 1, includes the following steps: S1: Construction and Data Organization of Spatiotemporal Database of Shield Tunnel Structural Status This corresponds to the spatiotemporal data and knowledge support layer of the shield tunnel structure status in the system architecture.

[0062] First, based on the total mileage of the shield tunnel, the tunnel lining structure is spatially hierarchically divided. Specifically, the tunnel is divided longitudinally into several lining rings, and each lining ring is assigned a unique ring number. Within each lining ring, the precast segments constituting that lining ring are numbered. At the same time, the circumferential joints formed between adjacent lining rings, as well as the longitudinal joints formed between adjacent segments within the same lining ring, are numbered, thus forming a structural indexing system with "lining ring-segment-joint" as the basic spatial unit.

[0063] During the operation of the shield tunnel, three-dimensional laser scanning of the tunnel was carried out at different time points to obtain high-density point cloud data of the tunnel's inner surface. By registering the point cloud data from multiple periods in a unified coordinate system and combining it with the tunnel design geometric model, feature extraction and geometric fitting calculations were performed on the point cloud data to obtain the structural deformation indices of each lining ring, segment, and joint at each time point.

[0064] In this embodiment, the structural deformation index includes at least: (1) The whole ring convergence deformation index characterizing the overall cross-sectional shape change of the lining ring; (2) The segment flexural deformation index characterizing the degree of bending deformation of a single segment; (3) The index of circumferential joint misalignment, which characterizes the relative displacement between adjacent lining rings; (4) The longitudinal joint misalignment angle index characterizes the relative misalignment between adjacent segments within the same lining ring.

[0065] The aforementioned deformation indices, along with their corresponding time information and spatial numbers, are stored together in the shield tunnel structural state spatiotemporal database, thus forming a structural state dataset with clear time and spatial attributes, providing a data foundation for subsequent structural state analysis.

[0066] S2: Construction and Retrieval Enhancement Mechanism of the Shield Tunnel Engineering Knowledge Base It corresponds to the spatiotemporal data and knowledge support layer of shield tunnel structure status, and provides knowledge support for the structural status understanding and reasoning layer.

[0067] A knowledge base for shield tunnel engineering will be constructed to store engineering knowledge related to the structural condition analysis of shield tunnels. This engineering knowledge includes, but is not limited to, shield tunnel design specifications, operation and maintenance technical procedures, structural calculation standards, descriptions of typical defects, and analysis reports of existing engineering cases.

[0068] To facilitate the large model's retrieval of relevant knowledge during the analysis process, the engineering knowledge content is structured and organized according to dimensions such as "structure type - deformation type - engineering significance - disposal suggestions," and a searchable index system is established.

[0069] During structural state analysis, the system automatically retrieves knowledge fragments related to the current analysis object and analysis indicators from the engineering knowledge base according to the content of the analysis task, and inputs the knowledge fragments as context information into the structural state understanding and reasoning layer, thereby realizing structural state analysis based on retrieval enhancement.

[0070] S3: Structural State Understanding and Reasoning Process Based on Large Models This corresponds to the structural state understanding and reasoning layer in the system architecture.

[0071] When the system needs to analyze the structural state of a certain tunnel section, the task planning and tool scheduling layer first retrieves the structural deformation data of the target lining ring, segments and joints within a specified time range from the shield tunnel structural state spatiotemporal database.

[0072] Subsequently, the structural deformation data and the engineering knowledge obtained through the retrieval enhancement mechanism are input into the large language model. The large language model is then guided to analyze the data in the role of a shield tunnel structural engineering and operation and maintenance expert through preset prompt word templates.

[0073] During the analysis process, the large language model can simultaneously consider the numerical magnitude of structural deformation, temporal variation trends, spatial distribution characteristics, and engineering specification constraints, perform comprehensive reasoning on the structural state, and output an engineering semantic description of the structural state, such as whether the structural state is stable, whether there is an abnormal evolution trend, and possible engineering causes.

[0074] This implementation method enables structural state analysis results to go beyond numerical descriptions and form analytical conclusions that are understandable in engineering contexts.

[0075] S4: Large Model Agent-Driven Task Planning and Analysis Process Execution This corresponds to the large model agent cognition layer and the task planning and tool scheduling layer in the system architecture.

[0076] When engineers submit a structural state analysis request via natural language, the natural language interaction layer first receives the input and then passes it to the large model agent cognition layer.

[0077] The large model's agent cognitive layer performs semantic understanding on user input, identifies the implicit structural state analysis intent, and transforms this intent into explicit analysis objectives. For example, "Is the recent structure of this section abnormal?" is transformed into a task of analyzing structural deformation trends and identifying anomalies in a specified section.

