An interactive picture-text combined wisdom evolution construction progress generation method and system

Abstract

The application provides an interactive picture-text combined wisdom evolution construction progress generation method and system, relates to the technical field of engineering construction project management, and the method comprises the following steps: based on a hierarchical structure perception mechanism, hierarchical reading of engineering drawings is carried out; a corresponding relationship between graphic elements and text annotations is established; cross-view entities are cross-verified and corrected, and the structured semantic representation of the engineering drawings is output; the structured semantic representation is analyzed in parallel and interactively by a graph intelligent agent, a text intelligent agent, a specification review intelligent agent and a historical experience intelligent agent in sequence, and a consensus scheme is generated; based on the consensus scheme, tasks are decomposed step by step, the construction period is estimated, the task dependency relationship is established, the resource allocation is optimized, and the construction progress plan is generated. The application can realize intelligent understanding of engineering drawings, and automatically generate a construction progress plan by using multi-agent deep analysis of the structured semantic representation of the engineering drawings, thereby improving the plan compilation efficiency and enhancing the feasibility of the plan.

Description

Technical Field

[0001] This application relates to the field of engineering construction project management technology, and in particular to an interactive graphic and textual joint intelligent evolution method and system for generating construction progress. Background Technology

[0002] Construction schedule planning is the core of project management, and its scientific nature and accuracy are directly related to the project's cost, quality, and schedule.

[0003] Traditional construction schedule planning relies heavily on the personal experience of planning engineers. This involves manually interpreting large amounts of unstructured documents such as contracts, drawings, and specifications, and combining this experience with work breakdown, process logic analysis, and time estimation. This method suffers from high subjectivity, low efficiency, and difficulty in handling complex projects and dynamic changes in multiple factors.

[0004] In recent years, with the development of information technology, project management methods based on BIM (Building Information Modeling) and methods for schedule preparation using project management software (such as Microsoft Project) have emerged. These technologies have achieved information visualization and integration to a certain extent, but the core "plan generation" process is still dominated by manual processes and lacks sufficient intelligence. At the same time, existing solutions generally suffer from data silos, making it difficult to effectively connect and intelligently analyze BIM data and contract text data in the design phase with schedule data in the construction phase. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide an interactive graphic and text-based intelligent evolution method and system for generating construction progress. Through hierarchical drawing reading, text annotation corresponding to graphic elements, and cross-view entity verification and error correction, it achieves intelligent understanding of engineering drawings. It also adopts multi-agent parallel analysis and interactive analysis of the structured semantic representation of engineering drawings to automatically generate construction progress plans, effectively improving the efficiency of plan preparation. At the same time, it integrates the information contained in the engineering drawings into the plan preparation, effectively improving the practicality, adaptability, and feasibility of the plan in the construction process.

[0006] In a first aspect, embodiments of this application provide an interactive graphic-text combined intelligent evolution method for generating construction progress, the method comprising: The engineering drawings, construction contract documents, technical specifications, design specifications, and historical project references of the target project are preprocessed to obtain a standardized multimodal document set. Based on a hierarchical structure perception mechanism, standardized engineering drawings are viewed hierarchically, identifying macro layout, main drawing layer, and layer components from macro to micro perspectives and extracting detailed features of the components; establishing a correspondence between graphic elements and text annotations, wherein the text annotations originate from engineering drawings and text documents; utilizing the geometric constraints between multiple views, cross-validation and error correction are performed on cross-view entities, and a structured semantic representation of the engineering drawings is output. The structured semantic representation is simultaneously input into graph agents, text agents, specification review agents, and historical experience agents deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme. Based on the consensus scheme, tasks are decomposed step by step and the duration is estimated. Task dependencies are established, resource allocation is optimized, and a construction schedule plan for the target project is generated.

[0007] In one possible implementation, the hierarchical structure perception mechanism performs hierarchical reading of standardized engineering drawings, sequentially identifying the macro layout, main drawing layer, and layer components from macro to micro, and extracting detailed features of the components, including: Identify the frame, title block, legend area, and main drawing area of ​​standardized engineering drawings; Based on the prior knowledge-guided attention mechanism of engineering drawing standards, the layer structure of the main drawing area is separated to obtain multiple layers; wherein, the layers include at least a grid separation layer, a wall layer, a column layer, a beam layer, and a label layer; A structure-aware target detection network is used to identify components and determine their spatial locations from each layer. Based on small target detection and a dedicated engineering symbol recognizer with an engineering symbol knowledge base, detailed features are extracted from the component detail area; the detailed features include at least material symbols, connection nodes, reinforcement details, and surface treatment markings.

