A method and system for optimizing tunnel spatial layout and extrapolating digital twin parameters, along with storage media and computer program products.

Abstract

This invention discloses a method and system for tunnel spatial layout optimization and digital twin parameter extrapolation intelligent agent, along with storage media and computer program products. It revolves around a three-layer operational architecture—parameter information layer, physical constraint layer, and interaction tool layer—with intelligent agent scheduling as the core. The system analyzes design objectives, transforms them into spatial parameter combinations, generates candidate cross-section schemes under different conditions, solves for cross-section layout, verifies geometric feasibility, constructs cross-section layout evaluation indicators, and achieves comprehensive ranking and recommendation. Finally, the recommended schemes are transformed into two-dimensional vector cross-section diagrams, three-dimensional BIM models, digital twin visualization models, extrapolation and verification records, and engineering delivery documents. It also supports the construction of standard tunnel cross-section models, the assembly of models along the tunnel route, and the construction of operational simulation scenarios. The advantage of this invention is that it achieves a closed-loop forward design and generative design process, encompassing the entire process from design requirement input, intelligent agent spatial parameter extrapolation, spatial layout verification and evaluation, to digital twin delivery.

Description

Technical Field

[0001] This invention relates to the fields of digital forward design, parametric modeling, engineering scheme evaluation and artificial intelligence-assisted generative design in tunnel engineering, and particularly to a method and system for tunnel spatial layout optimization and digital twin parameter extrapolation intelligent agent, storage medium and computer program product. Background Technology

[0002] With the continuous expansion of urban transportation infrastructure construction, engineering forms such as highway tunnels, urban underground roads, integrated transportation tunnels, and composite utility tunnels are becoming increasingly common. As a complex space where traffic functions, equipment systems, structural boundaries, and operation and maintenance space all interact, the rationality of tunnel layout directly affects project costs, space utilization efficiency, operational safety, and ease of future maintenance. Traditional tunnel design typically involves formulating schemes based on specifications, standard cross-section drawings, and existing engineering experience, followed by manual verification of constraints to arrive at a feasible cross-section design.

[0003] With the development of BIM and digital twin technologies, tunnel and underground space design is gradually shifting from two-dimensional drawing design to three-dimensional parametric forward design and full-process visual interaction. Through parametric models, different cross-sectional schemes can be quickly generated based on different tunnel design parameters, and further realized through interaction and expression on BIM or digital twin platforms, including component-level model expression, visual inspection and optimization of schemes, and traceable management of design results.

[0004] However, existing tunnel cross-section layout design methods still have certain limitations. Existing parametric modeling methods mostly focus on model generation after given parameters, i.e., rapid modeling based on manually preset cross-section dimensions. However, they lack the ability to extrapolate solutions for complex combinations of design parameters and boundary constraints, and struggle to automatically determine which parameter combinations are more feasible and superior. Even when multiple cross-section models are generated during the design phase, designers often still need to rely on their experience to judge their space utilization efficiency, boundary safety margins, and overall layout rationality.

[0005] Specifically, the existing technology mainly has the following problems: 1) Inefficient spatial layout parameter extrapolation: Existing methods usually rely on manual experience, standard cross sections or fixed parameterized templates for repeated calculations, making it difficult to automatically decompose design tasks based on design parameters such as the number of lanes, tunnel radius, utility tunnel requirements, and clearance requirements, and generate candidate schemes under different spatial layout combinations. 2) Lack of traceable feedback in trial and error: In the process of traditional manual trial calculation or ordinary parametric modeling, when a spatial layout scheme fails to meet the boundary, clearance or structural constraints, designers usually have to check it item by item. There is no mechanism to automatically record the reasons for failure, the location of constraint conflicts and the direction of parameter adjustment. 3) Lack of closed-loop linkage and optimization mechanism between model generation and design evaluation: Existing digital modeling processes are mostly used to express the final solution, and have not yet fully served the solution derivation process itself. It is difficult to unify agent scheduling, parameterized solution, index calculation, 3D modeling and digital twin visualization, and solution optimization into the same workflow.

[0006] The above are the areas that need to be improved in this application. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a method and system for optimizing tunnel spatial layout and digital twin parameter extrapolation, as well as storage media and computer program products, to improve the automation level, space utilization efficiency evaluation capability, and full-process traceable visualization expression capability of the cross-section and spatial layout scheme design of highway tunnels, urban underground roads, and integrated transportation tunnels, and to meet the engineering design needs of rapid generation, comparison, verification, and display of tunnel cross-section and spatial layout schemes.

[0008] To address the above technical problems, this invention provides an intelligent agent system for tunnel spatial layout optimization and digital twin parameter inference. The intelligent agent system constructs a three-layer operational architecture centered around the agent scheduling core: a parameter information layer, a physical constraint layer, and an interaction tool layer. Wherein: The intelligent agent uses a large language model as its reasoning kernel to perform semantic parsing of design goals, task decomposition, tool scheduling, result reading, feedback iteration, and solution interpretation. The parameter information layer is used to organize the various engineering information received, processed, recorded, and output during the intelligent agent's deduction process; The physical constraint layer is used to limit the system process constraints, physical boundary constraints, and user preference constraints that the intelligent agent's inference process must satisfy. The interactive tool layer provides capabilities for cross-sectional parameter analysis, evaluation quantification calculation, parameter modeling, and output delivery of executable results.

[0009] The parameter information layer includes design target information, spatial parameter information, simulation status information, and final scheme information; The design target information includes tunnel type, number of lanes, preference for single or double cross-section, cross-section radius range, utility tunnel requirements, equipment layout requirements, and operational scenario requirements. The spatial parameter information includes vehicle clearance, lane width, structural slab thickness, pipe gallery width, pipe gallery height, equipment installation space, safety margin, inter-floor height, and cross-sectional layout. The simulation status information includes candidate scheme parameters, tool call results, feasibility status, failure reasons, indicator calculation results, parameter adjustment trajectory, and manual review comments; The final solution information includes recommended cross-sectional schemes, scheme ranking results, evaluation index results, two-dimensional cross-sectional diagrams, three-dimensional models, digital twin visualization scenes, and engineering delivery documents.

