A method for intelligently optimizing arrangement of a shield construction site

Abstract

The application provides a method for intelligently optimizing a shield construction site. The method comprises the following steps: S1, component parameter modeling, knowledge rule base establishment and component semantic atlas establishment are performed on various components of the shield construction site; S2, spatial modeling of the shield construction site and component adaptation are performed; S3, on the basis of parameter modeling and rule modeling, a composite target function integrating multiple engineering attributes is constructed, and a component arrangement scheme meeting the global optimization target, spatial constraint and engineering logic is output; S4, a man-machine collaborative arrangement interface is set, and user participation in adjustment is supported; and S5, based on the component arrangement scheme and user feedback behavior, an arrangement reasoning engine integrating historical samples and feedback learning mechanism is called, and the arrangement strategy is optimized again. The application improves arrangement efficiency and space utilization, can meet the arrangement demand of a complex irregular site, realizes multi-target collaborative optimization and conflict automatic detection, and has arrangement man-machine collaborative adjustment and strategy self-learning capabilities.

Description

Technical Field

[0001] This invention relates to the fields of civil engineering and intelligent construction technology, and in particular to a method and system for intelligent optimization of the layout of a tunnel boring machine (TBM) construction site. Background Technology

[0002] Shield tunneling has been widely applied in underground engineering projects such as urban subways, river-crossing tunnels, and municipal utility tunnels. The rational layout of the construction site plays a crucial role in ensuring construction progress, reducing resource consumption, and controlling operational risks. Generally, a shield tunneling construction site needs to be divided into multiple functional zones, such as production areas, storage areas, and living areas, with corresponding components or facilities within each zone, such as mixing plants, segment storage yards, and rebar processing sheds. The arrangement of these components affects material transportation efficiency, on-site operational safety, and space utilization efficiency.

[0003] Currently, the layout of components at tunnel boring machine (TBM) construction sites mainly relies on designers manually completing the layout based on experience in a computer-aided design (CAD) environment. Although some projects utilize parametric design or rule-based layout tools to generate initial layouts, the following technical limitations still commonly exist:

[0004] 1. The layout of components lacks standardized rules; existing methods mostly focus on the overall functional area division, and have not yet formed a systematic expression for the size adaptation, functional attributes and relative positional relationships of components, resulting in poor repeatability of layout schemes and reliance on individual experience.

[0005] 2. Limited adaptability to irregular sites; tunnel boring often occurs in urban centers or narrow sites with complex boundary shapes and diverse construction constraints. Traditional methods are difficult to achieve high-density, low-conflict arrangement of components under such conditions.

[0006] 3. Lack of intelligent optimization and conflict verification mechanisms; most existing layout tools do not have automatic optimization capabilities and cannot effectively assess the spatial conflicts, safety distances or transportation path rationality between components, making it difficult to meet the dual requirements of modern construction sites for layout quality and efficiency.

[0007] 4. Insufficient human-machine collaboration and layout evolution capabilities; In actual construction, there are dynamic needs such as changes in site conditions and the addition of temporary facilities. Existing technologies lack feedback mechanisms and automatic re-optimization functions for user adjustments, which is not conducive to building a closed-loop iterative intelligent layout system.

[0008] In summary, existing technologies for the layout of components at tunnel boring machine (TBM) construction sites still have shortcomings in terms of systematicness, intelligence, and adaptability. There is an urgent need to introduce new technologies and methods with the capabilities of rule understanding, spatial reasoning, automatic optimization, and dynamic adaptation to improve layout efficiency, standardization, and engineering practicality. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides an intelligent optimization layout method and system for shield tunneling construction sites. Based on component parameter modeling, the system integrates spatial adaptation algorithms, multi-objective optimization strategies, interactive feedback mechanisms, and artificial intelligence reasoning models to achieve intelligent, efficient, dynamic layout and continuous optimization of component-level construction facilities in complex sites. This can solve the problems of component layout relying on manual experience, low standardization, low space utilization efficiency, and delayed on-site adjustment response in current shield tunneling construction.

[0010] To achieve the above technical objectives, this invention provides an intelligent optimization layout method for tunnel boring machine (TBM) construction sites, specifically including the following steps:

[0011] S1. Component parameter modeling is performed for the physical parameters of various components required for the shield tunneling construction site. Component attributes are standardized, and the layout rules for various components required for the shield tunneling construction site are standardized. A knowledge rule base for the layout rules of various components is established. Based on the component parameter modeling and layout rules, a component semantic graph is established to provide support for subsequent layout optimization and reasoning generation. This serves as the basis for driving spatial layout logic and guiding intelligent sample generation. The knowledge rule base is a set of rules formed by component layout rules, safety constraints, and process logic abstraction.

[0012] S2. After the component parameter modeling and layout rules are established, the spatial modeling of the shield tunneling site and its adaptation to the components are carried out. Specifically, this includes four steps: standardized modeling of site boundaries, functional zoning, spatial capacity estimation and component matching and allocation. This provides clear and reasonable spatial boundaries and layout units for subsequent layout optimization steps.

[0013] S3. Based on the spatial boundary modeling, functional zoning, capacity estimation and component matching results formed in step S2, and combined with the component parameter modeling and rule modeling results in step S1, construct a composite objective function that integrates multiple engineering attributes, optimize the algorithm design and solution mechanism, and design a spatial conflict detection and dynamic repair mechanism to output a component layout scheme that meets the global optimization objective, spatial constraints and engineering logic. This scheme has local adjustability and dynamic repair capabilities.

[0014] S4. Set up a human-machine collaborative layout interface to support user participation in adjustments and extract layout behavior information from user feedback for model feedback learning. Achieve continuous evolution of intelligent optimization of component-level layout through feedback learning strategy.

[0015] S5. After step S4 is completed, based on the spatial structure model generated in step S2 and the component layout scheme output in step S3, the user feedback behavior in step S4 is integrated, and the layout inference engine that integrates historical samples and feedback learning mechanism is invoked to further optimize the layout strategy, outputting a final component layout scheme with better space utilization, transportation path efficiency and safety regularity; the layout inference engine adopts a retrieval-enhanced generation mechanism, using the component semantic graph and knowledge rule base constructed in step S1 as knowledge indexes, and drives the layout strategy to generate a language model based on the Transformer structure through embedded prompt input, realizing intelligent reasoning and continuous optimization of component layout logic, forming an adaptive layout evolution and continuous optimization mechanism, and having the ability to continuously evolve layout rules and optimization strategies through user behavior feedback.

[0016] The preferred technical solution of this invention: In step S1, the components of the shield tunneling construction site include a mixing plant, segment storage yard, rebar processing shed, temporary warehouse, office buildings, and substation; the component parameter modeling of the various components required for the shield tunneling construction site specifically involves uniformly describing each type of component as a five-tuple structure, representing the construction type, geometric dimensions, functional area, layout priority, and layout rule set; the five-tuple form of each type of component is as follows: , in: : No. Structural description of each component; Component type identifier; Component geometry; including triples: length ,width ,high ; The functional area to which the component belongs; Component placement priority, used to control the placement sequence; : A set of arrangement constraints with other components or site boundaries;

[0017] The specific process of standardizing the construction layout rules in step S1 is as follows:

[0018] The rules are standardized and modeled using the "precondition-constraint" (IF-THEN) expression structure: , in: : No. Arrangement rules; : Rule premise; : Rule constraints;

[0019] A weighting factor was also introduced into the rules. This is used to control the degree of influence of the rule on the layout optimization objective function: ;

[0020] The semantic graph construction based on the layout rules in step S1 is as follows: , in: Component layout semantic diagram; nodes : Represents all components; edge : Represents spatial constraints;

[0021] Component information should be stored in a standardized manner using a unified data structure, and JSON should be used as the underlying interaction format.

[0022] The preferred technical solution of this invention: In step S2, the site boundary standardization modeling involves standardizing the format, validating the validity, and geometrically standardizing the input two-dimensional planar boundary, specifically including:

[0023] S201, Boundary Closure Verification: The received site boundary is a sequence of point coordinates. , in: An ordered set of polygon boundary points, representing the boundary outline of the construction site; : The coordinates of the starting point of the boundary, i.e., the first vertex of the boundary; : The coordinates of the endpoint of the boundary, i.e., the last vertex of the boundary;

[0024] To ensure that the boundary is a closed polygon, the following must be satisfied: The closure condition indicates that the first and last points coincide to form a closed outline; if it is not closed, the first point is automatically added to close the outline, and the points are rearranged in a counterclockwise direction.

