Shield construction site automatic layout method based on artificial intelligence

Abstract

The invention provides a shield construction site automatic layout method based on artificial intelligence. The method specifically comprises the following steps: S1, carrying out basic data acquisition and standardization processing on a construction site; s2, constructing a fine-grained field layout rule knowledge base; s3, inputting the basic data of the construction site after standardization processing, and based on the logic guidance and constraint of the fine-grained site layout rule knowledge base, intelligently generating functional region division, facility arrangement and traffic flow organization schemes of the shield construction site by utilizing an artificial intelligence large model to fuse data understanding and logic deduction; and S4, completely converting the site layout reasoning result in the step S3 from data to a graph, generating a construction site layout scheme, and visually displaying the construction site layout scheme. According to the method, standardized data input and knowledge-driven reasoning are constructed, and an artificial intelligence large model is combined, so that automation and standardization of functional area division, facility arrangement and traffic flow organization of the shield construction site are realized, and the design efficiency and consistency are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent engineering construction, and in particular to an automated layout method and system for tunnel boring machine (TBM) construction sites based on artificial intelligence. Background Technology

[0002] Tunnel boring machine (TBM) construction, as an important method in underground engineering, is widely used in infrastructure construction such as urban subways, highways, and water pipelines. The layout design of the TBM construction site directly affects the rationality of construction organization, the efficiency of material transportation, and the safety of the operation.

[0003] The current layout of tunnel boring machine (TBM) construction sites mainly relies on manual experience, with engineers manually designing the sites based on site conditions, construction techniques, and established rules of thumb. This method has the following drawbacks:

[0004] (1) Manual design takes a long time and is subject to the experience level of the designers, which can easily lead to strong subjectivity and low standardization in the layout scheme;

[0005] (2) In complex or irregular site shapes, traditional layout methods are difficult to quickly achieve reasonable division of functional areas and efficient arrangement of facilities, which affects construction progress and resource utilization efficiency.

[0006] (3) In the layout process, the systematic consideration of optimizing the construction flow and safety distance constraints is often neglected, which can easily lead to conflicts in the flow of personnel and materials and increase construction safety risks.

[0007] In recent years, although some studies have attempted to introduce CAD-aided design, rule-based reasoning, or optimization algorithms into the site layout process, existing technologies are mostly limited to single functional area division or static layout generation. They lack system support for dynamic adjustment, adaptation to complex scenarios, and continuous optimization capabilities, and cannot meet the development needs of modern shield tunneling construction for rapid changes, refined management, and intelligent construction.

[0008] Therefore, there is an urgent need for a new type of automated layout technology for shield tunneling sites that can intelligently reason based on input data, automatically generate site layout schemes, support human-computer interaction optimization, and adapt to diverse construction scenarios. Summary of the Invention

[0009] To address the problems of existing construction site layouts relying on manual experience, low layout efficiency, poor standardization, and difficulty in coping with complex and ever-changing site requirements, this invention provides an automated layout method and system for shield tunneling construction sites based on artificial intelligence. This method combines rule-based reasoning with large-scale artificial intelligence model inference to automatically complete the construction site layout design and supports dynamic adjustment and intelligent optimization.

[0010] To achieve the above technical objectives, this invention provides an automated layout method for tunnel boring machine (TBM) construction sites based on artificial intelligence, specifically including the following steps: S1. Basic data collection and standardization processing of the construction site; acquiring multidimensional data of the construction site, and standardizing it to form a unified input vector, providing a clear and accurate input basis for rule retrieval and model reasoning;

[0011] S2. Construct a fine-grained knowledge base of site layout rules based on the standardized construction site data processed in step S1. Combine this with retrieval enhancement generation technology and knowledge graph reasoning mechanism to realize the rule retrieval, logical deduction and decision support required for automated construction site layout.

[0012] S3. Input the basic construction site data after the standardization process in step S1. Based on the logical guidance and constraints of the fine-grained site layout rule knowledge base built in step S2, use the artificial intelligence big model to integrate data understanding and logical deduction to intelligently generate the functional area division, facility layout and traffic flow organization plan of the shield tunneling construction site.

[0013] S4. After completing the site layout reasoning generation and optimization, the site layout reasoning results in step S3 are completely transformed from data to graphics through standardized layout data parsing, automatic generation of drawing scripts, one-click rendering of CAD drawings and standardized storage of output files, generating a construction site layout plan and displaying it intuitively and visually.

[0014] A further technical solution of the present invention includes:

[0015] S5. Introduce a human-machine collaboration mechanism, allowing users to interactively adjust the construction site layout plan generated in step S4, including adjusting functional area boundaries, fine-tuning facility locations and orientations, and editing and replanning circulation paths; and record all user adjustment operations in real time, extract standardized feedback data as incremental fine-tuning data, and perform the following adaptive optimization:

[0016] (1) Model fine-tuning: Locally update the inference strategy of the large model through low-rank adaptation (LoRA);

[0017] (2) Rule update: Adjust rule priority or add specific layout rules according to user preferences;

[0018] (3) Reasoning evolution: Dynamically integrate and adjust experience in the next round of layout reasoning to improve the model's adaptability and layout generation effect.

[0019] The preferred technical solution of this invention is as follows: The multidimensional data in step S1 includes boundary data of the construction site, area ratio data of functional areas, facility configuration parameters, flow lines and safety constraints, and special construction requirements; the sources of multidimensional data collection include, but are not limited to: CAD topographic map analysis, GIS data extraction, UAV aerial surveying, construction site mapping, or input by the user through an interface; the standardization processing of multidimensional data in step S1 includes standardizing the boundary data of the construction site into a closed polygon and calculating its total area; normalizing the area ratio of functional areas to ensure that the sum of the ratios is 1; standardizing the facility configuration parameters into a structural model; and performing preliminary spatial constraint processing on the separation requirements of personnel flow lines and material flow lines, as well as the safety buffer zone model.

[0020] Further technical solutions of the present invention: In step S2, the construction of a fine-grained site layout rule knowledge base includes rule knowledge modeling and classification, knowledge base structure design, retrieval and reasoning algorithm construction, dynamic update mechanism and reasoning optimization method; the rules related to the layout of the shield tunneling construction site are divided into the following five categories: functional area division rules, facility layout rules, circulation organization rules, safety distance rules and special constraint rules; each rule adopts a three-part representation of premise-constraint-output; all rules are organized in the form of a hierarchical knowledge graph, where nodes represent functional areas, facilities, and circulation elements, and edges represent the constraint relationships between elements; the construction of the retrieval and reasoning algorithm combines RAG technology and knowledge graph reasoning mechanism, including four steps: ① Condition encoding and retrieval. During the retrieval process, the cosine similarity between the input vector and the rule premise vector is calculated according to the following formula, and the set of rules with the highest matching degree is selected. :

[0021]

[0022] in: : Input condition vector; : No. Vector representation of a rule.

[0023] ② Multi-rule fusion and conflict resolution: If multiple rules simultaneously meet the search criteria, they are fused and conflict resolved according to rule priority pip_ipi, and the rule with the highest weighted return is selected.

[0024] in: Optimal rule; Weighting parameters (empirically, 0.6 to 0.7 are used); : No. Rule priority;

[0025] : No. The degree of matching between the rule and the input conditions; : A set of matching rules;

[0026] ③ Reasoning chain construction: Combining the knowledge graph, based on the currently selected rule, continue to retrieve relevant nodes to form a reasoning chain:

[0027]

[0028] ④ Layout preliminary decision output: Based on the retrieval and reasoning results, a preliminary functional area division strategy, facility layout principles and flow organization plan are formed.