[0010] The physical constraint layer includes system process constraints, physical boundary constraints, and user preference constraints. The system process constraints, namely system prompts and predefined deduction process constraints, are used to limit the agent to execute according to the preset task parsing, candidate generation, cross-section analysis, evaluation decision, feedback iteration, recommended solution selection and engineering delivery process. The physical boundary constraints, namely physical boundaries and code hard constraints, are formulated around geometric boundaries and codes to limit the candidate cross-section schemes to meet vehicle clearance constraints, lining boundary constraints, pipe gallery space constraints, equipment installation space constraints, structural thickness constraints, and safety margin constraints. The user preference constraints are soft constraints of user preferences and indicator weights. The intelligent agent inference scheme satisfies the hard constraints while also meeting human-centered design requirements, including priority of space utilization, traffic capacity, safety margin, vertical balance, engineering cost, operation and maintenance, as well as the evaluation weights corresponding to functional space utilization rate, ineffective cavity rate, single lane cross-sectional area, minimum boundary margin, net height margin, vertical balance, and user-defined indicators. The agent performs constraint identification, conflict judgment, parameter adjustment and recommended scheme ranking on the candidate cross-section schemes according to the physical constraint layer, and writes the constraint verification results, constraint conflict reasons and soft constraint weight configuration into the inference state information.

[0011] The interactive tool layer includes parameter analysis tools, evaluation quantification calculation tools, parameter modeling tools, and result delivery and export tools. The parameter analysis tool is used to calculate the positional relationship of the vehicle space, pipe gallery space, structural slab, equipment space, arch space and remaining space based on the candidate cross-section parameters, and output the geometric feasibility status. The evaluation and quantitative calculation tool is used to calculate functional space utilization, ineffective cavity rate, single-lane cross-sectional area, minimum boundary margin, net height margin, vertical balance, and user-defined indicators. The parametric modeling tool is used to generate two-dimensional vector cross-sectional views, three-dimensional BIM models, and digital twin visualization models based on the recommended scheme; The deliverables export tool is used to output cross-sectional layout parameter tables, evaluation index calculation tables, constraint verification results, records of reasons for failure of infeasible solutions, agent inference logs, and recommended solution description documents.

[0012] The intelligent agent system, through the aforementioned three-layer operating architecture, continuously reads and updates the deduction state at the parameter information layer, determines whether candidate solutions meet hard constraints and user preference constraints at the physical constraint layer, dynamically calls engineering calculation and modeling tools at the tool interaction layer, and generates parameter adjustment strategies based on the tool's returned results. This enables an interpretable, traceable, and scalable intelligent agent-driven deduction process, from design target input to candidate solution generation, cross-section analysis verification, indicator evaluation and recommendation, feedback iteration, and digital twin delivery.

[0013] This invention also provides a method for optimizing tunnel spatial layout and for digital twin parameter inference using an intelligent agent, comprising the following steps: Step S1: Design target input; Input the tunnel cross-section design objectives and basic constraints in text form. The tunnel cross-section design objectives include tunnel type, number of lanes, cross-section layout, tunnel inner contour radius or radius range, vehicle clearance height, lane width, utility tunnel space requirements, equipment installation space requirements, lining boundary requirements, safety margin requirements, and operation and maintenance space requirements. The cross-sectional layout includes single-layer and double-layer layouts. The single-layer layout includes an upper vehicular space and a lower utility tunnel space; the double-layer layout includes an upper vehicular space, a lower vehicular space, and a lower utility tunnel space.

[0014] Step S2: Preliminary plan organization; Based on the tunnel cross-section design objectives input in step S1, the intelligent agent parses and semantically processes the tunnel cross-section design task, decomposes the overall design requirements into candidate parameter generation, cross-section layout analysis, evaluation index calculation, feedback iteration, recommended scheme selection and engineering delivery sub-tasks, and generates candidate cross-section schemes for single-layer cross-sections, double-layer cross-sections and different radii and lane combinations. In this step, the agent transforms the tunnel cross-section design target into structured parameters and determines the range of candidate parameters based on preset rules or the reasoning results of a large language model: determining the road width based on the number of lanes, determining the vehicle space height based on the clearance requirements, determining the bottom pipe gallery size based on the pipe gallery requirements, and generating candidate inputs for single-layer or double-layer cross-section layouts.

[0015] The i-th candidate cross-section scheme is represented as a design variable vector. The continuous variables include the inner contour radius of the tunnel. Elevation of each vehicular road slab Vehicle clearance height Structural plate thickness Pipe gallery width pipe gallery height Equipment installation space dimensions and safety margin; discrete variables include the number of lanes. Single-layer or double-layer layout And the component layout category. Its parametric expression is: (1); In equation (1), The initial search domain is composed of the input range, the allowed intervals, and the constructed set of values. The agent performs combined enumeration based on discrete variables and uses uniform sampling or adaptive sampling based on historical inference results for continuous variables to form a candidate set. This transforms the natural language design objective into a unified data structure that can be directly solved by the cross-sectional parameter analysis tool.

[0016] Step S3: Cross-sectional and spatial layout analysis and hard constraint verification; For each candidate cross-section scheme obtained in step S2, the agent calls the cross-section parameter analysis tool to perform spatial layout calculations and verifies the feasibility of the calculation results based on geometric boundaries and hard constraints of specifications. The cross-section parameter analysis tool calculates the positional relationships of the vehicle space, structural slab space, pipe gallery space, equipment installation space, arch space, and remaining space in the cross-section based on the tunnel's inner contour boundary, vehicle clearance, lane width, structural slab thickness, pipe gallery dimensions, and safety margin, and records the feasibility status and reasons for infeasibility of the candidate schemes.

[0017] For a single-layer cross-section, the cross-section parameter analysis tool calculates the upper vehicle space, the lower utility tunnel space, and their relative positions with the tunnel's circular boundary. For a double-layer cross-section, the cross-section parameter analysis tool calculates the upper vehicle space, the lower vehicle space, the lower utility tunnel space, the intermediate partition, and their relative positions with the tunnel's circular boundary.

[0018] A two-dimensional rectangular coordinate system is established with the center of the tunnel cross-section as the origin. The inner contour of the tunnel adopts a radius of... The circle or equivalent boundary described by a parametric curve; for the inner contour of the circle, given the vertical coordinates Available half-width for: (2); To ensure consistent verification of whether the vehicle clearance, utility tunnel space, equipment installation space, and structural slab functional areas exceed the tunnel's internal contour, each of these functional areas is represented by a rectangular envelope completely covering its outer boundary. For the j-th functional area, the coordinates of the center of its rectangular envelope are... Half width is Half height is The preset safety margin is The inner contour inclusion is checked using the four corner points or discrete boundary sampling points of the component. The boundary constraints under a circular cross-section are expressed as follows: (3); In equation (3), and These values ​​are 1 or -1, representing the four corner points of the rectangular envelope; when When, it indicates that the functional area and its safety margin do not exceed the tunnel's inner contour; when When a candidate scheme is found to have a boundary conflict, it indicates that at least one corner point exceeds the allowed boundary.