[0025] S202. Site area calculation: The area of ​​a standard polygon is calculated using the Shoelace formula. , in: : The calculated area of ​​the polygonal site; : No. The coordinates of the boundary points; Total number of boundary points (sequence of closed points);

[0026] S203. Concavity / convexity and legality detection; To prevent component layout from being chaotic or falling into unusable areas, the continuous three-point cross product method is used to determine whether the boundary is a convex polygon: , in: : No. Two-dimensional coordinates of the boundary points =0,1,...,n;

[0027] If all If the signs are the same, it is a convex boundary; otherwise, a Graham scan or QuickHull algorithm is executed to generate the minimum convex hull boundary for subsequent layout and trimming.

[0028] The preferred technical solution of this invention is as follows: In step S2, the functional zoning involves dividing the tunnel boring machine (TBM) construction site into a production area, a storage area, a living area, and an auxiliary area. The method for functional zoning is to use longitudinal or transverse regular cutting for regular rectangular sites; for irregular sites, a Voronoi region allocation method based on polygon segmentation and centroid guidance is used, with each functional zone division result bearing a region label. This is for component matching and judgment;

[0029] The user or system inputs the functional area proportions according to task requirements. The system is based on the total area of ​​the site. Target area for allocation of functional zones : , in: The total area ratio of all functional areas should be 1 to ensure full coverage of the site area without overlap;

[0030] To ensure the validity of the area ratio, the system performs automatic normalization: , in, : Area ratio normalization result, applicable to cases where the input ratio does not satisfy the condition that the sum is 1; Total number of functional areas, the number of types of functional areas in the system (such as production area, storage yard area, living area, auxiliary area). The sum of the proportions of all functional areas, used for normalization.

[0031] The preferred technical solution of the present invention is as follows: In step S2, the space capacity estimation is to estimate the effective layout capacity and layout density control of each functional area, and match the number of components that can be laid out based on this.

[0032] The specific calculations for site space capacity estimation and layout density control are as follows:

[0033] Set up functional areas The area is ,member The projected area is: ,

[0034] Combined with layout density coefficient Estimate the maximum capacity: , in: The floor symbol;

[0035] Arrangement density Dynamic adjustments are made based on the regularity of the site, the regularity of the components, and their rotatability.

[0036] When the number of arrangements exceeds When this happens, the system issues an early warning and suggests re-planning the priority of component layout or adjusting the area ratio.

[0037] The preferred technical solution of this invention is as follows: In step S2, the component matching and allocation strategy is to sort all components from high to low priority; each component is calculated in all functional areas. The region corresponding to the maximum value is selected as the placement target; if the placement of a component causes overcapacity, an attempt is made to allocate it to the suboptimal region; these components are marked as candidate components, and their region ratios are adjusted or inserted in subsequent iterations; the component region adaptation and preliminary allocation algorithm is as follows:

[0038] In the known list of components Functional area collection and layout rules Under the premise of this, execute the preliminary component allocation algorithm:

[0039] Matching scoring function: Component and functional areas The matching scoring function is defined as follows:

[0040] , in: Component type and functional area matching degree (binary / fuzzy); : The fit between component dimensions and area;

[0041] : Does it meet the layout rules (e.g., is it adjacent to the target area)? Factor weights (empirically set to 0.4 / 0.3 / 0.3).

[0042] The preferred technical solution of this invention is as follows: The composite objective function that integrates multiple engineering attributes in step S3 is as follows: , in: Overall layout optimization target value; : Overlap Penalty; : Total length of component transportation route (Logistics Distance); Safety Violation: Penalty for safety distance conflicts between components. : Alignment / Compactness penalty for layout irregularities; , , , : Corresponding weighting coefficients, set according to project requirements;

[0043] (1) Overlapping area term Let the set of components be... Each component is arranged in an area of ​​100 square meters. The overlap area between the components is: , in: :member With components The overlapping area; Area calculation function; Total number of components;

[0044] The overlapping area is calculated using Boolean geometry operations;

[0045] (2) Total length of transportation route The components must meet the principle of shortest path in construction logistics. The system models the distance from the component's center point to the transportation node. , in The coordinates of the logistics entry point for the component's function;

[0046] The system supports weighted paths, such as introducing weights for different components with different transportation frequencies. : , in: :member Transportation weights (frequency / importance); :member To the origin of the transportation route The Euclidean distance.

[0047] (3) Security conflict items If the component With components shortest boundary distance If a safety violation occurs, it will be recorded. , in: :member and Minimum boundary distance between them; Minimum safe distance threshold; : A Boolean conditional function that returns 1 if the condition is true and 0 otherwise;

[0048] The system can set a dynamic minimum safety distance based on the type of component;

[0049] (4) Arrangement regularity item The system introduces regularity objectives: row and column alignment difference; variance of inter-component spacing; and overall component distribution centroid offset; expressed as: , in: :member The center coordinates; Functional area to which the component belongs The geometric center of gravity is determined by the goal of placing components as close as possible to the center of the area to reduce fragmented space. Euclidean distance squared, representing the dispersion.

[0050] The preferred technical solution of this invention is as follows: the optimization strategy in step S3 adopts a heuristic hierarchical layout strategy, a nested objective function search strategy, and a layout graph search engine; the spatial conflict detection and dynamic repair mechanism in step S3 includes:

[0051] (1) Geometric overlap determination; Boolean overlap determination is performed using the component boundary polygons:

[0052]

[0053] The system highlights conflict areas and records the overlapping area and component pairs;

[0054] (2) Detection of insufficient safe distance; set up ,like: If a conflict occurs, a red alarm box will be marked, and the component will be automatically triggered to retreat or move to avoid it.

[0055] (3) Detection of traffic path obstruction; determined by simulating the intersection of transport line segments and component boundary polygons:

[0056] The system marks conflict locations on the route map, supporting the replanning of logistics channels or the automatic generation of avoidance channels.

[0057] The preferred technical solution of this invention is as follows: In step S4, a CAD platform is used as the interactive environment, and a layout control interface is embedded through a plug-in method. Combined with graphical operation and real-time feedback, the component layout is made visible, adjustable, and controllable. The extracted user feedback layout behaviors include dragging the position of various components, rotating the angle, adding components, deleting components, modifying area constraints, and viewing layout conflicts. The extracted layout behaviors are then used to generate standardized behavior vectors. Record the original state, target state, and intent label for each operation;

[0058] Based on user behavior and feedback, a low-rank adaptation parameter fine-tuning technique is used to incrementally update the deployment strategy of a large model, constructing a user preference vector. User preference vectors are injected into the intermediate layer of the large layout generation model, and the strategy weights are fine-tuned through the LoRA mechanism to achieve personalized layout generation.

[0059] After each layout optimization or user interaction, a layout snapshot version is automatically generated, supporting layout scheme comparison, viewing score changes, and one-click rollback, enhancing project transparency and decision controllability.

[0060] The preferred technical solution of this invention: In step S5, the overall architecture of the inference engine includes five layers: an input preprocessing layer, a graph neural network encoding layer, a policy generation layer, a decoder deployment layer, and an optimization scheduling layer; the inference steps are as follows:

[0061] (1) Component context encoding; Encode the type, size, rules, and functional area location information of the components to be arranged into a vector sequence. ;

[0062] (2) Vector retrieval; retrieve historical samples of similar component layouts or manually labeled rule logic from the rule knowledge base, and return a set of matching results. ;

[0063] (3) Strategy fusion generation; using a large language model, the current component state and retrieval results are used as context input to generate a candidate set of placement locations. ;

[0064] (4) Feasibility screening: The overlap, safety, path logic and other indicators of the candidate layout results are quickly evaluated by objective function, and the one with the highest score is selected.

[0065] The adaptive layout evolution and continuous optimization mechanism in step S5 includes:

[0066] The system automatically collects high-quality layout schemes confirmed by users to form a structured sample pool for evolutionary training data.