[0008] In one possible implementation, establishing the correspondence between graphic elements and text annotations includes: The system employs intelligent guideline tracking to establish a correspondence between the graphic elements at the starting point of the guideline and the first text label at the ending point of the guideline; wherein, the first text label originates from the engineering drawings. Using the nearest matching principle, the geometric center of the second text annotation is calculated, and all graphic elements within a preset radius of the geometric center that match the annotation type of the second text annotation are searched. A correspondence is established between the second text annotation and the nearest searched graphic element; wherein, the second text annotation originates from the engineering drawing. Input the graphic features of engineering drawings, the text features of text documents, and the spatial relationship matrix. Using a cross-attention mechanism, the graphic features are used as queries, the text features as keys, and the text semantics as values. The third text annotations of the graphic elements are output and a corresponding relationship is established. The third text annotations are derived from the text documents, which include construction contract documents, technical specification documents, design specifications, and historical project reference materials.

[0009] In one possible implementation, the step of utilizing geometric constraints between multiple views to perform cross-validation and error correction on cross-view entities includes: Based on preset view features, the view type is identified; wherein, the view type includes plan view, elevation view, and section view; Matching criteria based on consistency of axis position, numbering and labeling, and geometric dimensions allow for cross-view matching of the same entity; Based on a pre-defined consistency rule base, it is determined whether the representation of the same entity in different views conforms to the consistency rules. If it does not conform, a biased representation is identified and marked.

[0010] In one possible implementation, the graph agent analyzes geometric structures and spatial relationships, understands the spatial layout of building components, and calculates engineering quantities and dimensional parameters based on architectural drawing standards and structural mechanics knowledge. The text agent, based on an engineering terminology dictionary and a contract template library, understands text annotations and technical specifications, extracts constraints on schedule, quality, and cost, and parses contract terms and material specifications. The specification review agent retrieves applicable technical specifications based on national standards, industry standards, and local regulations, performs compliance checks, and identifies conflicts between designs and specifications. The historical experience agent retrieves similar historical projects based on the interactive memory field, provides successful case references, and warns of potential risks.

[0011] In one possible implementation, the multi-agent system sequentially performs parallel and interactive analysis on the structured semantic representation to generate a consensus scheme, including: Each intelligent agent independently analyzes the structured semantic representation and outputs preliminary conclusions; Each agent fills in the blind spots by asking and answering questions with other agents; the preliminary conclusions output by each agent are verified from multiple perspectives by other agents; if the verification fails, the agent will state its position, present and demonstrate evidence, conduct third-party arbitration, update its position, and so on until a consensus solution is generated.

[0012] In one possible implementation, the method further includes: The project metadata involved in generating the construction schedule plan, as well as the interaction records of the intelligent agent, decision nodes, and preliminary knowledge extraction results, are stored in the memory pool. When the target project analysis is completed, the construction schedule and its reasoning process are stored in the historical interactive memory; during the execution of the target project, the actual progress and problems of the project are stored in the historical interactive memory; after the target project is completed, the effect evaluation and lessons learned are stored in the historical interactive memory. Based on the two-layer memory architecture of the memory pool and the historical interaction memory bank, the collective intelligence evolution of the intelligent agent is carried out, including: identifying the success pattern of the target project, extracting the failure lessons of the target project, and optimizing or generating the reasoning rules of the intelligent agent based on the success pattern and the failure lessons.

[0013] Secondly, embodiments of this application provide an interactive, graphic-text-based, intelligently evolving construction progress generation system, the system comprising: The multimodal document collection generation module is used to preprocess the input engineering drawings, construction contract documents, technical specification documents, design specifications and historical project reference materials of the target project to obtain a standardized multimodal document collection. The structured semantic representation generation module is used to perform hierarchical reading of standardized engineering drawings based on a hierarchical structure perception mechanism. It identifies the macro layout, main drawing layer, and layer components from macro to micro perspectives and extracts the detailed features of the components. It establishes the correspondence between graphic elements and text annotations, wherein the text annotations are derived from engineering drawings and text documents. It uses the geometric constraint relationship between multiple views to cross-validate and correct cross-view entities and outputs the structured semantic representation of the engineering drawings. The consensus scheme generation module is used to simultaneously input the structured semantic representation into graph agents, text agents, normative review agents, and historical experience agents deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme. The construction schedule generation module is used to decompose tasks step by step based on the consensus scheme, estimate the construction period, establish task dependencies, optimize resource allocation, and generate the construction schedule for the target project.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the interactive graphic-text joint intelligent evolution construction progress generation method described in any of the first aspects.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the interactive graphic-text joint intelligent evolution construction progress generation method described in any one of the first aspects.