[0019] The spacing constraints, vehicle clearance constraints, and pipe gallery dimension constraints between adjacent components j and l are expressed as follows: (4); After completing the spatial layout calculation, the intelligent agent performs a geometric feasibility check on the candidate cross-section schemes based on preset hard constraints; the hard constraints include vehicle clearance constraints, pipe gallery space constraints, equipment installation space constraints, lining boundary constraints, structural thickness constraints, and safety margin constraints.

[0020] If a candidate cross-section scheme satisfies the above hard constraints, the agent marks it as a feasible scheme and records its cross-sectional spatial parameters, component positional relationships, and hard constraint verification results; if a candidate cross-section scheme does not satisfy the above hard constraints, the agent marks it as an infeasible scheme and records the reasons for infeasibility; the reasons for infeasibility include, but are not limited to, insufficient vehicle clearance, insufficient pipe gallery space, insufficient equipment installation space, insufficient cross-sectional boundary margin, insufficient inter-layer distance of structural layers, and layout space exceeding the inner contour boundary of the tunnel.

[0021] The intelligent agent further incorporates feasibility status, reasons for infeasibility, and hard constraint conflict information into the inference status information for solution iteration, multi-index evaluation, and recommended solution selection.

[0022] Step S4: Multi-indicator evaluation and decision making; Evaluation indicators are calculated for the feasible cross-section schemes verified in step S3 to form cross-section layout evaluation results for lateral comparison of schemes. The evaluation indicators include functional space utilization rate, ineffective cavity rate, single-lane cross-sectional area, minimum boundary margin, net height margin, and vertical balance; wherein: The functional space utilization rate is used to characterize the proportion of vehicle space and utility tunnel space to the tunnel's inner contour area. It is calculated by dividing the sum of the road space area and the utility tunnel area by the tunnel's inner contour area. A higher value indicates more efficient use of the cross-sectional space. The ineffective cavity ratio is used to characterize the proportion of remaining space within the cross-section that is not utilized by the vehicle space and the utility tunnel space. It is calculated by dividing the remaining area by the outline area inside the tunnel. The lower the value, the less space is wasted. The single-lane cross-sectional area is used to characterize the tunnel cross-sectional area consumption corresponding to a unit number of lanes. It is calculated by dividing the tunnel inner contour area by the total number of lanes. The lower the value, the less cross-sectional space is required per unit of traffic function. The minimum boundary margin is used to characterize the minimum safe distance between the boundary of each functional component in the cross section and the inner contour of the tunnel lining. It is calculated as min(distance from the boundary of each functional component to the inner contour of the tunnel lining). The higher the value, the safer the boundary arrangement and the larger the adjustment margin. The headroom margin is used to characterize the minimum margin between the actual vehicle headroom height and the design headroom requirement. It is calculated as min(actual vehicle headroom height of each vehicle level - design headroom requirement). The higher the value, the more sufficient the vehicle headroom. The vertical balance is used to characterize the degree of balance between the top and bottom remaining spaces of the cross section. It is obtained by normalizing the difference between the top and bottom remaining spaces. The closer the value is to 1, the more balanced the upper and lower remaining spaces are.

[0023] For the j-th index of the i-th solution in the feasible solution set Dimensionless normalization is performed using the range method. The positive and negative indices are calculated using the following formulas: (5); (6); Weights are set based on user preferences. The comprehensive evaluation score of the candidate solutions Represented as: (7); When a certain indicator has a range of zero in the current feasible solution set, its normalized value is set to 1 or it is temporarily removed from the current round of discriminative ranking, and the remaining weights are renormalized. In addition to weighted scoring, Pareto non-dominated ranking is used to retain candidate solutions with different trade-offs between spatial efficiency, safety margin, and engineering economy, and then the recommended solution is selected from the Pareto frontier based on user preferences.

[0024] Step S5: Agent feedback iteration; Based on the feasibility status, infeasibility reasons, evaluation index results, user preferences, and manual review opinions of the candidate cross-section schemes, the intelligent agent generates the next round of parameter adjustment strategy, and calls the cross-section parameter analysis tool, hard constraint verification tool, and evaluation quantification calculation tool again, forming a closed-loop iterative process of candidate generation, cross-section analysis, hard constraint verification, index evaluation, and feedback adjustment. For infeasible candidate solutions, the intelligent agent determines the corresponding parameter adjustment direction based on the cause of infeasibility. When the cause of infeasibility is insufficient vehicular clearance, the intelligent agent adjusts the tunnel's inner contour radius, road slab elevation, inter-layer height, or cross-sectional layout. When the cause of infeasibility is insufficient utility tunnel space, the intelligent agent adjusts the utility tunnel width, utility tunnel height, utility tunnel layer, or the relative position between the vehicular space and the utility tunnel. When the cause of infeasibility is insufficient equipment installation space, the intelligent agent adjusts the equipment layout area, arch space, lateral safety margin, or the dimensions of auxiliary components. When the cause of infeasibility is insufficient lining boundary margin, the intelligent agent increases the cross-sectional radius, adjusts the number of lanes, changes the single-layer or double-layer layout, or reduces non-critical functional space. When the cause of infeasibility is insufficient structural inter-layer distance, the intelligent agent adjusts the structural slab thickness, inter-layer clearance, or the elevation relationship between upper and lower road layers.

[0025] For feasible candidate solutions that already meet the hard constraints, the agent determines the optimization direction based on the evaluation index results. When the functional space utilization rate is low or the ineffective cavity rate is high, the agent adjusts the layout ratio of the vehicle space, the utility tunnel space, and the remaining space. When the cross-sectional area of ​​a single lane is too large, the agent attempts to optimize the lane combination, cross-sectional radius, or single-layer / double-layer layout. When the minimum boundary margin or net height margin is low, the agent increases the corresponding safety margin or adjusts the component position. When the vertical balance is low, the agent adjusts the upper arch space, the bottom utility tunnel space, and the elevation of the vehicle layer to make the distribution of the remaining cross-sectional space more balanced.