[0067] Policy replay and fine-tuning, based on the reinforcement learning framework, executes a policy replay mechanism (Replay Buffer), and updates the policy gradient by incorporating user feedback to continuously optimize the practicality of the generated results.

[0068] By summarizing and migrating regional scenarios, construction sites are clustered according to boundary shape, functional area ratio, etc., to establish a set of typical scenarios. The strategy is then fine-tuned to improve the model's cold start capability in new projects.

[0069] The beneficial effects of this invention are:

[0070] (1) Significantly improve layout efficiency and space utilization; This invention effectively realizes the automation and high-density arrangement of component-level layout in the construction site by constructing a component five-tuple model and a multi-functional area adaptation strategy, combined with a heuristic layout optimization algorithm; Compared with the traditional manual CAD layout method, the system can output a high-quality layout scheme within seconds, improving layout efficiency by more than 80% and space utilization by 15-25%.

[0071] (2) Meets the requirements of complex and irregular site layout; This invention supports the modeling and functional zoning of construction sites with concave corners, polygons and irregular boundaries. It adopts geometric algorithms such as Voronoi partitioning and minimum convex hull reconstruction, which breaks through the dependence of traditional layout systems on regular rectangular sites and realizes the layout adaptability of non-standard sites such as urban shield shafts and river-crossing tunnel entrances and exits.

[0072] (3) Achieve multi-objective collaborative optimization and automatic conflict detection; This invention designs a layout optimization objective function covering multiple dimensions such as overlap, transportation path, regularity, and safety distance. The system can automatically search for the optimal solution and embed geometric conflict detection and path occlusion recognition mechanisms to ensure that the generated scheme meets spatial constraints while taking into account construction safety and scheduling efficiency.

[0073] (4) It has the ability to arrange human-machine collaborative adjustment and strategy self-learning; the present invention integrates CAD visualization interaction plugin, allowing users to participate in component fine-tuning through operations such as "drag, rotate, add", and can record user behavior and convert it into preference vector. With the help of LoRA lightweight fine-tuning model, the arrangement strategy is adaptive, supporting the "personalized + intelligent" hybrid arrangement mode.

[0074] (5) Construct a large-scale layout model reasoning engine to realize intelligent strategy generation and continuous evolution; This invention proposes a large-scale layout model reasoning method based on component graph + knowledge retrieval enhanced generation (RAG), which can automatically generate site component layout schemes without human intervention, and optimize strategies through reinforcement learning and user acceptance feedback, and has the evolutionary ability of "the more it is used, the smarter it becomes".

[0075] (6) It has good engineering integration and cross-industry promotion value; the system of this invention supports three operating modes: local deployment, private cloud platform and AutoCAD / Revit plugin embedding. It is compatible with various engineering system interfaces such as project management, material scheduling and BIM platform. In addition to shield tunneling construction, it can also be extended to various engineering layout scenarios such as prefabricated bridge beam yard, mine temporary facilities and prefabricated component factory area, and has a wide range of engineering promotion prospects. Attached Figure Description

[0076] Figure 1 This is a flowchart of the present invention;

[0077] Figure 2 This is a flowchart of the component and functional area adaptation scoring and preliminary allocation process in the embodiment.

[0078] Figure 3 This is a logic diagram of spatial conflict detection and dynamic repair in the embodiment;

[0079] Figure 4 The following is a diagram illustrating the strategy reasoning process and RAG mechanism structure in this embodiment;

[0080] Figure 5 This is a schematic diagram of the multi-format output structure in the embodiment. Detailed Implementation

[0081] This invention discloses an intelligent optimization layout method and system for shield tunneling construction sites, aiming to solve the problems of component layout relying on manual experience, low standardization, low space utilization efficiency, and delayed on-site adjustment response in current shield tunneling construction. Based on component parameter modeling, the system integrates spatial adaptation algorithms, multi-objective optimization strategies, interactive feedback mechanisms, and artificial intelligence inference models to achieve intelligent, efficient, dynamic layout and continuous optimization of component-level construction facilities in complex sites.

[0082] The present invention provides an intelligent optimization layout method for shield tunneling construction sites, which consists of the following seven core components, forming a complete closed loop of "rule modeling—spatial matching—intelligent optimization—conflict detection—interactive feedback—reasoning generation—result output". The specific steps are as follows:

[0083] S1. Component Knowledge Modeling and Rule System Construction; In the automated layout process of shield tunneling construction sites, accurately describing the physical characteristics and layout logic of construction components is the foundation for achieving intelligent optimization. Therefore, this invention performs component parameter modeling for the physical parameters of various components required for shield tunneling construction sites, unifies component attributes, and standardizes the layout rules for various components required for shield tunneling construction sites. A knowledge rule base for the layout rules of various components is established, and a component semantic graph is built based on component parameter modeling and layout rules. This provides support for subsequent layout optimization and reasoning generation, serving as the basis for driving spatial layout logic and guiding intelligent sample generation. The knowledge rule base is a set of rules formed by component layout rules, safety constraints, and process logic abstraction. This step realizes the structured expression of component characteristics and standardized reasoning of spatial constraints, providing a data foundation and logical basis for subsequent layout generation, optimization decisions, and conflict identification. This step mainly includes four sub-tasks: component parameter modeling, layout rule modeling, semantic graph construction, and data standardization interface design.

[0084] S101. Component Parameter Modeling; The types of components at shield tunneling sites are diverse, including mixing plants, segment storage yards, rebar processing sheds, temporary warehouses, office buildings, substations, etc. The system needs to unify component attribute information to support subsequent layout determination and algorithm calls. This invention uses the following 5-tuple form to represent each type of component:

[0085] , in: : No. Structural description of each component; Component type identifier (e.g., mixing plant, stockpile, etc.); Component geometry (ternary: length) ,width ,high ); The functional area to which the component belongs (e.g., production area, living area, storage area); Component placement priority, used to control the placement sequence; : A set of arrangement constraints with other components or site boundaries.

[0086] The following section uses one component as an example to illustrate its modeling process in detail. For instance, the modeling of a mixing station is as follows: Type: Mixing station; Dimensions: 25m × 12m × 15m; Functional area: Production area; Priority: High; Layout rules: It needs to be close to the road, far from the living area, and a 10m operating space should be reserved around it. An example of its modeling is shown in Table 1.

[0087] Table 1: Examples of Typical Component Parameter Modeling

[0088]

[0089] S102. Component Layout Rule Modeling: Component layout depends not only on geometric properties but also on constraints such as safety regulations, construction techniques, and functional relationships. Therefore, this invention uses an "IF-THEN" structure to express the structure and standardize the modeling of the rules: , in, : No. Arrangement rules; : Rule prerequisites, such as "component type = mixing plant"; Rules include constraints (such as maintaining a distance of ≥15m from residential areas); weighting factors are also introduced into the rules. This is used to control the degree of influence of the rule on the layout optimization objective function: ;

[0090] Table 2 shows examples of component layout rule modeling.

[0091]

[0092] This expression method supports rule inheritance, combination, and version updates, making it easy for the system to dynamically update the layout logic and adapt to new scenario requirements.

[0093] S103. Component Semantic Graph Construction: To organize the spatial semantic relationships between components, this invention constructs a semantic graph based on layout rules. , in, Component layout semantic diagram; nodes : Represents all components; edge : Represents spatial constraints, such as adjacency, avoidance, and sequence dependency;

[0094] Each edge is defined with the following properties:

[0095] Type: such as adjacent, avoid, connect;

[0096] Intensity (weight): such as This indicates a strong preference for certain arrangements;

[0097] Function description: such as .

[0098] This graph can be used for multi-hop logic in reasoning systems, such as: "The mixing plant needs to be near the steel bar processing shed, and the steel bar processing shed needs to be near the segment stockpile ⇒ The mixing plant needs to be near the stockpile."

[0099] The component placement process can be supported by placement decisions on the graph using depth-first or graph convolutional inference algorithms.

[0100] S14. Component Data Standardization and Interface Design: In order to adapt to the requirements of algorithm scheduling and interactive calls, component information needs to be stored in a standardized manner using a unified data structure, with JSON as the underlying interactive format.