[0029] A further technical solution of the present invention: Step S3 specifically includes the design of the large model architecture, input encoding method, inference generation process, layout optimization strategy, output standardization and inference results, as detailed below:

[0030] S301. The large model architecture adopts a large-scale artificial intelligence inference model based on the Transformer structure, and supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) are performed for the construction scenario.

[0031] S302. The model input data includes the standardized construction site basic data in step S1, the rule retrieval results in step S2, and historical case prompts.

[0032] The site boundaries are represented by a set of polygonal points and embedded into the model using positional encoding.

[0033]

[0034] in: : No. The location encoding value of each boundary point; , : No. The coordinates of the points;

[0035] Encoding dimension index; Total coding dimensions;

[0036] The rule-based search results are used as search prompts and embedded in the Search Enhancement Generation (RAG) module:

[0037]

[0038] in, The set of rules retrieved;

[0039] S303. After the input encoding is completed, the reasoning generation stage begins. The reasoning strategies include functional area division reasoning, facility layout reasoning, and circulation line organization reasoning.

[0040] The functional area division reasoning is based on the input boundary and functional area ratio requirements, and generates a functional area division scheme with the goal of minimizing the irregularity of the region shape while maximizing the matching of the area ratio.

[0041] Define each function area target area Generate the actual area after division. Minimize the error objective function:

[0042] in: Number of functional areas;

[0043] The facility layout reasoning, based on the functional zoning, deduces the location of each facility, considering the following multi-factor optimization objectives: minimizing the internal space occupied by the facilities, minimizing the distance from the facilities to the flow entrances / exits, and avoiding mutual obstruction or conflict between facilities; the comprehensive layout optimization objective function is:

[0044]

[0045] in: Weight coefficients for each sub-objective; : No. The area of ​​overlap between this facility and other facilities;

[0046] : No. European distance from each facility to the entrance / exit; : No. The facility and the first Individual facility conflict values; Total number of facilities.

[0047] The traffic flow organization reasoning is used to reason about material flow and personnel flow, and the flow optimality evaluation index is:

[0048]

[0049] in: Material flow line Total path length; Personnel flow Total path length; Personnel passage priority coefficient; : A set of material flow path paths; : A set of personnel flow paths;

[0050] S304. After the initial layout is generated through inference, the system performs local optimization. The optimization mechanisms include: local perturbation and simulated annealing optimization and reinforcement learning feedback optimization.

[0051] The local perturbation and simulated annealing optimization involves randomly perturbing the functional area boundary / facility location, accepting variations that reduce the objective function, and gradually decreasing the temperature parameter TTT.

[0052] The reinforcement learning feedback optimization is based on layout scoring and updates the layout generation policy through a policy gradient method; the reinforcement learning objective function is:

[0053] in: Strategy parameters The expected reward below; Layout strategy parameters;

[0054] The reward value, or score, of the layout result; Layout scheme trajectory.

[0055] S305. The optimized layout scheme is output in the following standard format for the drawing module and subsequent construction applications: functional area boundary point set (represented by Polyline), facility layout coordinates and orientation (BoundingBox or center point form), material and personnel flow line nodes and path sequences, and layout logic description document (including inference chain and rule reference information).

[0056] A further technical solution of the present invention: Step S4 specifically includes drawing data structure design, drawing element encoding method, automated drawing script generation, drawing layout optimization and standard output management, and the specific process is as follows:

[0057] S401. Standardize the format of drawing data and structure the layout elements of functional areas, facilities, and circulation routes;

[0058] S402. Code the drawing of functional area boundaries, facility layouts, and circulation lines. The closure condition of functional area boundaries satisfies the consistency of start and end points, i.e.: S

[0059] in: As the first point, End point;

[0060] The connection order satisfies:

[0061] In the above formula: : The index of the current boundary point, traversing all boundary nodes; The total number of boundary points;

[0062] : No. Two-dimensional coordinates of the boundary points; From point arrive Boundary line segments;

[0063] : No. The coordinates of the boundary points are connected in sequence.

[0064] Facilities are represented by a center point, dimensions, and orientation; flow lines are drawn as polylines, connected sequentially by nodes, with path arrows indicating flow direction; flow line path node sequence. Connection formula:

[0065]

[0066] in: Streamline The coordinates of each node; Total number of streamline nodes;

[0067] S403. An automatic drawing module is created using Python scripts, with different line types and arrow styles used for different streamlines.

[0068] S404. Drawing layout optimization includes automatic layer management, automatic scaling, automatic text layout, and margin adjustment. Margin calculation formula:

[0069] in: White margin in the drawing area; : Site length; : Site width;

[0070] S405. The final output of the plotting data is uniformly in two standard formats: SCR plotting script and DXF graphic file.

[0071] A preferred technical solution of the present invention: In step S5, the functional area boundary adjustment is achieved by dragging boundary points to modify the shape and area of ​​the functional area, and the updated area is calculated in real time. Formula for calculating area change:

[0072]

[0073] in: : No. Changes in the area of ​​functional zones; Area of ​​the edited functional area; Original functional area area;

[0074] If the change exceeds the allowable deviation range The system will prompt you to confirm or make area normalization adjustments.

[0075] in: : Adjusted The area ratio of each functional zone; : No. The area after editing each functional area; Number of functional areas;

[0076] The facility's position and orientation adjustment is performed by calculating the new center coordinates of the facility in real time after it has been dragged and moved. And verify the minimum safe distance between facilities. Formula for detecting minimum distance between facilities:

[0077]

[0078] in: , : No. , A central point of the facility; Minimum safe distance.

[0079] If the safe distance is violated, a conflict will be highlighted on the interface; when rotating the facility, the orientation angle will be updated. :

[0080]

[0081] in, This provides the user with an increment for the rotation angle.

[0082] The streamline node adjustment involves inserting, deleting, or moving nodes during streamline replanning. After each adjustment, the total path length is recalculated. :

[0083]

[0084] in: : No. The coordinates of each node;

[0085] At the same time, check the minimum safe distance between the material flow line and the personnel flow line to ensure that the flow line separation constraint is met.

[0086] A further technical solution of the present invention: Standardizing the boundary data of the construction site into a structural model includes the following steps:

[0087] ① Boundary point closure specification: ensure that the sequence of site boundary vertices is arranged in counterclockwise order and that the start and end points are consistent, thus satisfying the closure condition;

[0088] ② Calculate the area of ​​the site using the Shoelace formula to calculate the area of ​​the polygon. For use in subsequent functional area division:

[0089] in: , : No. The coordinates of the vertices; , : No. The coordinates of the vertices;

[0090] : Number of boundary vertices, cyclically connected in the order of vertices (i.e. , );

[0091] ③ Boundary concavity / convexity verification: The cross product of three consecutive points is used to determine directional consistency. The formula is as follows:

[0092]

[0093] in: : No. The result of the cross product of the vertices of the group; , : No. Vertex coordinates;

[0094] , : No. Vertex coordinates; , : No. Vertex coordinates;

[0095] If all If the signs are the same, the boundary is convex; otherwise, the Graham scan method or QuickHull algorithm needs to be used to generate the convex hull and redefine the standard boundary.

[0096] The preferred technical solution of the present invention is as follows: the standardization of the area ratio of the functional area is specifically to divide the site into multiple functional areas according to the construction requirements. The area of ​​each area is initially specified by the user, the system automatically performs the ratio normalization process, and calculates the target area of ​​each functional area.