[0026] The agent generates new candidate cross-section parameter sets based on the above adjustment strategy, and calls the cross-section parameter analysis tool and evaluation quantification calculation tool again, forming a closed-loop iterative process of "candidate generation - cross-section analysis - hard constraint verification - indicator evaluation - feedback adjustment". Each iteration records the candidate scheme parameters, tool call results, feasibility status, failure reasons, indicator changes, parameter adjustment trajectory, and user review comments, forming a traceable deduction record.

[0027] The intelligent agent does not directly replace deterministic engineering calculations. Instead, it selects variables to be adjusted based on constraint violation rates and performance limitations, automatically infers, and returns new updated parameters. In round t, bounded updates are used for continuous design variables: (8); in, This is a parameter adjustment vector formed based on the main conflict constraints, the direction of indicator improvement, and historical effective adjustment records. To adjust the step size, This indicates that the updated results will be projected onto the initial search domain. For discrete variables such as layout and number of lanes, replacement is performed according to predefined neighborhood rules. New candidate solutions prioritize update directions that reduce constraint violation rates. When the preset number of iterations is reached, the number of feasible solutions that meet the requirements is obtained, or the current solution set is confirmed by manual review to meet the design requirements, the agent stops feedback iteration and outputs the final feasible candidate solution set and its evaluation index results to step S6. (9); The stopping condition for the above feedback iteration includes at least one of the following: (1) The number of iterations t reaches the set upper limit; (2) The improvement in the optimal comprehensive evaluation score for L consecutive rounds is less than the threshold. ; (3) Obtain no less than the preset number of feasible solutions; (4) Receive user confirmation instruction; By setting the above stopping conditions, invalid loops are avoided and the deduction process is guaranteed to be reproducible.

[0028] Step S6: Recommended solution selection; Based on the final feasible candidate scheme set and its evaluation index results obtained in step S5, multiple candidate cross-section schemes are ranked and filtered to obtain recommended cross-section schemes. The selection of recommended cross-section schemes is based on ranking according to a single evaluation index, or on comprehensive ranking based on a multi-index normalized weighted score. After the ranking results are output, the recommended schemes, index weights, or custom indicators are reviewed and corrected. The agent then returns to step S5 based on the feedback to iterate the parameters again.

[0029] In the comprehensive ranking based on multi-indicator normalized weighted scoring, the intelligent agent uses functional space utilization, ineffective cavity rate, single-lane cross-sectional area, minimum boundary margin, net height margin, and vertical balance as evaluation items, and also introduces new user-defined evaluation indicators. Users set the weights according to their actual design goals.

[0030] Step S7: Project delivery; The recommended cross-section scheme obtained in step S6 is input into the engineering delivery module to generate and output the corresponding engineering delivery results and extended application model. The engineering delivery results include a two-dimensional vector cross-section diagram, a three-dimensional BIM model and digital twin visualization scene, a cross-section layout parameter table, an evaluation index calculation table, constraint verification results, records of failure reasons for infeasible schemes, intelligent agent inference logs, parameter adjustment trajectories, and a recommended scheme description document.

[0031] In this step, the intelligent agent performs engineering transformation on the cross-section parameters based on the spatial parameters and component positional relationships of the recommended cross-section scheme, and automatically generates a standard tunnel cross-section model by calling a parametric modeling tool. The standard tunnel cross-section model includes tunnel lining, vehicle space, road slabs, utility tunnel space, partitions, equipment components, lighting components, signage components, and ancillary facility components. The cross-section layout parameters, evaluation indicators, constraint verification results, and deduction process records are associated with the model objects, thus forming a standard cross-section deliverable that can be used for scheme presentations, design reviews, and engineering communication.

[0032] The intelligent system receives tunnel alignment data and, based on the tunnel centerline, longitudinal slope, curve radius, mileage marker, and segment length information, arrays and assembles the standard tunnel cross-section model along the alignment direction to generate a tunnel alignment model. This tunnel alignment model is used to express the spatial continuity, component arrangement, and three-dimensional path adaptability of the recommended cross-section scheme under actual route conditions, and is used for subsequent detailed design, BIM collaboration, and digital twin platform integration.

[0033] The intelligent agent system receives tunnel operation environment design parameters and generates a tunnel operation simulation scenario based on the tunnel route model or standard cross-section model. The operation environment design parameters include lighting layout, traffic signs, equipment locations, vehicle travel paths, monitoring facilities, ventilation and fire protection facilities, maintenance access, and environmental performance parameters. By constructing a tunnel operation simulation scenario, the intelligent agent system simulates and represents the spatial perception, equipment layout, vehicle sight distance, maintenance access, and scene visualization effects of the recommended cross-section scheme during the operation phase.

[0034] Through the aforementioned intelligent agent method for optimizing tunnel spatial layout and extrapolating digital twin parameters, this invention not only completes conventional engineering delivery but also achieves continuous transformation from recommended cross-section schemes to standard tunnel cross-section models, from tunnel alignment data to tunnel alignment model assembly, and from operational environment design to tunnel operation simulation scenario construction. This forms an intelligent agent-driven extrapolation process that integrates scheme design, engineering refinement, digital twin, and operational simulation.

[0035] Furthermore, the present invention also provides a storage medium, which is a computer-readable storage medium, and stores a computer program thereon. When the computer program is executed by a processor, it implements the steps of the tunnel spatial layout optimization and digital twin parameter inference intelligent agent method as described above.

[0036] In addition, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the tunnel spatial layout optimization and digital twin parameter inference intelligent agent method as described above.

[0037] The one or more technical solutions proposed in this invention have at least the following beneficial effects: 1) Compared with traditional cross-section design methods that rely on human experience and fixed parameterized templates, this invention realizes a closed loop throughout the entire process from design requirement input, intelligent agent spatial parameter deduction, spatial layout verification and evaluation to digital twin delivery, providing an intelligent, traceable and comparable technical solution for tunnel cross-section design. 2) By introducing physical constraints, the engineering boundary conditions and deterministic verification results are integrated into the intelligent agent parameter derivation process. Based on the constraint conflicts and failure reasons, the parameters are guided to be adjusted iteratively, which improves the geometric feasibility of candidate solutions, reduces invalid trial calculations, and enhances the automation, evaluation interpretability and engineering delivery efficiency of the cross-section design of highway tunnels, urban underground roads and integrated transportation tunnels. 3) By introducing a human-computer interaction iterative mechanism, the designer's supplementary requirements, preference adjustments, and review opinions are continuously integrated into the scheme deduction. The intelligent agent assists in generating and revising the scheme, and the designer makes the final decision, so that the design process fully reflects human intentions and judgments and meets the human-centered engineering design requirements. Attached Figure Description

[0038] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the main components and interaction architecture of the intelligent agent in a specific embodiment of the present invention; Figure 3 This is a schematic diagram of the normalization and weighted scoring process for multi-index evaluation of cross-sectional layout in a specific embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the process of digital twin and its visual delivery according to a specific embodiment of the present invention; Numbering on the map: S1 - Design objective input; S2 - Preliminary scheme organization; S3 - Cross-sectional layout analysis; S4 - Multi-index evaluation and decision-making; S5 - Agent feedback iteration; S6 - Recommendation scheme selection; S7 - Digital Twin and Engineering Delivery. Detailed Implementation

[0039] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The executing entity of a specific embodiment of the present invention is a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, mobile phone, etc., or an electronic device that implements the above functions.