[0078] Subsequently, the large model agent cognition layer generates corresponding analysis strategies based on the analysis objectives and sends these strategies to the task planning and tool scheduling layer. The task planning and tool scheduling layer parses the analysis strategies, decomposes them into multiple sub-tasks, and sequentially performs operations such as data filtering, indicator extraction, time series analysis, and spatial correlation analysis according to a predetermined execution order.

[0079] The above method enables a shield tunnel structural state analysis process that is automatically driven by a large model agent, reducing manual intervention and improving analysis efficiency.

[0080] S5: Structural State Assessment and Decision Suggestion Generation under Natural Language Interaction This corresponds to the natural language interaction layer and the structural state understanding and reasoning layer in the system architecture.

[0081] The analysis results output by the structural state understanding and reasoning layer are further organized into structural state assessment information, including the current structural state level, major abnormal locations, deformation development trends, and potential risk warnings.

[0082] The structural status assessment information is fed back to the user through a natural language interaction layer. The user can obtain the system's analysis conclusions and decision suggestions in natural language, such as whether encrypted monitoring is needed, or whether special testing or maintenance is recommended.

[0083] Through natural language interaction, the structural state analysis results can be quickly understood and used by engineers from different professional backgrounds.

[0084] S6: System operation constraints and assurance of the reliability of analysis results This corresponds to the system coordination and constraint mechanisms in the system architecture.

[0085] During the large model agent-driven analysis process, the system constrains and controls the analysis task type and output content to ensure that the system is only used for shield tunnel structural status analysis and operation and maintenance decision support, and does not directly generate construction control or operation instructions.

[0086] Meanwhile, the system can verify the engineering rationality of the analysis results output by the large model. For example, it can compare the analysis conclusions with the limit requirements in the engineering specifications, mark or correct conclusions that are obviously inconsistent with common sense in engineering, thereby improving the credibility and engineering applicability of the analysis results.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A large-scale model proxy system for structural state analysis of shield tunnels, characterized in that: The system architecture adopts a layered and decoupled structure, including: (1) Spatiotemporal data and knowledge support layer of shield tunnel structure status As the underlying infrastructure of the system, it is used to uniformly support the multi-period spatiotemporal data resources and engineering professional knowledge resources required for the structural state analysis of shield tunnels; (2) Structural state understanding and reasoning layer It is used for engineering semantic understanding, state discrimination and reasoning analysis of spatiotemporal data of shield tunnel structure status, and is the core functional layer for realizing intelligent analysis and evaluation of shield tunnel structure status. (3) Task planning and tool scheduling layer It is used to address the needs of structural status analysis and evaluation of shield tunnels, and to perform engineering decomposition of analysis tasks, execution path planning, and automatic scheduling of analysis tools. (4) Large model agent cognitive layer It is used to build an intelligent agent with semantic understanding of shield tunnel engineering, cognitive ability to analyze structural status, and autonomous task planning capabilities. It is the core cognitive hub connecting user analysis needs and system analysis capabilities. (5) Natural Language Interaction Layer It is used to enable engineers and the system to interact with each other based on natural language for structural state analysis and to express evaluation results. Each layer interacts with the others through standardized data and semantic interfaces, and works collaboratively under the unified scheduling of the large model agent system, thereby enabling multi-source data understanding of the shield tunnel structure status, spatiotemporal correlation analysis, engineering semantic assessment, and decision support.

2. The large model proxy system for shield tunnel structural state analysis as described in claim 1, characterized in that: The shield tunnel structure status spatiotemporal data and knowledge support layer includes a shield tunnel structure status spatiotemporal database submodule and a shield tunnel structure status engineering knowledge base submodule. The shield tunnel structural state spatiotemporal database submodule takes the shield tunnel lining structure as the object, performs refined modeling of the tunnel structure according to the spatial hierarchy of lining rings, segments and joints, assigns a unique number to each lining ring, segment and its ring joint and longitudinal joint, and uses it as the spatial index unit of structural state data. The shield tunnel structure state spatiotemporal database submodule is used to store shield tunnel structure deformation data acquired at different times based on three-dimensional laser scanning technology. The shield tunnel structure deformation data includes at least whole ring convergence deformation data, segment deflection deformation data, circumferential joint misalignment angle data, and longitudinal joint misalignment angle data. The various types of data included in the shield tunnel structure deformation data are all stored in association with the corresponding time information and spatial number information, thereby constructing a spatiotemporal data system of shield tunnel structure status, supporting multi-condition query and retrieval according to time dimension, spatial dimension and structural index dimension; The shield tunnel engineering knowledge base submodule is used to store engineering knowledge data related to the structural state analysis of shield tunnels, including but not limited to shield tunnel design specifications, structural verification standards, operation and maintenance technical procedures, typical disease cases, and expert experience texts, which are used to provide professional knowledge constraints for understanding and reasoning about the structural state during the large model analysis process.