[0097] Preliminary area calculation formulas for each functional zone:

[0098] in: Total area of ​​the entire construction site; : No. Target area for each functional zone;

[0099] Functional Area The area ratio coefficient, ,and ;

[0100] The area ratio data should meet the normalization condition. If the user-input ratio does not meet the accuracy requirements, the target area of ​​the functional area will be automatically corrected according to the following formula:

[0101] in: : The normalized first The area ratio of each functional zone;

[0102] Original number The area ratio of each functional zone; Number of functional areas.

[0103] The preferred technical solution of this invention: Standardized structural modeling of facility configuration parameters includes facility size data modeling and estimation of the number of facilities; the size data modeling formula is as follows:

[0104]

[0105] in: Area occupied by a single facility; Facility length; : Width of the facility;

[0106] Arrangement quantity The estimation formula is:

[0107] in: : No. Functional area area; Area of ​​a single facility;

[0108] Layout density coefficient (empirical value of 0.7 to 0.85, set according to site regularity and facility flexibility).

[0109] The separation requirements for personnel flow and material flow are modeled, and preliminary spatial constraint processing is performed for the safety buffer model. The modeling formula for the safety buffer is as follows:

[0110] in: Area of ​​the safety buffer zone; : Perimeter of the functional area polygon; : The minimum buffer distance specified; optimize facility layout to minimize the total distance from the facility's center point to the functional area's center of gravity.

[0111]

[0112] in: : No. The facility to the The distance between the centers of gravity of each functional area;

[0113] , : No. The coordinates of the facility; , : No. Functional area centroid coordinates;

[0114] The goal is to minimize the facility's central point. To the center of the functional area The total distance, and the layout optimization objective function:

[0115]

[0116] in For the number of facilities.

[0117] This invention first acquires multi-dimensional input data, including boundary information, functional area requirements, facility configuration parameters, flow lines, and safety constraints of the construction site, through a data acquisition module. To ensure the accuracy of subsequent reasoning and generation, all input data undergoes standardization. Based on this standardization, the invention uses a built-in rule-based knowledge base retrieval and reasoning mechanism to perform preliminary logical deduction of the site layout. Rule-based reasoning not only improves the rationality and compliance of the layout scheme but also provides strong logical guidance for large-scale model reasoning generation. Building upon rule-based reasoning, this invention utilizes a large-scale Transformer architecture artificial intelligence model, integrating data input and reasoning prompts, to intelligently generate the construction site layout scheme. During the reasoning generation process, an objective function optimization strategy is introduced to minimize area deviation, layout overlap, and traffic path length, and reinforcement learning is employed. The system employs a feedback mechanism to dynamically optimize layout strategies. The inferred layout scheme is standardized and converted into drawing data, including functional area boundary point sets, facility layout center coordinates and orientations, and circulation path node sequences. Simultaneously, layout optimization strategies are introduced during the drawing phase, such as automatic scaling, layer management, and label avoidance, further enhancing the professionalism and readability of the layout diagram. Considering the dynamic changes at the construction site, this invention introduces a human-machine collaboration mechanism, allowing users to interactively adjust the initially generated layout scheme. This continuous learning and optimization mechanism ensures that the system can continuously adapt to different construction site conditions and dynamic changes, forming an efficient, intelligent, and flexible automated layout capability for construction sites.

[0118] The beneficial effects of this invention are:

[0119] (1) To realize the intelligent and standardized design of shield tunneling site layout; This invention realizes the automation and standardization of shield tunneling site functional area division, facility layout and traffic flow organization by constructing a standardized data input process and knowledge-driven reasoning system, combined with the artificial intelligence large model reasoning generation mechanism, which greatly reduces the dependence of traditional layout design on human experience and improves design efficiency and consistency.

[0120] (2) Improve the rationality and engineering applicability of the layout scheme; This invention uses a combination of rule knowledge base reasoning and large model deep inference to comprehensively consider multiple constraints such as functional area ratio, facility placement logic, safety distance control and flow organization optimization. The generated layout scheme has excellent performance in terms of space utilization, safety and construction convenience, and truly meets the actual needs of shield tunneling construction projects.

[0121] (3) Support dynamic human-machine collaboration and layout adaptive optimization; This invention introduces a user interaction editing mechanism, which allows designers to adjust the layout scheme according to the actual situation on site, and realizes the system's self-learning and intelligent evolution capabilities through real-time recording and standardized feedback mechanisms, thereby effectively responding to changes in the construction environment, changes in demand and other unexpected situations, and improving the flexibility and adaptability of the layout system.

[0122] (4) Forming layout optimization and drawing Figure 1 The integrated and automated output system: This invention achieves seamless integration from layout reasoning to construction drawing through layout data structure standardization and automated drawing script generation technology, automatically generating CAD drawings and layout description documents that meet construction standards, greatly improving design delivery efficiency and reducing drawing error rate and secondary processing costs.

[0123] (5) Significantly improve the level of construction safety assurance and construction organization efficiency; By systematically considering factors such as the separation of material flow lines and personnel flow lines, safety buffer zone control, and facility layout safety specifications during the layout reasoning and optimization process, this invention effectively reduces cross interference and safety hazards during construction, while optimizing material handling routes and operation flow lines, and improving the overall coordination efficiency and operation smoothness of construction organization.

[0124] (6) It has the ability to continuously evolve and migrate to multiple scenarios; the system of this invention has an incremental fine-tuning mechanism and a dynamic rule update mechanism, which can continuously optimize the reasoning strategy and knowledge system according to different project characteristics, construction organization methods and user preferences, so that the layout system has good portability and scalability, and is applicable to various shield tunneling construction scenarios such as subway, highway, and integrated pipe gallery, and has broad engineering application prospects and promotion value. Attached Figure Description

[0125] Figure 1This is a schematic diagram of the framework flow of an embodiment of the present invention;

[0126] Figure 2 This is a schematic diagram of the standardized input data framework for construction sites in an embodiment of the present invention;

[0127] Figure 3 This is a schematic diagram of the rule knowledge base retrieval and reasoning process in an embodiment of the present invention;

[0128] Figure 4 This is a schematic diagram of the artificial intelligence large model inference generation process in an embodiment of the present invention;

[0129] Figure 5 This is a schematic diagram of the functional area division and area ratio adjustment in an embodiment of the present invention;

[0130] Figure 6 This is a schematic diagram of the user interaction editing and feedback extraction process in an embodiment of the present invention;

[0131] Figure 7 This is a schematic diagram of the model incremental fine-tuning and layout evolution mechanism in an embodiment of the present invention;

[0132] Figure 8 This is a schematic diagram of the knowledge base structure in an embodiment of the present invention;

[0133] Figure 9 This is a schematic diagram of dynamic rule adjustment and reasoning evolution in an embodiment of the present invention. Detailed Implementation

[0134] The present invention will be further described below with reference to the accompanying drawings and embodiments. Figures 1 to 7 All accompanying drawings are simplified versions of embodiments and are intended only to clearly and concisely illustrate the embodiments of the present invention. The technical solutions shown in the drawings below are specific solutions of embodiments of the present invention and are not intended to limit the scope of the claimed invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0135] The embodiment provides an automated layout method for tunnel boring machine (TBM) construction sites based on artificial intelligence, the workflow of which is as follows: Figure 1 As shown, five core mechanisms—standardized data collection, rule-based knowledge reasoning, intelligent generation of large models, human-computer interaction feedback, and system adaptive evolution—form a complete closed-loop process from data input to layout optimization output. The specific steps are as follows: S1. Construction site basic data collection and standardization processing; Figure 2 As shown, multidimensional data of the construction site is obtained and standardized to form a unified input vector, providing a clear and accurate input basis for rule retrieval and model reasoning;

[0136] This paper begins with accurate and efficient site data collection and standardized input. Reasonable and standardized data input is the foundation for subsequent rule-based reasoning and intelligent layout generation. This step involves key aspects such as data collection and classification, data standardization modeling, preliminary spatial constraint processing, and integrity verification. The specific process is as follows:

[0137] (1) Data collection content and classification; the data collection at the construction site covers multi-dimensional information, and is classified according to purpose and type as shown in the table below:

[0138]

[0139] Data collection sources include, but are not limited to: CAD topographic map analysis, GIS data extraction, UAV aerial surveying, construction site mapping, or user input via interface.