[0016] The interactive graphic-text joint intelligent evolution construction progress generation method and system provided in this application embodiment achieves joint intelligent understanding of engineering drawings through hierarchical drawing reading, text annotation corresponding to graphic elements, and cross-view entity verification and error correction. It also employs multi-agent parallel analysis and interactive analysis of the structured semantic representation of engineering drawings to automatically generate construction progress plans, effectively improving planning efficiency. Furthermore, it integrates information contained in the engineering drawings into the planning process, effectively enhancing the practicality, adaptability, and feasibility of the plan during construction. In addition, this application embodiment utilizes a two-layer memory architecture based on a memory pool and a historical interactive memory bank to achieve collective intelligent evolution of agents, extracting reusable reasoning rules from project experience to enable automatic learning and evolution of agents.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This document illustrates a flowchart of an interactive graphic-text combined intelligent evolution method for generating construction progress, as provided in an embodiment of this application. Figure 2 This application provides a flowchart of a project tracking and feedback process according to an embodiment. Figure 3 This illustration shows a structural schematic diagram of an interactive graphic and text-based intelligent evolution construction progress generation system provided in an embodiment of this application; Figure 4 A schematic diagram of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] Construction schedule planning is the core of project management, and its scientific validity and accuracy directly affect the project's cost, quality, and schedule. Traditional construction schedule planning relies heavily on the personal experience of planning engineers, involving manual interpretation of numerous unstructured documents such as contracts, drawings, and specifications, combined with experience to break down tasks, clarify logical relationships between work processes, and estimate schedule duration. This method suffers from high subjectivity, low efficiency, and difficulty in handling complex projects and dynamic changes in multiple factors.

[0022] In recent years, with the development of information technology, project management methods based on BIM (Building Information Modeling) and methods for schedule preparation using project management software (such as Microsoft Project) have emerged. These technologies have achieved information visualization and integration to a certain extent, but the core "plan generation" process is still dominated by manual processes and lacks sufficient intelligence. At the same time, existing solutions generally suffer from data silos, making it difficult to effectively connect and intelligently analyze BIM data and contract text data in the design phase with schedule data in the construction phase.

[0023] Existing methods for automatically generating construction schedules generally suffer from the following problems: 1. Low utilization rate of project documents: There are a large number of unstructured documents in engineering projects, including engineering drawings, construction contracts, technical specifications, etc., and these unstructured documents have not been analyzed in depth.

[0024] 2. Difficulty in understanding drawings: Engineering drawings contain complex graphics, symbols and text annotations, which traditional OCR (Optical Character Recognition) and general vision models struggle to accurately understand professional semantics.

[0025] 3. Limitations of a single model: Complex engineering projects require expertise from multiple fields, which a single AI model cannot fully cover, and it lacks a self-verification mechanism.

[0026] 4. Difficulty in passing on experience: The successful experiences and lessons learned from past projects are difficult to extract and reuse effectively, requiring re-analysis each time.

[0027] 5. Lack of evolutionary capability: Most existing systems are static models, which cannot continuously learn and optimize themselves from practice.

[0028] The advantages and disadvantages of existing technical solutions are illustrated by the following two examples.

[0029] Example 1: Automatic construction schedule planning based on BIM and rule-based reasoning. This technical solution is designed for mixed concrete structures and aims to achieve automatic generation of component-level schedules. The specific technical solution is as follows: 1. Activity Decomposition and Coding: First, construction activities are decomposed and a coding system suitable for automatic scheduling is established to classify and identify components and construction activities.

[0030] 2. Rule-based reasoning generates logical relationships: Analyze the constraint rules (mainly spatial and physical constraints) between various components or construction activities, and infer the logical sequence (such as precedence and succession relationships) between construction activities based on these rules. For example, a rule might be defined as "the support of the floor slab must be installed after its formwork is installed".

[0031] 3. Extract quantities from BIM: Extract quantity information for each building component from the BIM model based on the IFC standard.