[0040] like Figure 1 As shown, the present invention also provides a method for optimizing tunnel spatial layout and for digital twin parameter inference intelligent agent, comprising the following steps: Step S1: Input Design Objectives; Input the tunnel cross-section design objectives and basic constraints. The tunnel cross-section design objectives include tunnel type, number of lanes, preference for single or double-layer cross-section, tunnel inner contour radius or radius range, vehicle clearance height, lane width, utility tunnel space requirements, equipment installation space requirements, lining boundary requirements, safety margin requirements, and operation and maintenance space requirements. Specifically, design requirements are input through natural language, parameter tables, or interactive interfaces. The target is "urban underground road, two-way four lanes, with a bottom integrated utility tunnel required, prioritizing space utilization and boundary safety margin". These design requirements are used as the initial input for intelligent agent inference and form structured design target information and spatial parameter information.

[0041] Step S2: Preliminary scheme organization; Based on the tunnel cross-section design objectives input in Step S1, the intelligent agent parses and semantically processes the tunnel cross-section design task, decomposes the overall design requirements into candidate parameter generation, cross-section layout analysis, evaluation index calculation, feedback iteration, recommended scheme selection and engineering delivery sub-tasks, and generates candidate cross-section schemes for single-layer cross-section, double-layer cross-section and different radii and different lane combinations. Specifically, the intelligent agent generates several candidate cross-section schemes based on tunnel type, number of lanes, utility tunnel requirements, clearance requirements, and radius range. These candidate cross-section schemes include single-layer and double-layer schemes, as well as parameter combinations under different tunnel inner contour radii, different lane combinations, different utility tunnel sizes, and different equipment space configurations. Each candidate scheme has an independent simulation status record for subsequent tool calls, result feedback, and version tracking; such as... Figure 2As shown, to support the agent in completing the above-mentioned deduction process, this invention constructs an agent system, which is based on a three-layer operating architecture consisting of a parameter information layer, a physical constraint layer, and an interaction tool layer. The parameter information layer organizes design target information, spatial parameter information, deduction status information, and final solution information; the physical constraint layer limits system prompts and deduction processes, geometric boundaries and standard hard constraints, and user preferences and indicator weight soft constraints; the interaction tool layer provides parameter analysis tools, evaluation quantification calculation tools, parameter modeling tools, and result delivery export tools. The agent uses a large language model as its reasoning kernel, continuously reading information, judging constraints, calling tools, and updating the deduction status according to the three-layer architecture.

[0042] Step S3: Cross-section and spatial layout analysis and hard constraint verification; For each candidate cross-section scheme obtained in Step S2, the agent calls the cross-section parameter analysis tool to perform spatial layout calculations, and verifies the feasibility of the calculation results based on geometric boundaries and standard hard constraints. The cross-section parameter analysis tool calculates the positional relationships of the vehicle space, structural plate, pipe gallery space, equipment installation space, arch space, and remaining space in the cross-section based on parameters such as tunnel inner contour boundary, vehicle clearance, lane width, structural plate thickness, pipe gallery dimensions, equipment installation space, and safety margin, and records the feasibility status and infeasibility reasons of the candidate schemes; For a single-layer cross-section, the cross-section parameter analysis tool calculates the upper vehicle space, lower pipe gallery space, equipment installation space, and their relative positional relationships with the tunnel circular boundary; For a double-layer cross-section, the cross-section parameter analysis tool calculates the upper vehicle space, lower vehicle space, lower pipe gallery space, middle partition, equipment installation space, and their relative positional relationships with the tunnel circular boundary; After completing the spatial layout calculation, the agent verifies the geometric feasibility of the candidate cross-section schemes based on preset hard constraint conditions. The hard constraints include vehicle clearance constraints, utility tunnel space constraints, equipment installation space constraints, lining boundary constraints, structural thickness constraints, and safety margin constraints. If a candidate cross-section scheme meets the above hard constraints, it is marked as a feasible scheme. If a candidate cross-section scheme does not meet the above hard constraints, it is marked as an infeasible scheme, and the reasons for infeasibility are recorded: insufficient vehicle clearance, insufficient utility tunnel space, insufficient equipment installation space, insufficient cross-sectional boundary margin, insufficient inter-layer spacing of the structure, or the arrangement space exceeding the tunnel's inner contour boundary.

[0043] Step S4: Multi-index evaluation and decision-making; Evaluation indices are calculated for the feasible cross-section schemes verified by the hard constraints in Step S3, forming a cross-section layout evaluation result for horizontal comparison of schemes. The evaluation indices include functional space utilization rate, ineffective cavity rate, single-lane cross-section area, minimum boundary margin, clearance margin, and vertical balance. Specifically, the functional space utilization rate characterizes the proportion of vehicle space and utility tunnel space to the tunnel's inner contour area; the ineffective cavity rate characterizes the proportion of unused remaining space within the cross-section; the single-lane cross-section area characterizes the cross-section area consumption per unit number of lanes; the minimum boundary margin characterizes the minimum distance between the boundary of each functional component in the cross-section and the inner contour of the tunnel lining; the clearance margin characterizes the minimum margin of actual vehicle clearance height relative to the design requirement clearance height; and the vertical balance characterizes the degree of balance between the remaining space at the top and bottom of the cross-section. Figure 3 As shown, the agent normalizes the above indicators and calculates the comprehensive evaluation score based on the user-defined indicator weights or the system's default weights. For positive indicators where larger values ​​indicate a better solution, forward normalization is used; for negative indicators where smaller values ​​indicate a better solution, backward normalization is used. Custom evaluation indicators are introduced based on engineering requirements: project cost, construction difficulty, ease of operation and maintenance, or traffic carrying capacity efficiency, and corresponding weights are assigned to participate in the comprehensive scoring.