[0140] (2) Site boundary data standardization modeling: The construction site is represented in the form of a two-dimensional polygon boundary. Boundary standardization modeling includes the following steps:

[0141] ① Boundary point closure specification: Ensure that the sequence of site boundary vertices is arranged in counterclockwise order and that the start and end points are consistent, thus satisfying the closure condition:

[0142] in: As the first point, End point;

[0143] To distinguish it from the formula used below, this formula states that the sequence of polygon boundary points must satisfy a closure condition. That is, the last point of the boundary... It should equal the first point. This is the basic condition for determining whether a polygon is closed, and the formula is part of geometric validity verification.

[0144] ② Calculate the area of ​​the site using the Shoelace formula to calculate the area of ​​the polygon. For use in subsequent functional area division:

[0145] in: No. The coordinates of the vertices; : No. The coordinates of the vertices;

[0146] : Number of boundary vertices, cyclically connected in the order of vertices (i.e. ).

[0147] ③ Boundary convexity check: To avoid layout chaos, it is necessary to determine whether the polygon is convex. The cross product of three consecutive points is used to determine directional consistency; the formula is:

[0148]

[0149] in: : No. The result of the cross product of the vertices of the group; : Vertex coordinates;

[0150] : No. Vertex coordinates; : No. Vertex coordinates;

[0151] If all If the signs are the same, the boundary is convex; otherwise, the Graham scan method or QuickHull algorithm needs to be used to generate the convex hull and redefine the standard boundary.

[0152] (3) Standardization of functional area ratios: Based on construction needs, the site needs to be divided into multiple functional areas (such as storage area, production area, living area, etc.). The area of ​​each area is initially specified by the user; the system automatically performs ratio normalization and calculates the target area of ​​each functional area; the initial area calculation formula is as follows:

[0153] in: Total area of ​​the entire construction site; : No. Target area for each functional zone;

[0154] Functional Area Area ratio coefficient ( ,and );

[0155] The area ratio data should meet the normalization conditions. If the user-input ratio does not meet the accuracy requirements, the system will automatically correct it.

[0156]

[0157] in: : The normalized first The area ratio of each functional zone; Original number The area ratio of each functional zone; Number of functional areas.

[0158] Example table: Standardized area ratio of functional areas

[0159]

[0160] (Note: The initial ratio of the living area was 0.18, which was adjusted to a standardized ratio of 0.20 to meet the requirements for ratio normalization.)

[0161] By standardizing the area ratio, we can ensure that the functional area division meets the overall area ratio requirements.

[0162] (4) Parameterized standard modeling of facility configuration: The following parameter information needs to be input for each facility and standardized:

[0163]

[0164] Facility size parameters are used for subsequent space filling optimization; the size data is modeled as follows:

[0165]

[0166] in: Area occupied by a single facility; Facility length; : Width of the facility.

[0167] At the same time, based on the facility layout density requirements, the initial number of facilities to be arranged in the functional area. The estimation formula is:

[0168]

[0169] in: : No. Functional area area; Area of ​​a single facility;

[0170] Layout density coefficient (empirical value of 0.7 to 0.85, set according to site regularity and facility flexibility).

[0171] (5) Traffic flow line and safety distance constraint modeling: To ensure smooth and safe construction organization, this invention introduces the following preliminary spatial constraint modeling method:

[0172] ① Safety buffer zone setting; Set the buffer zone width according to the facility type and functional area boundaries. Typical rules include:

[0173] The safe distance between the storage area and the living area shall be ≥15 meters; and the operating space of ≥10 meters shall be reserved around the mixing plant.

[0174] Buffer modeling formula:

[0175]

[0176] in: Area of ​​the safety buffer zone; : Perimeter of the functional area polygon; : This is the specified minimum buffer distance.

[0177] ② Separate modeling of material and personnel flow lines, defining the material flow line set. With personnel flow The minimum distance constraint between two sets in space is:

[0178]

[0179] in: Nodes in the material flow line; Nodes in the personnel flow; :node and The shortest Euclidean distance between them; The minimum safe distance is generally set at 2 meters or more.

[0180] If space is insufficient, the traffic optimization submodule will be used to replan the flow of traffic, prioritizing safety.

[0181] ③ Facility optimal proximity modeling: In order to reduce material handling distances and optimize facility layout, the total distance from the facility center point to the center of gravity of the functional area is minimized.

[0182] in: : No. The facility to the The distance between the centers of gravity of each functional area;

[0183] : No. The coordinates of the facility; : No. Functional area centroid coordinates;

[0184] The goal is to minimize the facility's central point. To the center of the functional area The total distance.

[0185] Layout optimization objective function:

[0186] in For the number of facilities;

[0187] By handling the spatial constraints in the preliminary way, we can ensure that the subsequent automatic layout is feasible and construction-friendly.

[0188] (6) Data integrity check and unified storage: After standardization and modeling, all input data must be checked for integrity, including: site boundary closure check; area ratio normalization check; facility parameter integrity verification (name, size, quantity, functional area); and flow line and safety distance constraint satisfaction judgment.

[0189] After the check is passed, the standardized data will be stored in a unified JSON format as the data source for subsequent intelligent inference modules. An example of field organization is shown below:

[0190] The standardized stored data will serve as the basic input for subsequent intelligent reasoning and rule retrieval modules, ensuring data consistency and system stability throughout the site layout process.

[0191] S2. A fine-grained site layout rule knowledge base is constructed using the standardized construction site data from step S1. This knowledge base is then combined with retrieval-augmented generation (RAG) technology and a knowledge graph reasoning mechanism to achieve the rule retrieval, logical deduction, and decision support required for automated construction site layout. This step includes key components such as rule knowledge modeling, knowledge base structure design, retrieval and reasoning algorithm construction, dynamic update mechanisms, and reasoning optimization methods. The specific process is as follows:

[0192] (1) In order to ensure that the layout scheme complies with construction specifications, engineering practices and safety management requirements, this invention systematically sorts out and models the relevant rules for shield tunneling site layout, which are mainly divided into the following five categories:

[0193]

[0194] (2) Knowledge base structure design and storage method:

[0195] ① Rule representation: Each rule uses a consistent three-part representation of premise-constraint-output for easy retrieval and reasoning.

[0196] If (Condition) ⇒ Then (Constraint / Action)

[0197] Example rule expression:

[0198] If: Functional area type = living area, adjacent functional area = storage yard area;

[0199] Then: The distance between the two boundaries is ≥15 meters.

[0200] Furthermore, the rules use structured data storage, with the basic unit shown in the table below:

[0201]

[0202] ② Knowledge base organization structure: This invention organizes all rules in the form of a hierarchical knowledge graph. Nodes represent elements such as functional areas, facilities, and circulation lines, and edges represent the constraint relationships between elements. The specific structure is as follows: Figure 8 As shown, this organizational method supports multi-hop retrieval and complex reasoning, improving the system's reasoning flexibility and layout rationality.