[0032] 4. Construction Period Calculation and Schedule Generation: Using the extracted quantities of work, the duration of each construction activity is calculated and determined. Finally, these activities with logical relationships and construction periods are integrated to automatically generate a component-level construction schedule for construction simulation.

[0033] The core process of this solution can be summarized as follows: BIM model (IFC standard) → activity decomposition and coding → logical reasoning based on predefined constraint rules → quantity extraction and schedule calculation → generation of schedule plan.

[0034] While Example 1 introduces rule-based reasoning, representing a step forward from pure case or template methods, it still suffers from the following significant drawbacks: 1. Limited and static rule coverage: The rules relied upon by this scheme mainly focus on spatial and physical constraints between components (such as support relationships). It fails to effectively understand and incorporate a large amount of unstructured knowledge existing in contracts, specifications, and construction organization designs (such as process requirements, resource allocation constraints, safety regulations, and preferences for specific construction methods). Its rule base is predefined and static, making it difficult to adapt to complex and ever-changing actual construction scenarios and new technologies, and lacking self-learning and adaptive capabilities.

[0035] 2. Lack of ability to process unstructured engineering documents: This solution relies entirely on structured BIM (IFC standard) models as data input sources. The system lacks the ability to understand and extract the rich schedule constraint information contained in crucial unstructured documents (such as construction contracts, technical specifications, and drawings), potentially leading to the omission of key business logic and constraints in the generated plans.

[0036] 3. The reasoning mechanism is simplistic and lacks collaboration and deep reasoning: The reasoning process of this solution is essentially matching predefined rules with BIM components, which is a relatively unidirectional and linear process. It lacks the division of labor, cooperation, and game theory among multiple specialized "intelligent agents," and cannot simulate the complex decision-making process in actual construction management that requires balancing multiple factors (such as cost, schedule, resources, and safety). It also struggles to handle rule conflicts or perform common-sense reasoning.

[0037] 4. Limited precision and practicality of the generated schedule: Due to the limitations of rules and data, although the schedule generated by this method is more detailed at the "component level", it is still insufficient in terms of the completeness of task decomposition, the precision of logical relationships, and the fit with the actual construction organization. It is difficult to directly use it to guide the on-site execution of complex projects, and there is a gap between it and the ultimate goal of "a schedule that can guide project execution".

[0038] Example 2: A deep learning-based system and method for predicting the progress of construction projects. The core of this technical solution lies in using a deep learning model to learn progress patterns from historical project data to predict the progress of new projects. The technical solution specifically includes three processes: 1. Data preparation: Extract data from historical projects and perform preprocessing such as cleaning, correction, supplementation and regularization on these data to form the training dataset.

[0039] 2. Model Training and Validation: An initial deep learning prediction model is constructed and trained using preprocessed historical data. This approach places particular emphasis on validation with "special data" (such as data from special scenarios like project delays and resource interruptions), and the validated model is used as the final progress prediction model.

[0040] 3. Schedule Prediction Process: Acquire the characteristic information of the current actual project and input it into the trained schedule prediction model. The model outputs predicted project schedule information (such as total duration or key milestone duration) that is related to the actual project, and can provide early warnings for potential risks.

[0041] This approach focuses on mining patterns from structured historical data. The process can be summarized as follows: collect historical data → data preprocessing → train and validate deep learning models → input new project features → output predicted progress and early warnings.

[0042] Although Example 2 utilizes advanced deep learning techniques, it has the following inherent drawbacks: 1. "Black box" decision-making with poor interpretability: The schedule predictions made by deep learning models lack clear logical explanations and reasoning processes. Project managers cannot understand why the model gives a specific timeframe or warning, and it is difficult to trace the basis of its decision-making (e.g., whether it is based on process logic or resource constraints), resulting in low trust in the prediction results and hindering its application and promotion in actual project management.

[0043] 2. Functional limitations: This method is essentially a macro-level regression prediction or classification problem, and its output is usually the total project duration, key milestone dates, or delay risk level. It cannot generate an executable schedule that includes a complete Work Breakdown Structure (WBS), detailed procedures, logical relationship networks, and resource allocation, therefore it cannot replace traditional schedule preparation work.

[0044] 3. Heavy reliance on large amounts of high-quality structured historical data: The model's performance is highly dependent on the quantity, quality, and consistency of historical data. For new or special projects with scarce data, inconsistent data formats, or poor data quality, the model's predictive performance will significantly decrease or even become unusable. It cannot make direct predictions using the large amounts of unstructured documents (such as contracts and drawings) generated in the early stages of a project.