3. The large model proxy system for shield tunnel structural state analysis as described in claim 2, characterized in that: The structural state understanding and reasoning layer is built on a large language model and uses retrieval-enhanced generation technology. It combines the relevant knowledge content in the shield tunnel structural state engineering knowledge base sub-module with the shield tunnel structural deformation data to be analyzed, so that the large language model has both data-driven capability and engineering knowledge constraint capability during the analysis process.

4. The large model proxy system for shield tunnel structural state analysis as described in claim 3, characterized in that: The structural state understanding and reasoning layer is capable of comprehensive analysis and reasoning about the following: 1) The evolution trend of structural deformation data at different time scales; 2) Spatial distribution characteristics and relationships among different lining rings, segments, and joints; 3) The synergistic change characteristics among multiple structural deformation indices; 4) The matching relationship between structural deformation state and engineering specifications and empirical thresholds; In this way, the large language model can not only describe numerical changes, but also make comprehensive judgments on the structural state of shield tunnels with engineering semantics.

5. The large model proxy system for shield tunnel structural state analysis as described in claim 4, characterized in that: The task planning and tool scheduling layer includes: 1) Task parsing unit, used to receive the analysis target from the large model agent cognition layer and parse it into a structural state analysis task; 2) Task decomposition unit, used to decompose a comprehensive analysis task into multiple sub-tasks, wherein the sub-tasks include at least data screening task, indicator extraction task, time series analysis task and spatial correlation analysis task; 3) Execution order planning unit, used to determine the execution order of subtasks based on the logical dependencies between them; 4) Data and tool scheduling unit, used to retrieve relevant data from the shield tunnel structure state spatiotemporal database according to task requirements, and transmit the processing results to the structure state understanding and reasoning layer; By setting up this layer, the system can automatically complete the process of analyzing the structural status of complex shield tunnels without human intervention.

6. The large model proxy system for shield tunnel structural state analysis as described in claim 5, characterized in that: The large model agent cognitive layer is built on a large language model, and its functions include: 1) Perform semantic understanding on the user's natural language input and identify the implicit intention to analyze the structural state of the shield tunnel; 2) Mapping engineering problems expressed in natural language into structural state analysis objectives, including structural safety assessment, deformation trend analysis, abnormal component identification, and risk assessment; 3) Generate corresponding analysis strategies based on the analysis objectives, and pass the analysis strategies to the task planning and tool scheduling layer; 4) Through prompt word constraints and context control, the large model agent always participates in the analysis as a shield tunnel structural engineering and operation and maintenance expert, ensuring that the analysis results conform to engineering logic and professional common sense.

7. The large model proxy system for shield tunnel structural state analysis as described in claim 6, characterized in that: The natural language interaction layer enables users to obtain professional analysis conclusions on the structural status of shield tunnels through natural language interaction, without having to directly operate complex data systems or analysis software.

8. The large model proxy system for structural state analysis of shield tunnels as described in any one of claims 1-7, characterized in that: This large model agent system also includes a system coordination and constraint mechanism that runs through all the above layers. This mechanism is used to impose engineering rationality constraints on the analysis process and output results, ensuring that the system is used only for structural state analysis and decision support, and does not directly generate construction or control instructions.

9. A method for analyzing the structural state of shield tunnels, characterized in that: Using a large model proxy system for structural state analysis of shield tunnels as described in any one of claims 1-8, the method includes the following steps: S1: Spatiotemporal data construction of shield tunnel structural status Acquire 3D laser scanning data of the shield tunnel at different time points; Using the lining ring, segments and joints as basic spatial units, the scanned data were registered, fitted and deformation indexes were extracted. Construct a spatiotemporal database of shield tunnel structural status containing time and spatial identifiers; S2: Construction of a Knowledge Base for Shield Tunnel Structural Status Engineering Collect design specifications, operation and maintenance procedures, disease cases and expert experience related to the structural condition analysis of shield tunnels; The engineering knowledge is structured and a searchable index is established to form an engineering knowledge base for the structural status of shield tunnels; S3: Understanding the Requirements for Shield Tunnel Structural Status Analysis Receive user requests for structural state analysis or evaluation in natural language form; By using the large model of shield tunnel structural state analysis to act as a cognitive layer, semantic parsing of user input is performed to identify the analysis intent and generate structural state analysis targets. S4: Task Planning for Shield Tunnel Structural Status Analysis The analysis task is decomposed according to the structural state analysis objectives; Automatically plan the execution flow of data retrieval, indicator analysis, time series analysis, and spatial correlation analysis; S5: Intelligent Understanding and Assessment of Shield Tunnel Structural Status Retrieve spatiotemporal data of structural status and related engineering knowledge; The input to the large language model is generated using a retrieval-enhanced approach; To comprehensively understand the structural status, determine trends, and conduct engineering semantic assessments; S6: Generation and Expression of Shield Tunnel Structural Status Assessment Results Generate structural condition assessment conclusions and decision recommendations; The structural state analysis results are output to the user in natural language.

Images