[0203] (3) Rule retrieval and reasoning algorithm design: This invention combines RAG technology and knowledge graph reasoning mechanism to construct an efficient and accurate rule retrieval and reasoning method; the reasoning process consists of four steps:

[0204] ① Conditional encoding and retrieval: Based on user-inputted information such as site boundaries, functional area requirements, and facility configuration, the information is encoded into conditional vectors. During the retrieval process, the cosine similarity between the input vector and the rule premise vector is calculated, and the set of rules with the highest matching degree is selected. :

[0205]

[0206] in: : Input condition vector; : No. Vector representation of a rule;

[0207] ② Multi-rule fusion and conflict resolution: If multiple rules simultaneously meet the search criteria, the system performs fusion and conflict resolution based on rule priority pip_ipi, selecting the rule with the highest weighted benefit.

[0208]

[0209] in: Optimal rule; Weighting parameters (empirically, 0.6 to 0.7 are used); : No. Rule priority;

[0210] : No. The degree of matching between the rule and the input conditions; : A set of matching rules.

[0211] ③ Reasoning chain construction: By combining with the knowledge graph, the system can continue to search for relevant nodes based on the currently selected rule, forming a reasoning chain.

[0212]

[0213] For example, the "material flow direction planning" can be deduced from "the storage area is close to the transportation channel", and then the "setting of a one-way circulation route" can be deduced.

[0214] Depth of the inference chain Dynamic adjustment, usually with a maximum depth set. To ensure reasoning efficiency.

[0215] ④ Initial layout decision output: Based on the retrieval and reasoning results, a preliminary functional area division strategy, facility layout principles, and flow organization plan are formulated. Example of initial decision:

[0216]

[0217] (4) Dynamic update mechanism and adaptive reasoning

[0218] This invention takes into account the characteristics of variable actual construction scenarios and dynamic adjustment of requirements, and designs a dynamic update and adaptive reasoning mechanism:

[0219] New rule learning: Supports users to manually add special rules, such as priority layout logic for specific temporary roads.

[0220] Rule conflict awareness: The system can automatically detect conflicts between different rules during the layout process and provide solutions.

[0221] Adaptive inference strategy: Dynamically adjust rule retrieval weights and inference depth based on user feedback (such as layout result satisfaction ratings).

[0222] Formula Example: User Feedback Score Update weights Method:

[0223]

[0224] in: Updated weights; Original weights;

[0225] Learning rate (range 0.1 to 0.3); User feedback score.

[0226] (5) Example of retrieval and reasoning (complete process demonstration). Taking a real construction site as an example, the input information is as follows:

[0227] Site boundary: Irregular polygon; Functional area requirements: 30% for storage yard, 50% for production area, and 20% for living area;

[0228] Facility requirements: 2 mixing plants, 1 office building, and a 5,000㎡ tunnel segment storage yard;

[0229] Special requirements: Fire lanes must run north-south;

[0230] The system's retrieval and reasoning process is as follows: Ⅰ. Conditional encoding and retrieval of matching rules; Ⅱ. Selection of high-priority functional area division rules; Ⅲ. Formation of a three-section layout: storage area - production area - living area; Ⅳ. Derivation of unidirectional material flow lines and fire lane design based on flow line rules; ⤩. Output of preliminary layout decision scheme. The layout effect is as follows: Figure 5 As shown.

[0231] S3. Input the standardized construction site basic data from step S1. Based on the logical guidance and constraints of the fine-grained site layout rule knowledge base constructed in step S2, utilize an artificial intelligence large-scale model to integrate data understanding and logical deduction to intelligently generate functional area division, facility layout, and traffic flow organization schemes for the tunnel boring machine (TBM) construction site. This step includes five main components: large-scale model architecture design, input encoding method, reasoning generation process, layout optimization strategy, output standardization, and evaluation of reasoning results, as detailed below:

[0232] (1) Large-scale model architecture design; This invention adopts a large-scale artificial intelligence inference model based on the Transformer structure, and specifically performs supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) optimization for construction scenarios. The main features of the model architecture are as follows:

[0233]

[0234] Through the above design, the model of this invention not only has powerful reasoning and layout generation capabilities, but also can integrate rule knowledge in real time and dynamically adapt to changing construction needs scenarios.

[0235] (2) Input coding and scene understanding methods: The model input includes standardized site data, rule reasoning results, and historical case prompts, which need to be uniformly coded and processed.

[0236] ① Boundary and functional area inputs are vectorized. Site boundaries are represented by polygon point sets and embedded into the model using positional encoding.

[0237]

[0238] in: : No. The location encoding value of each boundary point; : No. The coordinates of the points;

[0239] Encoding dimension index; Total coding dimensions;

[0240] The area ratio of the functional area is input as a normalized floating-point array, and functional labels are embedded (e.g., Storage, Production, Living, etc.).

[0241] Rule-based reasoning prompts are integrated, with rule-based retrieval results serving as retrieval prompts, embedded in the Retrieval Enhancement Generation (RAG) module:

[0242]

[0243] in, The set of rules retrieved.

[0244] When integrating prompts to guide large-scale model reasoning, priority should be given to satisfying established engineering specifications and empirical logic.

[0245] (3) Reasoning generation and layout scheme construction: After the input encoding is completed, the reasoning generation stage begins. The present invention adopts the following reasoning strategy:

[0246] ① Functional Zone Division Reasoning: Based on the input boundaries and functional zone ratio requirements, a functional zone division scheme is generated through reasoning. The goal is to minimize the irregularity of the region shape while maximizing the matching of area ratios.

[0247] Define each function area target area Generate the actual area after division. Minimize the error objective function:

[0248]

[0249] in: Number of functional areas;

[0250] The initial partitioning was achieved using the Voronoi diagram generation method, combined with iterative optimization through area adjustment.

[0251] Facility layout reasoning involves reasoning about the location of each facility based on the functional zoning, taking into account the following multi-factor optimization objectives: minimizing the internal space occupied by the facility; minimizing the distance from the facility to the flow entrance / exit; and avoiding mutual obstruction or conflict between facilities.

[0252] The overall layout optimization objective function is:

[0253]

[0254] in: Weight coefficients for each sub-objective; : No. The area of ​​overlap between this facility and other facilities; : No. European distance from each facility to the entrance / exit; : No. The facility and the first Individual facility conflict values; Total number of facilities.

[0255] Traffic flow organization reasoning, including material flow and personnel flow, ensures smooth and safe operation, with the following constraints:

[0256] The material flow line is unidirectional and avoids backflow and crossflow.

[0257] Personnel flow lines are independent, and the minimum distance between them and material flow lines is ≥2 meters.

[0258] Shortest path optimization (SOP) combined with a traffic simulator is used to infer streamline layout. The streamline optimality evaluation index is:

[0259]

[0260] in: Material flow line Total path length; Personnel flow Total path length;

[0261] Personnel passage priority coefficient; : A set of material flow path paths;

[0262] : A set of personnel flow paths.

[0263] (4) Layout optimization and dynamic adjustment mechanism: After the initial layout is generated through reasoning, the system performs local optimization. The optimization mechanism includes:

[0264] Local perturbation and simulated annealing optimization: Randomly perturb the functional area boundary / facility location, accept the variation that makes the objective function decrease, and control the temperature parameter TTT to decrease gradually.

[0265] Reinforcement learning feedback optimization: Based on layout scores (e.g., functional area compactness, streamline efficiency, security, etc.), the layout generation strategy is updated using a policy gradient method. Reinforcement learning objective function:

[0266]

[0267] in: Strategy parameters The expected reward below; Layout strategy parameters;

[0268] : The reward value (score) of the layout result; Layout scheme trajectory.