[0045] 4. Disconnected from the schedule generation process: This solution is an independent forecasting system. Its forecast results need to be manually interpreted by the planning engineer and then manually integrated into the schedule preparation process. It fails to achieve end-to-end automatic generation from data to plan, and the degree of automation is limited.

[0046] Therefore, the construction engineering field urgently needs an intelligent system that can accurately understand engineering drawings, deeply integrate artificial intelligence technology to achieve multi-perspective collaborative reasoning, and possess self-evolution capabilities, thereby enabling automatic understanding of engineering documents, intelligent reasoning of construction logic, and dynamic generation and optimization of schedule plans.

[0047] To address the aforementioned issues, this application provides an interactive, text-based, and intelligently evolving method and system for generating construction schedules. Through hierarchical drawing interpretation, text annotation corresponding to graphic elements, and cross-view entity verification and error correction, it achieves intelligent understanding of engineering drawings. Furthermore, it employs multi-agent parallel analysis and interactive analysis of the structured semantic representation of engineering drawings to automatically generate construction schedule plans, effectively improving planning efficiency. Simultaneously, it integrates information contained within the engineering drawings into the planning process, effectively enhancing the plan's practicality and adaptability during construction. Moreover, based on a two-layer memory architecture of a short-term memory pool and a long-term historical interactive memory bank, the agents extract reusable reasoning rules from project experience, continuously learning and self-optimizing from practice, making it an intelligent system with self-evolutionary capabilities.

[0048] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0049] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0050] To facilitate understanding of this embodiment, a detailed description of an interactive graphic and text-based intelligent evolution method for generating construction progress, as disclosed in this application embodiment, will be provided first.

[0051] See Figure 1 As shown, Figure 1 A flowchart of an interactive graphic-text combined intelligent evolution method for generating construction progress is provided in this application embodiment. The method includes the following steps: S101. Preprocess the engineering drawings, construction contract documents, technical specifications, design specifications and historical project references of the input target project to obtain a standardized multimodal document set.

[0052] Step S101: Document input and preprocessing.

[0053] The target project is any construction project that requires the generation of a construction schedule. During the implementation of such a project, numerous written and drawing materials will be involved, including but not limited to: engineering drawings (PDF, DWG, JPG, etc.), construction contract documents, technical specifications, design specifications, and historical project references. To facilitate subsequent processing of these written and drawing materials, preprocessing is required. Step S101, the preprocessing process, specifically includes: S1011. Automatic document type recognition: (1) Identification of drawing types: plan view, elevation view, section view, detail view, etc.

[0054] (2) Classification of text documents: contracts, specifications, instructions, etc.

[0055] S1012, Quality Assessment and Enhancement: Image sharpness detection; Document integrity check; Low-quality image enhancement processing.

[0056] S1013, Format Unification Conversion: Converts different formats to a standard processing format; extracts editable text and image data.

[0057] After processing through steps S1011-S1013, a standardized multimodal document set is obtained. Multimodal means that the document set contains data such as text, images, and tables.

[0058] S102. Based on the hierarchical structure perception mechanism, the standardized engineering drawings are viewed hierarchically, and the macro layout, main drawing layer, and layer components are identified sequentially from macro to micro and the detailed features of the components are extracted; the correspondence between graphic elements and text annotations is established, wherein the text annotations are derived from the engineering drawings and text documents; using the geometric constraint relationship between multiple views, cross-validation and error correction are performed on cross-view entities, and the structured semantic representation of the engineering drawings is output.

[0059] Step S102: Intelligent understanding of engineering drawings.

[0060] Engineering drawings differ from natural images and have the following unique characteristics: 1. Symbolic representation: Use standard engineering symbol system.

[0061] 2. Precise measurement: Dimensions are accurate to the millimeter level.

[0062] 3. Tight coupling of text and graphics: Text annotations directly identify the semantic meaning of graphics.

[0063] 4. Multi-view association: Plan view, elevation view, and section view represent the same entity.

[0064] 5. Specialized semantics: Domain knowledge is required for correct understanding.

[0065] General-purpose visual models cannot effectively handle these characteristics. Therefore, this application uses a dedicated model that combines structure awareness and semantic fusion (SASF Model) to extract the structured semantic representation of engineering drawings.