[0101] Example: Constructing a JSON data structure

[0102]

[0103] This step loads all component information into memory and constructs a layout map before component placement, realizing an integrated processing flow from "attribute understanding—rule invocation—spatial reasoning". This step constructs a knowledge modeling and rule expression system suitable for automatic component-level layout in shield tunneling construction sites, realizing standardized modeling of component structural parameters; logical and weighted expression of spatial layout rules; visual organization of multi-component layout semantic graphs and unified standardization of data structures and interfaces; providing basic data and reasoning logic support for subsequent layout optimization, reasoning generation and conflict detection modules, ensuring that the layout scheme has engineering adaptability, logical rationality and system scalability.

[0104] S2. Site Modeling and Component Adaptation Algorithm: After the component knowledge modeling and layout rule construction are completed, the spatial modeling and component adaptation of the shield tunneling construction site is a key starting point for intelligent layout optimization. This step involves spatial modeling and component adaptation of the shield tunneling construction site after the component parameter modeling and layout rule construction are completed. Specifically, it includes four steps: standardized site boundary modeling, functional zoning, spatial capacity estimation, and component matching and allocation. This provides clear and reasonable spatial boundaries and layout units for subsequent layout optimization steps. The specific steps include:

[0105] S201. Site Boundary Modeling and Geometric Standardization: Shield tunneling sites commonly face practical problems such as irregular terrain, limited space, and complex topography. Therefore, the input two-dimensional plane boundary must first be standardized in format, validated for legality, and geometrically modeled to ensure that the component layout is calculable, controllable, and optimizable.

[0106] (1) Boundary closure check: The system receives a set of point coordinate sequences for the site boundary: , in: An ordered set of polygon boundary points, representing the boundary outline of the construction site; : The coordinates of the starting point of the boundary, i.e., the first vertex of the boundary; : The coordinates of the endpoint of the boundary, i.e., the last vertex of the boundary;

[0107] To ensure that the boundary is a closed polygon, the following must be satisfied:

[0108] The closure condition indicates that the first and last points coincide to form a closed outline; if it is not closed, the first point is automatically added to close the outline, and the points are rearranged in a counterclockwise direction.

[0109] (2) Calculation of site area (Shoelace formula); The area of ​​a standard polygon is calculated using the Shoelace formula:

[0110] , in, : The calculated area of ​​the polygonal site; : No. The coordinates of the boundary points; Total number of boundary points (sequence of closed points);

[0111] This area value is used for subsequent functional area allocation and density estimation.

[0112] (3) Concavity and convexity and legality detection: To prevent the components from being arranged in a chaotic manner or falling into non-practical areas, it is necessary to determine whether the boundary is a convex polygon. The three-point cross product method is used for determination. , in: : No. Two-dimensional coordinates of the boundary points =0,1,...,n;

[0113] If all If the signs are the same, it is a convex boundary; otherwise, a Graham scan or QuickHull algorithm is executed to generate the minimum convex hull boundary for subsequent layout and trimming.

[0114] S02. Functional Zoning and Area Label Assignment; Shield tunneling sites are generally divided into the following functional areas:

[0115] Production area: mixing plant, steel bar processing shed, etc.; storage area: tunnel segment stacking and slag temporary storage.

[0116] Living area: office space and warehouse; auxiliary area: power distribution room, temporary roads, fire lanes, etc.

[0117] The user or system inputs the functional area proportions according to task requirements. The system is based on the total area of ​​the site. Target area for allocation of functional zones : , in: The total area ratio of all functional areas should be 1 to ensure full coverage of the site area without overlap;

[0118] To ensure the validity of the area ratio, the system performs automatic normalization: , in, : Area ratio normalization result, applicable to cases where the input ratio does not satisfy the condition that the sum is 1; Total number of functional areas, the number of types of functional areas in the system (such as production area, storage yard area, living area, auxiliary area). The sum of the proportions of all functional areas, used for normalization.

[0119] The zoning method employs two approaches: for regular rectangular sites, longitudinal / transverse regular division is used; for irregular sites, a Voronoi zoning method based on polygon segmentation and centroid guidance is used. Each functional zone division result is labeled with a zone label. This is for component matching and judgment.

[0120] S203. Site space capacity estimation and layout density control: In order to avoid overly dense layout or wasted space, the system needs to estimate the effective layout capacity of each functional area and match the number of components that can be laid out based on this.

[0121] Set up functional areas The area is ,member The projected area is: ;

[0122] Combined with layout density coefficient Estimate the maximum capacity: ; in: : Round down symbol; Layout density The adjustments can be made dynamically based on the regularity of the site, the regularity of the components, and their rotatability, as shown in the table below:

[0123] Table 3: Recommendations for Selecting Empirical Parameters for Layout Density

[0124]

[0125] When the number of arrangements exceeds When this happens, the system issues an early warning and suggests re-planning the priority of component layout or adjusting the area ratio.

[0126] S204. Component Region Adaptation and Preliminary Allocation Algorithm: Based on a known component list... Functional area collection and layout rules Under the premise of this, execute the preliminary component allocation algorithm:

[0127] (1) Matching scoring function, component and functional areas The matching scoring function is defined as follows:

[0128] in: Component type and functional area matching degree (binary / fuzzy); : The fit between component dimensions and area; : Does it meet the layout rules (e.g., is it adjacent to the target area)? Factor weights (empirically set to 0.4 / 0.3 / 0.3).

[0129] (2) Allocation strategy: All components are sorted from highest to lowest priority; each component is calculated in all functional areas. Select the area corresponding to the maximum value as the deployment target;

[0130] If placing a component causes overcapacity, attempt to allocate it to the suboptimal region; mark it as a candidate component, and insert or adjust the region ratio in subsequent iterations.

[0131] (3) Output structure: The preliminary component allocation results are output in a structured manner as follows:

[0132]

[0133] This result will serve as the initial state input for the layout optimization module, which will be used for further search space generation and path optimization.

[0134] This step establishes a standardized spatial modeling and component adaptation method for complex construction sites, achieving geometric standardization and boundary closure modeling for construction sites of arbitrary shapes; multi-functional area division and labeling management; component quantity estimation and density control mechanisms; and a component-area allocation strategy driven by multi-constraint matching. This step provides a clear, logically rigorous, and spatially feasible initial state for subsequent layout optimization algorithms, effectively improving the stability, universality, and intelligence of the layout system.

[0135] S3. Multi-objective layout optimization and conflict detection: After completing component rule modeling and preliminary area allocation in step S2, further issues such as layout rationality, spatial feasibility, and construction process logic still need to be addressed. This invention designs a multi-objective layout optimization algorithm integrating "safety, efficiency, and spatial coordination." Based on the spatial boundary modeling, functional zoning, capacity estimation, and component matching results formed in step S2, and combined with the component parameter modeling and rule modeling results from step S1, a composite objective function integrating multiple engineering attributes is constructed. The algorithm design and solution mechanism are optimized, and a spatial conflict detection and dynamic repair mechanism is designed to output a component layout scheme that satisfies the global optimization objective, spatial constraints, and engineering logic. This scheme has local adjustability and dynamic repair capabilities. The core of this step includes the construction of the optimization objective function, the expression of spatial constraints, the algorithm solution mechanism, the conflict detection method, and the fault tolerance strategy.

[0136] S301. Construction of Multi-Objective Optimization Function; Component placement must not only be "suitable for placement," but also "suitable for placement." Therefore, this embodiment defines a composite objective function that integrates multiple engineering attributes: , in: Overall layout optimization target value; : Overlap Penalty; : Total length of component transportation route (Logistics Distance); Safety Violation: Penalty for safety distance conflicts between components. : Alignment / Compactness penalty for layout irregularities; , , , : Corresponding weight coefficients, set according to project requirements (suggested values ​​are shown in the table below);

[0137] Recommendations for setting objective function weight parameters (example)

[0138]

[0139] (1) Overlapping area term Let the set of components be... Each component is arranged in an area of ​​100 square meters. The overlap area between the components is: , in: :member With components The overlapping area; Area calculation function; Total number of components;

[0140] The overlapping area is calculated using Boolean geometry to avoid structural installation conflicts caused by overlapping components.