[0269] (5) Standardized output of layout scheme. The optimized layout scheme is output in the following standard format for the drawing module and subsequent construction applications: functional area boundary point set (represented by Polyline); facility layout coordinates and orientation (BoundingBox or center point form); material and personnel flow nodes and path sequences; layout logic description document (including reasoning chain and rule reference information);

[0270] Example data structure (simplified version):

[0271]

[0272] S4. After completing the site layout reasoning generation and optimization, the site layout reasoning results from step S3 are completely transformed from data to graphics through standardized layout data parsing, automatic drawing script generation, one-click rendering of CAD drawings, and standardized storage of output files. This generates a construction site layout scheme and displays it intuitively and visually. The specific technical process of this step includes drawing data structure design, drawing element encoding methods, automated drawing script generation, drawing layout optimization, and standard output management, as detailed below:

[0273] (1) Drawing data structure design: To ensure that the layout reasoning results can be efficiently converted into renderable graphics, this invention uniformly establishes a standard format for drawing data, which structurally describes layout elements such as functional areas, facilities, and flow lines. The drawing data mainly includes the following parts:

[0274]

[0275] (2) Drawing element encoding method:

[0276] ① Functional area boundary coding: Each functional area is represented by a polygon, and the boundary point set is arranged in clockwise order to form a closed region. The formula for drawing a closed polygon is as follows:

[0277]

[0278] To distinguish it from the formula mentioned above, this formula is used to illustrate how to represent functional area boundaries using polygons (Polylines) when drawing. It is emphasized again that to draw a closed region, the start and end coordinates must be the same; this formula, identical to the previous one, further emphasizes the technical requirements at the drawing coding level; this formula is used for data structure validity checks during the implementation of the drawing algorithm.

[0279] in: Coordinates of the final boundary point; : Coordinates of the initial boundary point;

[0280] And the connection order satisfies:

[0281] In the above formula: : The index of the current boundary point, traversing all boundary nodes; The total number of boundary points;

[0282] : No. Two-dimensional coordinates of the boundary points; From point arrive Boundary line segments;

[0283] : No. The coordinates of the boundary points are connected in sequence.

[0284] Polygon attribute encoding: fill color (indicating different functional area categories); border line style (solid line / dashed line); functional area name text label.

[0285] Facility layout coding: Facilities are represented by their center point + size + orientation (Bounding Box method). The coordinates of the four vertices of the facility rectangle are also included. Based on the center point Length and width with orientation angle calculate:

[0286]

[0287] Draw the facility outline polygon and add a facility type label at the center.

[0288] Streamline drawing is coded, and the streamlines are drawn in the form of polylines, connected in the order of nodes, with path arrows drawn to indicate the direction of flow.

[0289] Streamline path node sequence Connection formula:

[0290]

[0291] in: Streamline The coordinates of each node; Total number of streamline nodes;

[0292] Flow line attributes include: material flow lines (thick solid lines with arrows); personnel flow lines (thin dashed lines with arrows).

[0293] Minimum spacing between streamlines (to prevent crossing):

[0294]

[0295] in, Generally, a height of 2 meters or more is chosen.

[0296] (3) Automatic drawing script generation method: This invention uses Python language to write an automatic drawing script generation module. The core process is as follows: ① Function area drawing script generation: Traverse each function area element and generate drawing commands:

[0297]

[0298] Set the border line style and fill color, and label the function area with its name.

[0299] The facility drawing script is generated, the coordinates of the four vertices of the facility rectangle are calculated, and the command to draw the rectangle is generated.

[0300]

[0301] Configure facility fill styles and numbering identifiers.

[0302] Streamline drawing script generation: Generates polyline paths based on node sequences and adds arrows.

[0303]

[0304] Different types of streamlines use different line types and arrow styles.

[0305] (4) Drawing layout optimization strategy: In order to improve the clarity and readability of the layout diagram, this invention designs a drawing layout optimization strategy:

[0306] Automatic layer management: Function areas, facilities, and workflows are each assigned an independent layer for easy individual editing;

[0307] Automatic scaling: Automatically sets the drawing scale according to the actual size of the site to maintain visual consistency;

[0308] Automatic text layout: Function area / facility labels dynamically adjust their position based on the center point to avoid obstruction;

[0309] Boundary margin adjustment: Leave a margin of bbb around the drawing area. A typical setting is b = 5% × b = 5% × maximum area size. Margin calculation formula:

[0310] in: White margin in the drawing area; : Site length; : Width of the site.

[0311] (5) Standardized output and storage of drawing files. The final output of drawing data is uniformly in the following two standard formats:

[0312]

[0313] All plotting output files should be named using the project number followed by the date, for example:

[0314]

[0315] At the same time, a layout specification document is generated, describing the layout reasoning logic, functional area definitions, facility parameters, and circulation organization scheme, which is convenient for construction units to understand and use.

[0316] S5. A human-machine collaboration mechanism is introduced, allowing users to interactively adjust the construction site layout scheme generated in step S4, including adjusting functional area boundaries, fine-tuning facility locations and orientations, and editing and replanning flow paths. Although the automated layout system for tunnel boring machine (TBM) construction sites proposed in this invention can generate layout schemes based on data input and intelligent reasoning, considering unforeseen factors such as dynamic adjustments to requirements, changes in construction organization, and differences in the site environment in actual engineering scenarios, relying solely on automatic reasoning results may not fully meet all practical application requirements. Therefore, this invention further designs a complete user interaction and feedback optimization mechanism to achieve human-machine collaborative optimization of the layout scheme, dynamically improving the rationality of the layout and the system's intelligent adaptability.

[0317] This step is as follows: Figure 6 As shown, this includes interactive interface design, visual layout editing, feedback data extraction, model fine-tuning and incremental learning, dynamic rule adjustment and inference self-evolution, etc., as detailed below:

[0318] (1) User interface design: To facilitate construction engineers and designers in intuitively understanding and modifying the layout scheme, this invention has developed a dedicated user interface module, whose core functions include:

[0319]

[0320] The interactive interface supports CAD plug-in deployment or web-based visual operation, and is compatible with common platforms such as AutoCAD and Revit. (2) Layout visualization editing method: Users can make detailed adjustments to the layout scheme through the interactive interface. The main editing operations include:

[0321] ① Function area boundary adjustment: Users can modify the shape and area of ​​the function area by dragging boundary points, and the system calculates the updated area in real time. Formula for calculating area change:

[0322] in: : No. Changes in the area of ​​functional zones; Area of ​​the edited functional area; Original functional area area;

[0323] If the change exceeds the allowable deviation range The system prompts the user to confirm or perform area normalization adjustments:

[0324]

[0325] in: : Adjusted The area ratio of each functional zone; : No. The area after editing each functional area; Number of functional areas.

[0326] Facility location and orientation adjustment: After a facility is dragged and moved, the new center coordinates of the facility are calculated in real time. And verify the minimum safe distance between facilities. Formula for detecting minimum distance between facilities:

[0327]

[0328] in: : No. A central point of the facility; Minimum safe distance.

[0329] If the safe distance is violated, a conflict will be highlighted on the interface. When rotating the facility, the orientation angle will be updated. :

[0330]

[0331] in, This provides the user with an increment for the rotation angle.

[0332] Streamline node adjustment: During streamline replanning, nodes are inserted, deleted, or moved. After each adjustment, the total path length is recalculated. :

[0333]

[0334] in: : No. The coordinates of each node;

[0335] At the same time, check the minimum safe distance between the material flow line and the personnel flow line to ensure that the flow line separation constraint is met.