[0044] Step S5: Agent Feedback Iteration; Based on the feasibility status and infeasibility reasons obtained in Step S3, and the evaluation index results obtained in Step S4, the agent identifies the constraint conflicts and index shortcomings of the candidate solutions and generates the parameter adjustment strategy for the next round. For infeasible candidate solutions, the agent determines the direction of parameter adjustment based on the infeasibility reasons: If the infeasibility reason is insufficient vehicle clearance, the agent adjusts the tunnel inner contour radius, road slab elevation, inter-layer height, or cross-sectional layout; if the infeasibility reason is insufficient utility tunnel space, the agent adjusts the utility tunnel width, utility tunnel height, utility tunnel layer, or the relative position between the vehicle space and the utility tunnel; if the infeasibility reason is insufficient equipment installation space, the agent adjusts the equipment layout area, arch space, lateral safety margin, or auxiliary component dimensions; if the infeasibility reason is insufficient lining boundary margin, the agent increases the cross-sectional radius, adjusts the number of lanes, changes the single-layer or double-layer layout, or reduces non-critical functional space.

[0045] For feasible candidate solutions that already meet the hard constraints, the agent determines the optimization direction based on the evaluation index results: when the functional space utilization rate is low or the ineffective cavity rate is high, the agent adjusts the layout ratio of the vehicle space, the utility tunnel space, and the remaining space; when the cross-sectional area of ​​a single lane is too large, the agent optimizes the lane combination, cross-sectional radius, or single / double layer layout; when the minimum boundary margin or net height margin is low, the agent increases the corresponding safety margin or adjusts the component position; when the vertical balance is low, the agent adjusts the upper arch space, the bottom utility tunnel space, and the elevation of the vehicle layer.

[0046] The agent generates new candidate cross-section parameter sets based on the above adjustment strategy, and then calls the cross-section parameter analysis tool, hard constraint verification tool, and evaluation quantification calculation tool again, forming a closed-loop iterative process of "candidate generation - cross-section analysis - hard constraint verification - indicator evaluation - feedback adjustment". Each iteration records the candidate scheme parameters, tool call results, feasibility status, failure reasons, indicator changes, parameter adjustment trajectory, and user review comments.

[0047] Step S6: Recommended Scheme Selection; Based on the final feasible candidate scheme set and its evaluation index results obtained in Step S5, the multiple candidate cross-section schemes are sorted and filtered to obtain the recommended cross-section scheme. The selection of recommended cross-section schemes is based on a specified single evaluation index, or on a comprehensive sorting based on a multi-index normalized weighted score.

[0048] In the comprehensive ranking process using a multi-indicator normalized weighted scoring, the agent uses functional space utilization, ineffective cavity rate, single-lane cross-sectional area, minimum boundary margin, net height margin, and vertical balance as the main evaluation items, and also introduces user-defined new evaluation indicators. The agent calculates the comprehensive evaluation score based on the normalized score and corresponding weight of each evaluation indicator. The higher the comprehensive evaluation score, the better the overall performance of the candidate cross-section scheme. All feasible cross-section schemes are ranked in descending order according to the comprehensive evaluation score, and the scheme with the highest score is selected as the recommended cross-section scheme.

[0049] After the ranking results are output, the recommended scheme, indicator weights, custom indicators, or engineering constraints are reviewed and corrected. The agent returns to step S5 based on user feedback to iterate again and update the recommended scheme.

[0050] Step S7: Project Delivery; Input the recommended cross-section scheme obtained in Step S6 into the project delivery module to generate and output the corresponding project delivery results and extended application model. The project delivery results include a two-dimensional vector cross-section diagram, a three-dimensional BIM model and digital twin visualization scene, a cross-section layout parameter table, an evaluation index calculation table, constraint verification results, records of reasons for failure of infeasible schemes, agent inference logs, parameter adjustment trajectory, and a recommended scheme description document.

[0051] In this step, the intelligent agent performs engineering transformation on the cross-section parameters based on the spatial parameters and component positional relationships of the recommended cross-section scheme, and automatically generates a standard tunnel cross-section model by calling a parametric modeling tool. The standard tunnel cross-section model includes tunnel lining, vehicle space, road slabs, utility tunnel space, partitions, equipment components, lighting components, signage components, and ancillary facility components. The cross-section layout parameters, evaluation indicators, constraint verification results, and deduction process records are associated with the model objects, thus forming a standard cross-section deliverable that can be used for scheme presentations, design reviews, and engineering communication.

[0052] like Figure 4 As shown, the intelligent agent system further receives tunnel alignment data and, based on the tunnel centerline, longitudinal slope, curve radius, mileage marker, and segment length information, arrays and assembles the standard tunnel cross-section model along the alignment direction to generate a tunnel alignment model. The intelligent agent system receives tunnel operation environment design parameters and generates a tunnel operation simulation scenario based on the tunnel alignment model or the standard cross-section model. Through the above methods, this invention forms an integrated intelligent agent-driven deduction process for scheme design, engineering refinement, digital twinning, and operation simulation.

[0053] The present invention also provides a computer-readable storage medium having computer-readable program instructions stored thereon, the computer-readable program instructions being used to execute the tunnel spatial layout optimization and digital twin parameter inference intelligent agent method in the above embodiments.

[0054] The computer-readable storage medium provided in this application is, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium is any tangible medium that contains or stores a program that is used by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium is transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0055] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the tunnel spatial layout optimization and digital twin parameter inference intelligent agent method as described above.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A tunnel spatial layout optimization and digital twin parameter inference intelligent agent system, characterized in that: The intelligent agent system is constructed with a three-layer operational architecture centered around the intelligent agent scheduling core: a parameter information layer, a physical constraint layer, and an interaction tool layer; wherein: The intelligent agent uses a large language model as its reasoning kernel to perform semantic parsing, task decomposition, tool scheduling, result reading, feedback iteration, and solution interpretation of the design goals. The parameter information layer is used to organize the various types of engineering information received, processed, recorded, and output during the intelligent agent's deduction process; The physical constraint layer is used to limit the system process constraints, physical boundary constraints, and user preference constraints that the intelligent agent's inference process must satisfy. The interactive tool layer is used to provide cross-sectional parameter analysis, evaluation quantification calculation, parameter modeling, and output of executable results. The intelligent agent system, through the aforementioned three-layer operating architecture, continuously reads and updates the deduction state at the parameter information layer, determines whether candidate solutions meet hard constraints and user preference constraints at the physical constraint layer, and dynamically calls engineering calculation and modeling tools at the interaction tool layer, generating parameter adjustment strategies based on the tool's returned results. This enables an interpretable, traceable, and scalable intelligent agent-driven deduction process, from design target input to candidate solution generation, cross-sectional analysis verification, indicator evaluation and recommendation, feedback iteration, and digital twin delivery.