[0141] (2) Total length of transportation route The components must meet the principle of shortest path in construction logistics. The system models the components based on the Euclidean distance from the center point to the transportation nodes (such as the tunnel boring machine shaft opening and entrance / exit). , in, The coordinates of the logistics entry point for the component's function.

[0142] The system supports weighted paths, such as introducing weights for different components with different transportation frequencies. : , in, :member Transportation weights (frequency / importance); :member To the origin of the transportation route The Euclidean distance.

[0143] (3) Security conflict items If the component With components shortest boundary distance If a safety violation occurs, it will be recorded. , in: :member and Minimum boundary distance between them; Minimum safe distance threshold; Boolean conditional function (1 if the condition is true, 0 otherwise);

[0144] The system can set a dynamic minimum safety distance based on the type of component (e.g., ≥10m for mixing plants, ≥20m for office areas).

[0145] (4) Arrangement regularity item To improve the aesthetics of component distribution and site compactness, the system introduces a regularity target:

[0146] Poor row and column alignment; variance in spacing between components; and shift in the center of gravity of the overall component distribution.

[0147] The typical form of expression is: , in: :member The center coordinates; Functional area to which the component belongs The geometric center of gravity is determined by the goal of placing components as close as possible to the center of the area to reduce fragmented space. Euclidean distance squared (representing dispersion);

[0148] S302. Optimization Algorithm Design and Solution Mechanism: The above objective function constitutes a high-dimensional combinatorial optimization problem. Considering conditions such as variable component size (e.g., rotation), mobility, and trimming, the solution space is enormous. The embodiment adopts the following optimization strategy:

[0149] (1) Heuristic Two-Stage Layout Strategy:

[0150] Phase 1: Global initialization. Components are sorted by priority and filled into the functional area using a greedy algorithm or tiling method to avoid initial conflicts.

[0151] Phase 2: Local perturbation optimization. Based on the current layout solution, simulated annealing is performed through "move, rotate, and exchange" operations to search for local optima.

[0152] (2) Nested search strategy for objective functions: Different objective function sub-items are nested and executed according to priority: hard constraints are solved first. Then gradually guide soft targets Avoid getting trapped in early suboptimal solutions.

[0153] (3) Layout diagram search engine: Build layout diagrams based on component layout status. Each node represents a layout scheme, and the edges represent transformation operations. Path search is performed using A* search combined with a heuristic evaluation function.

[0154] S303. Spatial conflict detection and dynamic repair mechanism. The optimized layout scheme needs to undergo systematic conflict detection, including: (1) Geometric overlap determination, Boolean overlap determination (PolygonIntersection) is performed through the component boundary polygons:

[0155] ;

[0156] The system highlights conflict areas and records the overlapping area and component pairs.

[0157] (2) Detection of insufficient safe distance, set ,like: If a conflict is detected, a red alarm box will be displayed, which can automatically trigger the component to retreat or move to avoid it.

[0158] (3) Path obstruction detection is determined by simulating the intersection of transport line segments and component boundary polygons:

[0159] ;

[0160] The system can mark conflict locations on the route map, supporting the replanning of logistics channels or the automatic generation of "avoidance channels".

[0161] S304. Conflict tolerance and user interaction mechanism: The system allows minor violations (such as a safety spacing error of ±10%) but requires user confirmation; supports "locking components" to prevent automatic adjustment by the optimizer; provides a version comparison function, allowing users to compare the component layout status before and after optimization.

[0162] This step establishes a complete multi-objective optimization framework for component placement and a spatial conflict detection mechanism, enabling multi-dimensional quantitative modeling of indicators during component placement; a combined heuristic algorithm solution strategy; automatic conflict detection and path blocking identification; and a user-friendly fault-tolerance and iterative mechanism. This provides high-quality, implementable component placement results for subsequent layout diagram output and intelligent evolution.

[0163] S4. Human-Computer Interaction and Feedback Learning Mechanism; In the process of complex shield tunneling site layout, relying solely on algorithms to automatically complete component layout often fails to meet the diverse requirements of the engineering site for safety, construction convenience, and personalization. Therefore, this embodiment designs a human-computer collaborative layout interaction and feedback optimization mechanism, sets up a human-computer collaborative layout interface, supports user participation in adjustments, and extracts layout behavior information from user feedback for model feedback learning. Through a feedback learning strategy, the continuous evolution of intelligent optimization of component-level layout is achieved. This step includes the following core components: interactive layout interface design, user operation monitoring mechanism, fine-tuning feedback update algorithm, layout preference modeling and learning, and layout version management and comparison system.

[0164] S401. Interactive Layout Interface Design: The embodiment uses a CAD platform (such as AutoCAD, Revit) as the interactive environment and embeds the layout control interface through a plug-in method. Combined with graphical operation and real-time feedback, the component layout is made visible, adjustable and controllable. Users can perform the following operations on various components in the interface: (1) Position drag: After clicking on the component, it can be translated to any area in the two-dimensional plane; (2) Angle rotation: Double-clicking the component can call the rotation tool to set or accurately input the direction angle (0°~360°); (3) Component addition / deletion: Manually add specific types of components to the layout drawing or delete the selected components through the component library interface; (4) Area constraint modification: Clicking on the functional area boundary can adjust the area scale or redefine the function label in real time; (5) Layout conflict viewing: Automatically highlight the components with conflicts in the current layout and provide two processing paths: "automatic repair" and "manual adjustment". All interactions are linked with the system backend optimization engine in real time to ensure that the interactive behavior is traceable, quantifiable and learnable.

[0165] S402. User Operation Monitoring and Event-Driven Mechanism: The system has a built-in interactive behavior listener that records all user actions and extracts key information for model feedback and learning. The listening data structure is as follows:

[0166]

[0167] It supports capturing the following five types of operation events:

[0168]

[0169] Each event operation is bound to a timestamp, operator ID, component ID, and information on changes in state before and after the event. Users can also add explanations for the reasons (such as "avoiding the storage yard" or "approaching the transport port"), providing supervisory annotations for the system's learning.

[0170] S403. Fine-tuning Feedback and Model Update Mechanism: After collecting user deployment behavior, the system uses a lightweight parameter fine-tuning algorithm to update the deployment inference module. The main process is as follows:

[0171] (1) Behavior vector encoding; the system converts user operations into structured behavior vectors. The format is:

[0172] in, : No. A vector of operational behaviors; Operation type (e.g., move, rotate); Component type; Displacement; : Amount of change in angle; Function area changes; : Is it in a state of conflict?

[0173] (2) Preference vector aggregation: Aggregate and cluster the historical behaviors of the same user or multiple users, and extract and arrange the principal components of preferences. This represents the arrangement feature dimension of their preference.

[0174] (3) LoRA parameter injection: injecting user preference vectors The intermediate layer of the injection layout generation model (such as a component layout strategy network based on Transformer) is used to fine-tune the strategy weights through the LoRA (Low-Rank Adaptation) mechanism to achieve personalized layout generation.

[0175] (4) Adjust the layout optimization weights. The system updates the weight parameters in the optimizer's objective function to better match the user's layout style. For example: if the user prefers a compact layout, increase δ (regularity weight);

[0176] User preference for safe distance redundancy ⇒ Increase γ (safety weight);

[0177] Users frequently move components near the logistics path manually ⇒ Adjust path sensitivity.

[0178] S4O4. User Layout Preference Modeling and Personalization Evolution; The system supports the creation of layout preference profiles for different users, project teams, or construction units, defined as: , in, :user A profile of layout preferences; Commonly used component combination templates; : Set of preference rules (e.g., "the mixing plant is always in the top left corner", "the office area is a separate zone"); : Objective function preference weight vector.

[0179] This profile can be used for: transferring and reusing layout styles across multiple projects; quickly generating "experience-based layout templates" in the early stages of a project; and detecting layout style anomalies (such as team members' styles deviating significantly from the project's design).