[0336] (3) Feedback data extraction and standardization: The system automatically records and extracts standardized feedback data for all user adjustments made on the interactive interface, including:

[0337]

[0338] All extracted data is packaged in the standard JSON format and used as input for model fine-tuning and rule revision.

[0339] (4) Incremental model fine-tuning and learning update: This invention introduces a local incremental learning mechanism to dynamically fine-tune the layout inference model based on user-adjusted data, thereby improving the adaptability of subsequent inference results. Incremental fine-tuning employs a weight update strategy:

[0340] I. Collect the adjusted layout data pairs;

[0341] II. Using the original layout as input and the adjusted layout as the ideal output, a mini-supervised training set is formed;

[0342] III. Use the low-rank adaptation (LoRA) fine-tuning technique to fine-tune only a portion of the weight matrix while maintaining the stability of the original model.

[0343] Incremental fine-tuning of the objective function:

[0344]

[0345] in: : The model's current prediction output; : The target label for the layout adjusted by the user.

[0346] After training, the fine-tuned weights are merged into the main model to form adaptive evolution capability.

[0347] (5) Dynamic rule adjustment and inference evolution. In addition to model fine-tuning, the system also supports dynamically updating the rule knowledge base based on user-adjusted data:

[0348] New temporary rules: such as setting temporary safety buffer standards for specific construction sites;

[0349] Rule weight adjustment: Dynamically increase the priority of certain layout rules based on user preferences;

[0350] Conflict Detection and Optimization: Automatically detects conflicts between existing rules and user-adjusted layouts, prompting the user to confirm or update the rules. Rule Weight Update Formula:

[0351]

[0352] in: The updated version Rule priority; Original rule priority;

[0353] User-adjusted preference score; Update step size (typically 0.2);

[0354] The inference evolution diagram in the embodiment is as follows: Figure 9 As shown;

[0355] (6) Complete human-machine collaborative optimization process; This invention ultimately forms a complete closed-loop human-machine collaborative optimization process:

[0356] ① The system generates a preliminary layout based on data input and rule-based reasoning;

[0357] ② Users can visually edit layout details and make appropriate adjustments;

[0358] ③ The system records adjustment operations in real time and extracts feedback in a standardized manner;

[0359] ④ Fine-tune the model and dynamically correct rules based on feedback data;

[0360] ⑤ The next round of layout reasoning will automatically incorporate user preferences and adjustment experience.

[0361] This mechanism enables the system layout results to be continuously iterated and optimized to meet the actual needs of the construction site, greatly improving the system's intelligence level, practicality, and application breadth.

[0362] This invention proposes an automated layout method for tunnel boring machine (TBM) construction sites based on artificial intelligence, aiming to achieve intelligent, standardized, and dynamic optimization of functional area division, facility layout, and traffic flow organization at TBM construction sites. This method relies on five core mechanisms: standardized data acquisition, rule-based knowledge reasoning, intelligent generation of large-scale models, human-computer interaction feedback, and system adaptive evolution, forming a complete closed-loop process from data input to layout optimization output.

Claims

1. An automated layout method for tunnel boring machine (TBM) construction sites based on artificial intelligence, characterized in that, Specifically, the following steps are included: S1. Construction site basic data collection and standardization processing; acquire multi-dimensional data of the construction site, and after standardization processing, form a unified input vector to provide a clear and accurate input basis for rule retrieval and model reasoning; S2. Construct a fine-grained knowledge base of site layout rules based on the standardized construction site data processed in step S1. Combine this with retrieval enhancement generation technology and knowledge graph reasoning mechanism to realize the rule retrieval, logical deduction and decision support required for automated construction site layout. S3. Input the basic construction site data after the standardization process in step S1. Based on the logical guidance and constraints of the fine-grained site layout rule knowledge base built in step S2, use the artificial intelligence big model to integrate data understanding and logical deduction to intelligently generate the functional area division, facility layout and traffic flow organization plan of the shield tunneling construction site. S4. After completing the site layout reasoning generation and optimization, the site layout reasoning results in step S3 are completely transformed from data to graphics through standardized layout data parsing, automatic generation of drawing scripts, one-click rendering of CAD drawings and standardized storage of output files, generating a construction site layout plan and displaying it intuitively and visually.

2. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 1, characterized in that, The method further includes: S5. Introduce a human-machine collaboration mechanism, allowing users to interactively adjust the construction site layout plan generated in step S4, including adjusting functional area boundaries, fine-tuning facility locations and orientations, and editing and replanning circulation paths; and record all user adjustment operations in real time, extract standardized feedback data as incremental fine-tuning data, and perform the following adaptive optimization: (1) Model fine-tuning: Locally update the inference strategy of the large model through low-rank adaptation; (2) Rule update: Adjust rule priority or add specific layout rules according to user preferences; (3) Reasoning evolution: Dynamically integrate and adjust experience in the next round of layout reasoning to improve the model's adaptability and layout generation effect.

3. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 1, characterized in that: The multidimensional data in step S1 includes boundary data of the construction site, area ratio data of functional areas, facility configuration parameters, flow lines and safety constraints, and special construction requirements. Sources of multidimensional data acquisition include, but are not limited to: CAD topographic map analysis, GIS data extraction, UAV aerial surveying, construction site mapping, or user input via an interface. Standardization processing of the multidimensional data in step S1 includes: standardizing the boundary data of the construction site into a closed polygon structure and calculating its total area; normalizing the area ratios of functional areas to ensure the sum of the ratios is 1; standardizing the facility configuration parameters into a structural model; and performing preliminary spatial constraint processing on the separation requirements of personnel flow lines and material flow lines, as well as the safety buffer zone model.

4. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 1 or 2, characterized in that, Step S2 involves constructing a fine-grained site layout rule knowledge base, which includes rule knowledge modeling and classification, knowledge base structure design, retrieval and reasoning algorithm construction, dynamic update mechanism, and reasoning optimization method. The rules related to the shield tunneling site layout are divided into five categories: functional area division rules, facility layout rules, circulation organization rules, safety distance rules, and special constraint rules. Each rule is uniformly represented by a three-part structure of premise-constraint-output. All rules are organized in the form of a hierarchical knowledge graph, where nodes represent functional areas, facilities, and circulation elements, and edges represent the constraint relationships between elements. The retrieval and reasoning algorithm combines RAG technology with knowledge graph reasoning mechanisms, and includes four steps: ① Conditional encoding and retrieval. During the retrieval process, the cosine similarity between the input vector and the rule premise vector is calculated according to the following formula, and the set of rules with the highest matching degree is selected. : in: : Input condition vector; : No. Vector representation of a rule. ② Multi-rule fusion and conflict resolution: If multiple rules simultaneously meet the search criteria, they are fused and conflict resolved according to rule priority pip_ipi, and the rule with the highest weighted return is selected. in: Optimal rule; Weighting parameters, empirically taken as 0.6 to 0.7; : No. Rule priority; : No. The degree of matching between the rule and the input conditions; : A set of matching rules; ③ Reasoning chain construction: Combining the knowledge graph, based on the currently selected rule, continue to retrieve relevant nodes to form a reasoning chain: ④ Layout preliminary decision output: Based on the retrieval and reasoning results, a preliminary functional area division strategy, facility layout principles and flow organization plan are formed.

5. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 1 or 2, characterized in that, The S3 steps specifically include the design of the large model architecture, input encoding methods, inference generation process, layout optimization strategy, output standardization, and inference results, as detailed below: S301. The large model architecture adopts a large-scale artificial intelligence inference model based on the Transformer structure, and has been supervised fine-tuned and optimized by reinforcement learning for construction scenarios; S302. The model input data includes the standardized construction site basic data in step S1, the rule retrieval results in step S2, and historical case prompts. The site boundaries are represented by a set of polygonal points and embedded into the model using positional encoding. in: : No. The location encoding value of each boundary point; , : No. The coordinates of the points; Encoding dimension index; Total coding dimensions; The rule-based search results are used as search prompts and embedded in the search enhancement generation module. in, The set of rules retrieved; S303. After the input encoding is completed, the reasoning generation stage begins. The reasoning strategies include functional area division reasoning, facility layout reasoning, and circulation line organization reasoning. The functional area division reasoning is based on the input boundary and functional area ratio requirements, and generates a functional area division scheme with the goal of minimizing the irregularity of the region shape while maximizing the matching of the area ratio. Define each function area target area Generate the actual area after division. Minimize the error objective function: in: Number of functional areas; The facility layout reasoning, based on the functional zoning, deduces the location of each facility, considering the following multi-factor optimization objectives: minimizing the internal space occupied by the facilities, minimizing the distance from the facilities to the flow entrances / exits, and avoiding mutual obstruction or conflict between facilities; the comprehensive layout optimization objective function is: in: , , Weight coefficients for each sub-objective; : No. The area of ​​overlap between this facility and other facilities; : No. European-style distance from each facility to the entrance / exit; : No. The facility and the first Individual facility conflict values; Total number of facilities. The traffic flow organization reasoning is used to reason about material flow and personnel flow, and the flow optimality evaluation index is: in: Material flow line Total path length; Personnel flow Total path length; Personnel passage priority coefficient; : A set of material flow path paths; : A set of personnel flow paths; S304. After the initial layout is generated through inference, the system performs local optimization. The optimization mechanisms include: local perturbation and simulated annealing optimization and reinforcement learning feedback optimization. The local perturbation and simulated annealing optimization involves randomly perturbing the functional area boundary / facility location, accepting variations that reduce the objective function, and gradually decreasing the temperature parameter TTT. The reinforcement learning feedback optimization is based on layout scoring and updates the layout generation policy through a policy gradient method; the reinforcement learning objective function is: in: Strategy parameters The expected reward below; Layout strategy parameters; The reward value, or score, of the layout result; Layout scheme trajectory. S305. The optimized layout scheme is output in the following standard format for the drawing module and subsequent construction applications: functional area boundary point set, facility layout coordinates and orientation, material and personnel flow nodes and path sequences, and layout logic description document.

6. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 1 or 2, characterized in that: The S4 steps specifically include plotting data structure design, plotting element encoding methods, automated plotting script generation, plotting layout optimization, and standard output management. The specific process is as follows: S401. Standardize the format of drawing data and structure the layout elements of functional areas, facilities, and circulation routes; S402. Code the drawing of functional area boundaries, facility layouts, and circulation lines. The closure condition of functional area boundaries must satisfy the condition that the start and end points are consistent, i.e.: S in: As the first point, End point; The connection order satisfies: In the above formula: : The index of the current boundary point, traversing all boundary nodes; The total number of boundary points; : No. Two-dimensional coordinates of the boundary points; From point arrive Boundary line segments; : No. The coordinates of the boundary points are connected in sequence. Facilities are indicated by center point + size + orientation; Streamlines are drawn as polylines, connected sequentially by nodes, with path arrows indicating the flow direction; streamline path node sequence. Connection formula: in: Streamline Node coordinates; Total number of streamline nodes; S403. An automatic drawing module is generated by writing drawing scripts in Python, with different line types and arrow styles used for different types of streamlines; S404. Drawing layout optimization includes automatic layer management, automatic scaling, automatic text layout, and margin adjustment. Margin calculation formula: in: White margin in the drawing area; : Site length; : Site width; S405. The final output of the plotting data is uniformly in two standard formats: SCR plotting script and DXF graphic file.

7. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 2, characterized in that: In step S5, adjusting the functional area boundary involves dragging boundary points to modify the shape and area of ​​the functional area, and calculating the updated area in real time. Formula for calculating area change: in: : No. Changes in the area of ​​functional zones; Area of ​​the edited functional area; Original functional area area; If the change exceeds the allowable deviation range The system will prompt you to confirm or make area normalization adjustments. in: : Adjusted The area ratio of each functional zone; : No. The area after editing each functional area; Number of functional areas; The facility's position and orientation adjustment is performed by calculating the new center coordinates of the facility in real time after it has been dragged and moved. And verify the minimum safe distance between facilities. Formula for detecting minimum distance between facilities: in: , : No. , A central point of the facility; Minimum safe distance. If the safe distance is violated, a conflict will be highlighted on the interface; when rotating the facility, the orientation angle will be updated. : in, This is the increment of the user's rotation angle. The streamline node adjustment involves inserting, deleting, or moving nodes during streamline replanning. After each adjustment, the total path length is recalculated. : in: : No. Node coordinates; At the same time, check the minimum safe distance between the material flow line and the personnel flow line to ensure that the flow line separation constraint is met.

8. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 3, characterized in that, Standardizing the boundary data of the construction site into a structural model involves the following steps: ① Boundary point closure specification: Ensure that the sequence of site boundary vertices is arranged in counterclockwise order and that the start and end points are consistent, thus satisfying the closure condition; ② Site area calculation: Calculate the area of ​​the polygon using the Shoelace formula. For use in subsequent functional area division: in: , : No. The coordinates of the vertices; , : No. The coordinates of the vertices; : Number of boundary vertices, cyclically connected in the order of vertices (i.e. , ); ③ Boundary concavity / convexity verification: The cross product of three consecutive points is used to determine directional consistency. The formula is as follows: in: : No. The result of the cross product of the vertices of the group; , : No. Vertex coordinates; , : No. Vertex coordinates; , : No. Vertex coordinates; If all If the signs are the same, the boundary is convex; otherwise, the Graham scan method or QuickHull algorithm needs to be used to generate the convex hull and redefine the standard boundary.

9. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 3, characterized in that, The standardization of the functional area ratio specifically involves dividing the site into multiple functional areas based on construction needs. The user specifies the initial ratio for each area, and the system automatically performs ratio normalization and calculates the target area for each functional area. Preliminary area calculation formulas for each functional zone: in: Total area of ​​the entire construction site; : No. Target area for each functional zone; Functional Area The area ratio coefficient, ,and ; The area ratio data should meet the normalization condition. If the user-input ratio does not meet the accuracy requirements, the target area of ​​the functional area will be automatically corrected according to the following formula: in: : The normalized first The area ratio of each functional zone; Original number The area ratio of each functional zone; Number of functional areas.

10. The automated layout method for shield tunneling construction sites based on artificial intelligence according to claim 3, characterized in that, Standardized structural modeling of facility configuration parameters includes facility size data modeling and estimation of the number of facilities; the formula for the size data modeling is: in: Area occupied by a single facility; Facility length; : Width of the facility; Arrangement quantity The estimation formula is: in: : No. Functional area area; Area of ​​a single facility; Layout density coefficient, empirically taken as 0.7 to 0.85; Preliminary spatial constraint processing is performed on the modeling of the separation requirements between personnel flow lines and material flow lines, as well as the modeling of safety buffer zones. The modeling formula for the safety buffer zone is as follows: in: Area of ​​the safety buffer zone; : Perimeter of the functional area polygon; : This is the specified minimum buffer distance.

Images