2. The intelligent agent system for tunnel spatial layout optimization and digital twin parameter inference according to claim 1, characterized in that: The parameter information layer includes design target information, spatial parameter information, simulation status information, and final scheme information; The design target information includes tunnel type, number of lanes, preference for single or double cross-section, cross-section radius range, utility tunnel requirements, equipment layout requirements, and operational scenario requirements. The spatial parameter information includes vehicle clearance, lane width, structural slab thickness, pipe gallery width, pipe gallery height, equipment installation space, safety margin, inter-floor height, and cross-sectional layout. The simulation status information includes candidate scheme parameters, tool call results, feasibility status, failure reasons, indicator calculation results, parameter adjustment trajectory, and manual review comments; The final solution information includes recommended cross-sectional schemes, scheme ranking results, evaluation index results, two-dimensional cross-sectional diagrams, three-dimensional models, digital twin visualization scenes, and engineering delivery documents.

3. The intelligent agent system for tunnel spatial layout optimization and digital twin parameter inference according to claim 1, characterized in that: The physical constraint layer includes system process constraints, physical boundary constraints, and user preference constraints. The system process constraints are used to limit the intelligent agent to execute according to the preset task parsing, candidate generation, cross-section analysis, evaluation decision, feedback iteration, recommended solution selection and engineering delivery process; The physical boundary constraints are used to limit the candidate cross-section schemes to meet the constraints of vehicle clearance, lining boundary, pipe gallery space, equipment installation space, structural thickness, and safety margin. The user preference constraints mean that the intelligent agent inference scheme meets both hard constraints and human-centered design requirements, including priority of space utilization, traffic capacity, safety margin, vertical balance, engineering cost, operation and maintenance, as well as evaluation weights corresponding to functional space utilization rate, ineffective cavity rate, single lane cross-sectional area, minimum boundary margin, net height margin, vertical balance, and user-defined indicators. The agent performs constraint identification, conflict judgment, parameter adjustment and recommended scheme ranking on the candidate cross-section schemes according to the physical constraint layer, and writes the constraint verification results, constraint conflict reasons and soft constraint weight configuration into the inference state information.

4. The intelligent agent system for tunnel spatial layout optimization and digital twin parameter inference according to claim 1, characterized in that: The interactive tool layer includes parameter analysis tools, evaluation quantification calculation tools, parameter modeling tools, and result delivery and export tools. The parameter analysis tool is used to calculate the positional relationship of the vehicle space, pipe gallery space, structural slab, equipment space, arch space and remaining space based on the candidate cross-section parameters, and output the geometric feasibility status. The evaluation and quantitative calculation tool is used to calculate functional space utilization rate, ineffective cavity rate, single lane cross-sectional area, minimum boundary margin, net height margin, vertical balance, and user-defined indicators. The parametric modeling tool is used to generate two-dimensional vector cross-sectional views, three-dimensional BIM models, and digital twin visualization models based on the recommended scheme. The aforementioned deliverables export tool is used to output cross-sectional layout parameter tables, evaluation index calculation tables, constraint verification results, records of reasons for failure of infeasible solutions, agent inference logs, and recommended solution description documents.

5. A method for optimizing tunnel spatial layout and generating digital twin parameters using an intelligent agent, characterized in that: Includes the following steps: Step S1: Design target input; Input the tunnel cross-section design objectives and basic constraints; Step S2: Preliminary plan organization; The intelligent agent performs parsing and semantic processing of parameters based on the input tunnel cross-section design target, and generates candidate cross-section schemes for single-layer cross-sections, double-layer cross-sections, and different radii and lane combinations. Step S3: Cross-sectional and spatial layout analysis and hard constraint verification; The intelligent agent calls the cross-section parameter analysis tool to perform spatial layout calculations for each candidate cross-section scheme, and performs feasibility verification of the calculation results based on geometric boundaries and standard hard constraints; Step S4: Multi-indicator evaluation and decision making; Evaluation indicators are calculated for feasible cross-section schemes that have passed hard constraint verification, and cross-section layout evaluation results are generated for cross-sectional comparison of schemes. Step S5: Agent feedback iteration; Based on the feasibility status, infeasibility reasons, evaluation index results, user preferences, and manual review opinions of the candidate cross-section schemes, the intelligent agent generates the next round of parameter adjustment strategy, and calls the cross-section parameter analysis tool, hard constraint verification tool, and evaluation quantification calculation tool again, forming a closed-loop iterative process of candidate generation, cross-section analysis, hard constraint verification, index evaluation, and feedback adjustment. Step S6: Recommended solution selection; Multiple candidate cross-section schemes are sorted and filtered to obtain recommended cross-section schemes; after the sorting results are output, the recommended schemes, indicator weights or custom indicators are reviewed and corrected, and the agent returns to step S5 to iterate the parameters again based on the feedback. Step S7: Project delivery; Input the recommended cross-section scheme into the engineering delivery module to generate and output the corresponding engineering delivery results and extended application models.

6. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S1, the tunnel cross-section design objectives include tunnel type, number of lanes, cross-section layout, tunnel inner contour radius or radius range, vehicle clearance height, lane width, utility tunnel space requirements, equipment installation space requirements, lining boundary requirements, safety margin requirements, and operation and maintenance space requirements; the cross-section layout includes single-layer cross-section layout and double-layer cross-section layout, the single-layer cross-section layout includes upper vehicle space and lower utility tunnel space, and the double-layer cross-section layout includes upper vehicle space, lower vehicle space, and lower utility tunnel space; The intelligent agent transforms the tunnel cross-section design objectives into structured parameters, determines the range of candidate parameters based on preset rules or the reasoning results of a large language model, and generates candidate inputs for single-layer or double-layer cross-section layouts.

7. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S2, the intelligent agent breaks down the overall design requirements into candidate parameter generation, cross-section layout analysis, evaluation index calculation, feedback iteration, recommended scheme selection, and project delivery sub-tasks; the candidate cross-section schemes include combinations of schemes under different tunnel radii, different number of lanes, different cross-section layout forms, different pipe gallery sizes, and different clearance parameters. The i-th candidate cross-section scheme is represented as a design variable vector. The continuous variables include the inner contour radius of the tunnel. Elevation of each vehicular road slab Vehicle clearance height Structural plate thickness Pipe gallery width pipe gallery height Equipment installation space dimensions and safety margin; Discrete variables include the number of lanes Single-layer or double-layer layout And the component layout category, its parametric expression is as follows: （1）； In equation (1), The initial search domain is composed of the input range, the allowed interval of the standard, and the constructed set of values. The agent performs combined enumeration based on discrete variables and uses uniform sampling or adaptive sampling based on historical inference results for continuous variables to form a candidate set.

8. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S3, the positional relationships of the vehicle space, structural slab, utility tunnel space, equipment installation space, arch space, and remaining space in the cross-section are calculated, and the feasibility status and infeasibility reasons of the candidate schemes are recorded. For a single-layer cross-section, the cross-section parameter analysis tool calculates the relative positions of the upper vehicle space, the lower utility tunnel space, and the circular boundary of the tunnel. For a double-layer cross-section, the cross-section parameter analysis tool calculates the relative positions of the upper vehicle space, the lower vehicle space, the lower utility tunnel space, the intermediate partition, and the circular boundary of the tunnel. A two-dimensional rectangular coordinate system is established with the center of the tunnel cross-section as the origin. The inner contour of the tunnel adopts a radius of... The circle or equivalent boundary described by a parametric curve; for the inner contour of the circle, given the vertical coordinates Available half-width for: （2）； For the j-th functional region, the coordinates of its rectangular envelope center are: Half width is Half height is The preset safety margin is The inner contour inclusion is checked using the four corner points or discrete boundary sampling points of the component. The boundary constraints under the circular cross-section are expressed as follows: （3）； In equation (3), and These values ​​are 1 or -1, representing the four corner points of the rectangular envelope; when When, it indicates that the functional area and its safety margin do not exceed the tunnel's inner contour; when When this occurs, it indicates that at least one corner point exceeds the allowed boundary, and the candidate solution is marked as having a boundary conflict; The spacing constraints, vehicle clearance constraints, and pipe gallery dimension constraints between adjacent components j and l are expressed as follows: （4）； After completing the spatial layout calculation, the intelligent agent performs a geometric feasibility check on the candidate cross-section schemes based on preset hard constraints; the hard constraints include vehicle clearance constraints, pipe gallery space constraints, equipment installation space constraints, lining boundary constraints, structural thickness constraints, and safety margin constraints. If a candidate cross-section scheme satisfies the above hard constraints, the agent marks it as a feasible scheme and records its cross-sectional spatial parameters, component positional relationships, and hard constraint verification results. If a candidate cross-section scheme does not meet the above hard constraints, the agent marks it as an infeasible scheme and records the reason for infeasibility. The agent writes the feasibility status, the reason for infeasibility, and the hard constraint conflict information into the inference status information for scheme iteration, multi-index evaluation, and recommended scheme selection.

9. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S4, the evaluation indicators include functional space utilization rate, ineffective cavity rate, single lane cross-sectional area, minimum boundary margin, net height margin, and vertical balance. For the j-th index of the i-th solution in the feasible solution set Dimensionless normalization is performed using the range method; the positive and negative indices are calculated according to the following formulas: （5）； （6）； Weights are set based on user preferences. The comprehensive evaluation score of the candidate solutions Represented as: （7）； The functional space utilization rate is used to characterize the proportion of vehicle space and utility tunnel space to the tunnel's inner contour area. It is calculated by dividing the sum of the road space area and the utility tunnel area by the tunnel's inner contour area. The ineffective cavity ratio is used to characterize the proportion of remaining space within the cross-section that is not utilized by functional spaces such as vehicle space and utility tunnel space. It is calculated by dividing the remaining area by the outline area within the tunnel. The single-lane cross-sectional area is used to characterize the tunnel cross-sectional area consumption corresponding to a unit number of lanes. It is calculated by dividing the tunnel inner contour area by the total number of lanes. The minimum boundary margin is used to characterize the minimum safe distance between the boundary of each functional component in the cross section and the inner contour of the tunnel lining. The clearance margin is used to characterize the minimum margin of actual vehicle clearance height relative to the design clearance requirement. The vertical balance is used to characterize the degree of balance between the top and bottom remaining spaces of the cross-section, and it is obtained by normalizing the difference between the top and bottom remaining spaces.

10. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S5, the agent selects variables to be adjusted based on the constraint violation rate and the indicator weakness, automatically infers and returns new update parameters; in round t, bounded updates are used for continuous design variables: （8）； in, This is a parameter adjustment vector formed based on the main conflict constraints, the direction of indicator improvement, and historical effective adjustment records. To adjust the step size, This indicates that the updated results will be projected onto the initial search domain. ; When the preset number of iterations is reached, the number of feasible solutions that meet the requirements is obtained, or the current solution set is manually reviewed and confirmed to meet the design requirements, the agent stops feedback iteration and outputs the final feasible candidate solution set and its evaluation index results to step S6. （9）； The conditions for stopping feedback iteration include at least one of the following: (1) the number of iterations t reaches the set upper limit; (2) the improvement of the optimal comprehensive evaluation score for L consecutive iterations is less than the threshold. (3) Obtain no less than a preset number of feasible solutions; (4) Receive user confirmation instructions.

11. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S6, the selection of recommended cross-section schemes is based on a single evaluation index or a comprehensive ranking based on a multi-index normalized weighted score. In the comprehensive ranking based on a multi-index normalized weighted score, the intelligent agent uses functional space utilization rate, ineffective cavity rate, single-lane cross-sectional area, minimum boundary margin, net height margin, and vertical balance as evaluation items.

12. The tunnel spatial layout optimization and digital twin parameter inference intelligent agent method according to claim 5, characterized in that: In step S7, the engineering deliverables include two-dimensional vector cross-section diagrams, three-dimensional BIM models and digital twin visualization scenes, cross-section layout parameter tables, evaluation index calculation tables, constraint verification results, records of reasons for failure of infeasible solutions, agent inference logs, parameter adjustment trajectories, and recommended solution description documents.

13. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the tunnel spatial layout optimization and digital twin parameter inference intelligent agent method as described in any one of claims 5 to 12.

14. A computer program product, characterized in that: It includes a computer program that, when executed by a processor, implements the steps of the tunnel spatial layout optimization and digital twin parameter deduction intelligent agent method as described in any one of claims 5 to 12.

Images