[0180] S405. Layout version management and interactive comparison mechanism; To support the controllability and traceability of layout schemes during multiple rounds of adjustments, the system provides complete layout version records, difference comparison, and backtracking recovery functions. After each layout optimization or user interaction, the system automatically generates a layout snapshot version: layout component list and location; optimization objective function score ( Value); number and type of conflicts; user notes or selected strategy;

[0181] Users can view the differences between any two versions at any time in the "Layout History Panel", including: component displacement heatmap; changes in optimization indicators; changes in conflict points; and a visual comparison of overall compactness and distribution pattern.

[0182] It also supports one-click "revert to a specified version" to ensure rapid response to on-site construction changes and traceability of layout quality.

[0183] This step designs a highly engineering-adaptable, self-learning, and evolving component placement human-computer interaction mechanism with the following significant advantages: providing an intuitive two-dimensional interactive interface that supports free adjustment of the placement; enabling structured monitoring of user behavior and injection of deep learning parameters; supporting user preference modeling to generate personalized placement styles; and providing placement version tracking and visual comparison to enhance the controllability of placement management. The collaborative optimization paradigm of "algorithm decision-making + user experience" built in this step lays the foundation and is a key supporting module for the intelligent placement of components to move from "usable" to "adjustable, learnable, and evolvable".

[0184] S5. Intelligent Evolutionary Reasoning Mechanism and Output Structure; As component knowledge modeling, human-computer interaction optimization, and feedback learning mechanisms are gradually improved, the system gradually accumulates a large amount of structured layout data and layout behavior samples. To further enhance the system's adaptability and intelligence, after step S4, this embodiment, based on the spatial structure model generated in step S2 and the component layout scheme output in step S3, integrates user feedback behavior from step S4, calls the layout reasoning engine that integrates historical samples and the feedback learning mechanism, and further optimizes the layout strategy, outputting a final component layout scheme with better space utilization, transportation path efficiency, and safety regularity; The layout reasoning engine adopts a retrieval-enhanced generation mechanism, using the component semantic graph and knowledge rule base constructed in step S1 as knowledge indexes, and drives the layout strategy generation based on the Transformer structure language model through embedded prompt input, realizing intelligent reasoning and continuous optimization of component layout logic, forming an adaptive layout evolution and continuous optimization mechanism, and having the ability to continuously evolve layout rules and optimization strategies through user behavior feedback. This step includes the following: deploying the inference engine architecture, knowledge enhancement and generation mechanisms, deploying evolutionary learning methods, implementing adaptive deployment capabilities, designing output standards, and encapsulating engineering application interfaces; specifically as follows:

[0185] S501. Overall Architecture of the Inference Engine: The inference engine architecture consists of the following five layers: (1) Input Preprocessing Layer: Receives input data such as site boundaries, functional zoning, component library, and constraint rules, and transforms them into standardized structures; (2) Graph Neural Network Encoding Layer: Embeds component semantic graphs and rule graphs as nodes to capture multi-dimensional spatial relationships; (3) Strategy Generation Layer (Core of Large Model): Generates component layout order and relative position decisions based on Transformer or GAT structure; (4) Layout Decoder Layer: Maps the generated strategy to two-dimensional coordinates, orientation, and component rotation angle; (5) Optimization Scheduling Layer: Links the objective function and user preference module to dynamically adjust the layout order and spatial position, and outputs the final solution. This architecture supports asynchronous loading, batch generation, and user parameter injection, which facilitates parallel inference and rapid response in large-scale construction scenarios.

[0186] S502. Knowledge-Enhanced Layout Generation Strategy: Considering the complexity and variability of engineering scenarios, the diversity of components, and the numerous rules, this system introduces a Retrieval-Augmented Generation (RAG) mechanism, combining a knowledge base and a language model to generate layout strategies. The reasoning steps are as follows:

[0187] Step 1: Component Context Encoding; Encode the type, size, rules, functional area location, and other information of the components to be arranged into a vector sequence. .

[0188] Step 2: Vector retrieval; Search the knowledge base for historical samples of similar component layouts or manually labeled rule logic, and return a set of matching results. .

[0189] Step 3: Strategy Fusion Generation; Using large language models (such as DeepSeek, LLaMA, etc.) as contextual input, the current component state and retrieval results are used to generate a candidate set of placement locations. .

[0190] Step 4: Feasibility screening; quickly evaluate the overlap, safety, path logic and other indicators of the candidate layout results through the objective function, and select the one with the highest score.

[0191] This process organically combines semantic understanding, data memory, and physical spatial arrangement, significantly improving reasoning accuracy and generalization ability.

[0192] S503. Adaptive Layout Evolution and Continuous Optimization Mechanism: This system supports continuous layout optimization capabilities, that is, it continuously evolves the layout generation strategy through historical data, user preferences and site characteristics, and builds an inference engine that becomes "smarter with use" in engineering applications.

[0193] The main mechanisms include the following: (1) Construction of evolution training data pool: The system automatically collects high-quality layout schemes after user confirmation to form a structured sample pool, including component-region matching; conflict avoidance strategy; user adjustment path; and optimized score evolution sequence.

[0194] (2) Policy replay and fine-tuning: Based on the reinforcement learning framework, the policy replay mechanism (Replay Buffer) is executed, and the policy gradient is updated in combination with user correction feedback to continuously optimize the practicality of the generated results.

[0195] (3) Regional scene summarization and migration: The construction site is clustered according to the boundary shape, functional area ratio, etc., to establish a typical scene set, and the strategy is transferred and fine-tuned to improve the cold start capability of the model in new projects.

[0196] S504. Output Structure Design and Standard Format Support: The system supports various output formats to adapt to the needs of design, construction, operation and maintenance stages. Output structures include:

[0197] (1) Output of layout drawings (CAD drawings): Supports *.dxf / *.dwg formats; each component is expressed in the form of solid elements, including name, size, and layer classification; supports automatic generation of drawing frames, annotations and drawing scale control.

[0198] (2) Component parameter list (table / JSON format):

[0199] Each component outputs the following parameter fields:

[0200]

[0201] (3) Layout Analysis Report (PDF / HTML format): The system automatically generates a layout evaluation report containing the following content: overall layout diagram and zoning diagram; component statistics table (quantity, distribution, deviation); conflict detection results summary; layout optimization objective function score comparison; user adjustment behavior records and suggestions.

[0202] (4) BIM platform interface format: Supports IFC (Industry Foundation Classes) interface output to achieve interconnection with BIM platforms (such as Revit and Navisworks), which facilitates subsequent three-dimensional construction simulation and site collaborative management.

[0203] S505. Engineering Deployment and Platform Integration Capabilities; To ensure system deployment flexibility and application efficiency, this invention supports the following deployment architectures:

[0204]

[0205] It also provides a RESTful API interface, which supports interface integration with existing construction management and control platforms, material scheduling systems, and project progress systems, enabling intelligent support for the entire process of shield tunneling site design, scheduling, and management.

[0206] The above steps construct an intelligent reasoning generation and continuous evolution mechanism for the layout of shield tunneling components, which has the following advantages: combining large model generation capabilities with knowledge enhancement, it has good generalization and interpretability; it supports continuous strategy evolution and user preference injection, achieving "the more it is used, the more accurate it becomes"; it outputs results in multiple formats, covering drawings, tables, reports and 3D interfaces; it is flexible in deployment and has open interfaces, making it easy to integrate into engineering projects and implement in various scenarios.

[0207] Thus, the implementation plan completes a full closed loop from "rule-driven" to "intelligent generation" and then to "evolutionary optimization," constructing an intelligent layout platform for shield tunneling construction sites with self-learning capabilities, enhanced interaction capabilities, and high-precision layout effects.

[0208] To ensure that the intelligent optimization layout method for shield tunneling construction sites proposed in this invention has good engineering adaptability, deployability, and long-term operation capability, an intelligent optimization system for shield tunneling construction sites is set up to support the method. The system is designed in accordance with the three principles of modularity, service orientation, and platform compatibility, supports multiple deployment methods and platform integration strategies, and meets the needs of users in different industries.

[0209] Firstly, the intelligent optimization system for shield tunneling construction sites adopts a multi-level deployment architecture of "local deployment + private cloud platform + plug-in embedded integration". The overall deployment architecture includes (1) local deployment mode: suitable for units with high data security requirements and stable deployment environment (such as design institutes and general contracting departments); the model engine, optimizer, user interface and CAD plug-in all run on the same workstation or server; it supports local GPU inference and layout fast feedback, with fast response speed and layout generation delay <1s.

[0210] (2) Private cloud deployment mode: suitable for collaborative management and control needs of multiple projects and multiple users; build a centralized model service platform, and realize large model scheduling and load balancing through Docker / K8s container orchestration system; each project is configured with a computing resource pool in units of "site engineering tasks", supporting parallel deployment, centralized training and cross-project strategy migration.

[0211] (3) AutoCAD / Revit plug-in integration mode: The system is embedded into the mainstream drawing platform in the form of a plug-in; users can directly call the layout engine in the original design drawings to optimize the layout of components and make intelligent adjustments; the plug-in synchronously calls the backend capabilities such as optimization objective function, rule detection, and feedback learning.

[0212] The system supports communication with the following platforms via RESTful API, WebSocket, MQTT, and other communication protocols:

[0213]

[0214] Through the above interface integration, the system is no longer just a layout design tool, but is embedded in the entire construction information flow to achieve full-chain digital collaboration.

[0215] The system deployment process strictly adheres to engineering information security standards, including: user authentication and operation log auditing; hierarchical access control (administrator / designer / construction scheduler); project isolation mechanisms (resources, models, and layout plans of each project do not interfere with each other); and encrypted storage of layout models and parameter permission settings to prevent unauthorized copying and tampering.

[0216] To verify the practicality and promotional value of the intelligent layout method for tunnel boring machine (TBM) construction sites proposed in this invention, the R&D team has completed system testing and initial deployment in several typical projects, achieving significant results. The verification work covers four dimensions: accuracy indicators, efficiency indicators, user satisfaction, and engineering adaptability.

[0217] (I) The following are the actual verification results of the above method in two typical scenarios:

[0218] Project A: Urban subway tunnel boring machine construction site (rectangular site 80m × 45m):

[0219] Component types: 14 categories in total, including segment storage yards, mixing plants, steel bar processing sheds, and office buildings;

[0220] Average deployment generation time: 5.3 seconds;

[0221] Compared with manual design: space utilization increased by 17.6%; violations of component safety distances decreased by 92%; and the average length of transportation routes decreased by 23%.

[0222] Project B: Construction area of ​​the main shaft of the river-crossing tunnel (irregularly shaped site with concave corners):

[0223] Introduce knowledge enhancement and path-aware reasoning modules;

[0224] After the solution is automatically generated, only two minor adjustments need to be made manually.

[0225] User satisfaction survey: 98% of users agree that the initial system layout meets the requirements of construction specifications.

[0226] (II) Sample Accumulation and Algorithm Evolution Results: To date, this invention has accumulated 836 high-quality layout cases; user feedback behavior data exceeds 21,000 pieces (component movement, rotation, adjustment); the model has undergone 27 rounds of fine-tuning; and the inference accuracy (user acceptance rate) has increased from 68% initially to the current 92.4%. These data constitute the core support for the system's evolutionary capabilities, driving the system "from rule-driven to experience-guided" and achieving autonomous optimization.

[0227] This invention possesses significant potential for cross-industry and cross-scenario expansion, and can be widely applied to: automated layout design of tunnel boring machine (TBM) sites; intelligent stacking of prefabricated bridge beam yards; intelligent layout optimization of precast component factory areas; spatial collaborative optimization of temporary facilities in industrial sites and mines; and deployment of construction platforms in urban underground spaces (such as station enclosure sites and temporary platforms for TBM shafts). In the future, it can also collaborate with IoT terminal devices to achieve intelligent management of construction sites through "virtual-real mapping + intelligent adjustment."

[0228] The above-mentioned practical engineering applications have verified that the two dimensions complement the aforementioned technologies. It can be seen that the present invention has a closed-loop capability from theoretical design to engineering practice, and fully demonstrates the multi-terminal deployment flexibility, system engineering interface compatibility, large model inference effectiveness, field deployment performance advantages, and industry promotion and expansion potential of the present invention.

Claims

1. A method for intelligent optimization layout of a shield tunneling construction site, characterized in that, Specifically, the following steps are included: S1. Component parameter modeling is performed for the physical parameters of various components required for the shield tunneling construction site. Component attributes are standardized, and the layout rules for various components required for the shield tunneling construction site are standardized. A knowledge rule base for the layout rules of various components is established. Based on the component parameter modeling and layout rules, a component semantic graph is established to provide support for subsequent layout optimization and reasoning generation. This serves as the basis for driving spatial layout logic and guiding intelligent sample generation. The knowledge rule base is a set of rules formed by component layout rules, safety constraints, and process logic abstraction. S2. After the component parameter modeling and layout rules are established, the spatial modeling of the shield tunneling site and its adaptation to the components are carried out. Specifically, this includes four steps: standardized modeling of site boundaries, functional zoning, spatial capacity estimation and component matching and allocation. This provides clear and reasonable spatial boundaries and layout units for subsequent layout optimization steps. S3. Based on the spatial boundary modeling, functional zoning, capacity estimation and component matching results formed in step S2, and combined with the component parameter modeling and rule modeling results in step S1, construct a composite objective function that integrates multiple engineering attributes, optimize the algorithm design and solution mechanism, and design a spatial conflict detection and dynamic repair mechanism to output a component layout scheme that meets the global optimization objective, spatial constraints and engineering logic. This scheme has local adjustability and dynamic repair capabilities. S4. Set up a human-machine collaborative layout interface to support user participation in adjustments and extract layout behavior information from user feedback for model feedback learning. Achieve continuous evolution of intelligent optimization of component-level layout through feedback learning strategy. S5. After step S4 is completed, based on the spatial structure model generated in step S2 and the component layout scheme output in step S3, the user feedback behavior in step S4 is integrated, and the layout inference engine that integrates historical samples and feedback learning mechanism is invoked to further optimize the layout strategy, outputting a final component layout scheme with better space utilization, transportation path efficiency and safety regularity; the layout inference engine adopts a retrieval-enhanced generation mechanism, using the component semantic graph and knowledge rule base constructed in step S1 as knowledge indexes, and drives the layout strategy to generate a language model based on the Transformer structure through embedded prompt input, realizing intelligent reasoning and continuous optimization of component layout logic, forming an adaptive layout evolution and continuous optimization mechanism, and having the ability to continuously evolve layout rules and optimization strategies through user behavior feedback.

2. The intelligent optimization layout method for shield tunneling construction sites according to claim 1, characterized in that: In step S1, the components of the shield tunneling site include a mixing plant, segment storage yard, rebar processing shed, temporary warehouse, office buildings, and substation. Component parameter modeling of all required components at the shield tunneling site involves uniformly describing each component as a five-tuple structure, representing the component type, geometric dimensions, functional area, layout priority, and set of layout rules. The five-tuple form for each type of component is as follows: ; in: : No. Structural description of each component; Component type identifier; Component geometric dimensions; including triplets: length ,width ,high ; The functional area to which the component belongs; Component placement priority, used to control the placement sequence; : A set of arrangement constraints with other components or site boundaries; The specific process of standardizing the construction layout rules in step S1 is as follows: The rules are standardized and modeled using the "precondition-constraint" (IF-THEN) expression structure: ； in: : No. Arrangement rules; : Rule premise; : Rule constraints; A weighting factor was also introduced into the rules. This is used to control the degree of influence of the rule on the layout optimization objective function: ; The semantic graph construction based on the layout rules in step S1 is as follows: ; in: Component layout semantic diagram; nodes : Represents all components; edge : Represents spatial constraints; Component information should be stored in a standardized manner using a unified data structure, and JSON should be used as the underlying interaction format.

3. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: In step S2, site boundary standardization modeling involves standardizing the format, validating the validity, and geometrically standardizing the input two-dimensional planar boundary. Specifically, this includes: S201, Boundary Closure Verification: The received site boundary is a sequence of point coordinates. ； in: An ordered set of polygon boundary points, representing the boundary outline of the construction site; : The coordinates of the starting point of the boundary, i.e., the first vertex of the boundary; : The coordinates of the endpoint of the boundary, i.e., the last vertex of the boundary; To ensure that the boundary is a closed polygon, the following must be satisfied: ; The closure condition indicates that the first and last points coincide, forming a closed contour; If the point is not closed, the first point will be added automatically to close it, and the points will be rearranged in a counterclockwise direction. S202. Site area calculation: The area of ​​a standard polygon is calculated using the Shoelace formula. ； in: : The calculated area of ​​the polygonal site; : No. The coordinates of the boundary points; Total number of boundary points; S203. Concavity / convexity and legality detection; To prevent component layout from being chaotic or falling into unusable areas, the continuous three-point cross product method is used to determine whether the boundary is a convex polygon: ； in: : No. Two-dimensional coordinates of the boundary points =0,1,...,n; If all If the signs are the same, it is a convex boundary; otherwise, a Graham scan or QuickHull algorithm is executed to generate the minimum convex hull boundary for subsequent layout and trimming.

4. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: In step S2, the functional zoning involves dividing the tunnel boring machine (TBM) construction site into a production area, a storage area, a living area, and an auxiliary area. The method for functional zoning is as follows: for regular rectangular sites, longitudinal or transverse regular cutting is used; for irregular sites, a Voronoi region allocation method based on polygon segmentation and centroid guidance is used, with each functional zone division result labeled with a region label. This is for component matching and judgment; The user or system inputs the functional area proportions according to task requirements. The system is based on the total area of ​​the site. Target area for allocation of functional zones : ; in: The total area ratio of all functional areas should be 1 to ensure full coverage of the site area without overlap; To ensure the validity of the area ratio, the system performs automatic normalization: ； in: : Area ratio normalization result, applicable to cases where the input ratio does not satisfy the condition that the sum is 1; Total number of functional areas; the number of different types of functional areas the system divides into. The sum of the proportions of all functional areas, used for normalization.

5. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: In step S2, space capacity estimation involves estimating the effective layout capacity and layout density control for each functional area, and matching the number of available components based on this. The specific calculations for site space capacity estimation and layout density control are as follows: Set up functional areas The area is ,member The projected area is: ； For components The dimension along its length in a horizontally arranged plane; For components The dimension along the width direction within the horizontally arranged plane; Combined with layout density coefficient Estimate the maximum capacity: ； in: The floor symbol; Arrangement density Dynamic adjustments are made based on the regularity of the site, the regularity of the components, and their rotatability. When the number of arrangements exceeds When this happens, the system issues an early warning and suggests re-planning the priority of component layout or adjusting the area ratio.

6. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: In step S2, the component matching and allocation strategy is to sort all components by priority from high to low; each component is calculated in all functional areas. The region corresponding to the maximum value is selected as the placement target; if the placement of a component causes overcapacity, an attempt is made to allocate it to the suboptimal region; these components are marked as candidate components, and their region ratios are adjusted or inserted in subsequent iterations; the component region adaptation and preliminary allocation algorithm is as follows: In the known list of components Functional area collection and layout rules Under the premise of this, execute the preliminary component allocation algorithm: Matching scoring function: Component and functional areas The matching scoring function is defined as follows: ； in: Component type and functional area matching degree; : The fit between component dimensions and area; Does it meet the layout rules? The weights of each factor are empirically set to 0.4 / 0.3 / 0.

3.

7. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: The composite objective function that integrates multiple engineering attributes in step S3 is as follows: ； in: Overall layout optimization target value; : Overlapping area of ​​components; Total length of component transportation route; Penalty for safety distance conflicts between components; Penalty for irregular layout; , , , : Corresponding weighting coefficients, set according to project requirements; (1) Overlapping area term Let the set of components be... Each component is arranged in an area of ​​100 square meters. The overlap area between the components is: ； in: :member With components The overlapping area; Area calculation function; Total number of components; The overlapping area is calculated using Boolean geometric operations; (2) Total length of transportation route The components must meet the principle of shortest path in construction logistics. The system models the distance from the component's center point to the transportation node. ； in The system provides the coordinates of the logistics entry point for the component's function; it supports weighted paths, allowing for the introduction of weights for different components with varying transportation frequencies. : ； in: :member Transportation weight; :member To the origin of the transportation route The Euclidean distance; (3) Security conflict items If the component With components shortest boundary distance If a safety violation occurs, it will be recorded. ； in: :member and Minimum boundary distance between them; Minimum safe distance threshold; : A Boolean conditional function that returns 1 if the condition is true and 0 otherwise; The system can set a dynamic minimum safety distance based on the type of component; (4) Arrangement regularity item The system introduces regularity objectives: row and column alignment difference; variance of inter-component spacing; and overall component distribution centroid offset; expressed as: ； in: :member The center coordinates; Functional area to which the component belongs The geometric center of gravity is determined by the goal of placing components as close as possible to the center of the area to reduce fragmented space. Euclidean distance squared, representing the dispersion.

8. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: The optimization strategy in step S3 employs a heuristic hierarchical layout strategy, a nested objective function search strategy, and a layout graph search engine; the spatial conflict detection and dynamic repair mechanism in step S3 includes: (1) Geometric overlap determination; Boolean overlap determination is performed using the component boundary polygons: ； The system highlights conflict areas and records the overlapping area and component pairs; (2) Detection of insufficient safe distance; set up ,like: ； If a conflict is detected, a red alarm box will be displayed, and the component will be automatically triggered to retreat or move to avoid it. (3) Detection of traffic path obstruction; determined by simulating the intersection of transport line segments and component boundary polygons: ； The system marks conflict locations on the route map, supporting the replanning of logistics channels or the automatic generation of avoidance channels.

9. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: In step S4, a CAD platform is used as the interactive environment. A layout control interface is embedded via a plug-in, combining graphical operation with real-time feedback to achieve visible, adjustable, and controllable component layout. The extracted user feedback layout behaviors include dragging and dropping various components, rotating them, adding or deleting them, modifying area constraints, and viewing layout conflicts. These extracted layout behaviors are then used to generate standardized behavior vectors. Record the original state, target state, and intent label for each operation; Based on user behavior and feedback, a low-rank adaptation parameter fine-tuning technique is used to incrementally update the deployment strategy of a large model, constructing a user preference vector. User preference vectors are injected into the intermediate layer of the large layout generation model, and the policy weights are fine-tuned through the LoRA mechanism to achieve personalized layout generation; For users A profile of layout preferences; This is a template for commonly used component combinations; For a set of preference rules; : Objective function preference weight vector; After each layout optimization or user interaction, a layout snapshot version is automatically generated, supporting layout scheme comparison, viewing score changes, and one-click rollback, enhancing project transparency and decision controllability.

10. The intelligent optimization layout method for shield tunneling construction sites according to claim 1 or 2, characterized in that: The S5 step deploys the overall architecture of the inference engine, which includes five layers: an input preprocessing layer, a graph neural network encoding layer, a policy generation layer, a deployment decoder layer, and an optimization scheduling layer. The inference steps are as follows: (1) Component context encoding; Encode the type, size, rules, and functional area location information of the components to be arranged into a vector sequence. ; (2) Vector retrieval; retrieve historical samples of similar component layouts or manually labeled rule logic from the rule knowledge base, and return a set of matching results. ; (3) Strategy fusion generation; using a large language model, the current component state and retrieval results are used as context input to generate a candidate set of placement locations. ; in, Indicates the components to be arranged The set of candidate placement locations; Representing components The One candidate arrangement posture, =1, 2, ..., ; This indicates the number of candidate layout postures generated by the system; For each candidate arrangement posture: This represents the abscissa of the component's center point in the two-dimensional plane coordinate system of the construction site. This represents the ordinate of the component's center point in the two-dimensional plane coordinate system of the construction site. It represents the rotation angle of the component in the two-dimensional plane of the construction site, that is, the orientation angle of the component relative to the positive direction of the horizontal axis of the site plane coordinate system; (4) Feasibility screening: The overlap, safety and path logic indicators of the candidate layout results are quickly evaluated through the objective function, and the one with the highest score is selected. The adaptive deployment evolution and continuous optimization mechanism in step S5 includes the construction of an evolutionary training data pool, policy replay and fine-tuning, and regional scenario summarization and migration.

Images