[0066] Step S102 processes the standardized engineering drawings in the standardized multimodal document set obtained in step S101 to obtain a structured semantic representation of the engineering drawings, enabling intelligent understanding of the engineering drawings. The specific processing steps of the SASF Model include: employing a hierarchical structure perception mechanism to simulate the hierarchical drawing reading process of human engineers, analyzing the engineering drawings layer by layer from macro to micro (steps S1021-S1024); extracting graphic elements from the engineering drawings and extracting text annotations of the graphic elements from text documents such as engineering drawings, construction contract documents, technical specifications, design specifications, and historical project reference materials, and establishing a correspondence between graphic elements and text annotations (steps S1025-S1027); the information of the same entity in the plan view, elevation view, and section view should be consistent. If inconsistencies occur, the entity information of each view needs to be unified. In this embodiment, the geometric constraint relationship between views is used to unify the entity information (steps S1028-S10210). Through the processing of steps S1021-S10210, a structured semantic representation of the engineering drawings is obtained. The specific processing flow for step S102 includes: S1021. Identify the frame, title block, legend area, and main drawing area of ​​standardized engineering drawings.

[0067] First layer: Macro layout identification.

[0068] Input: Complete drawing (2048×2048 resolution).

[0069] Processing: Identify the title block, legend area, and main image area.

[0070] Output: Global layout structure F_global.

[0071] Function: To establish a holistic cognitive framework.

[0072] S1022. Based on the prior knowledge-guided attention mechanism of engineering drawing specifications, the layer structure of the main drawing area is separated to obtain multiple layers; wherein, the layers include at least a separated grid layer, a wall layer, a column layer, a beam layer, and a label layer.

[0073] Second layer: Layer structure separation.

[0074] Input: Main image area (1024×1024 resolution).

[0075] Processing: Separate the grid layer, wall layer, column layer, beam layer, and labeling layer, etc.

[0076] Technology: Attention-guided mechanism based on prior knowledge of engineering drawing standards.

[0077] Output: Layered feature representation F_layer={F_grid, F_wall, F_column, F_beam, ...}.

[0078] Function: To organize drawing elements according to engineering logic.

[0079] S1023, A structure-aware target detection network identifies components from each layer and determines their spatial location.

[0080] The third layer: component identification and positioning.

[0081] Input: Single layer area (512×512 resolution).

[0082] Processing: Identify specific components (columns, beams, walls, doors, windows, etc.).

[0083] Technology: Structure-aware target detection network.

[0084] Output: Component feature set F_component={f_c1, f_c2, ..., f_cn}.

[0085] Function: Extract all components and their spatial positions.

[0086] S1024. Based on small target detection and an engineering symbol-specific recognizer with an engineering symbol knowledge base, extract detailed features from the component detail area; the detailed features include at least material symbols, connection nodes, reinforcement details, and surface treatment markings.

[0087] Fourth layer: Detail feature extraction.

[0088] Input: Component detail area (256×256 resolution).

[0089] Processing: Identify material symbols, connection nodes, reinforcement details, and surface treatment markings.

[0090] Technology: Small target detection + engineering symbol recognition device.

[0091] Output: Detail feature set F_detail.

[0092] Function: To capture detailed information required for construction.

[0093] It should be noted that the engineering symbol recognition device used in this application has the advantages of a specially trained small target detection network (symbols are usually 10-50 pixels), rotation invariance design (symbols may be in different directions), and context awareness (the meaning of the symbol depends on the surrounding graphics), and can extract 70 material symbols, 120 construction symbols, 85 pipeline symbols, and 150 equipment symbols.

[0094] Steps S1021-S1024 are used to achieve hierarchical reading of engineering drawings.

[0095] S1025. Intelligent tracking of guide lines is adopted to establish a correspondence between the graphic elements located at the starting point of the guide line and the first text label located at the ending point of the guide line; wherein, the first text label originates from the engineering drawing.

[0096] Alignment Strategy 1: Intelligent tracking of guide lines.

[0097] Algorithm steps: 1. Inspection guide line (thin solid line, connecting graphics and text).

[0098] 2. Trace the guide line path (handle polylines and curves).

[0099] 3. Determine the starting point (graphic anchor point) and ending point (text position) of the guide line.

[0100] 4. Establish corresponding relationships.

[0101] Example: Graphical element: column, position (x=850, y=1200).

[0102] Guide lines: from (850, 1200) to (950, 1150).

[0103] Text annotation: Position (950, 1150), content "400×400".

[0104] Corresponding relationship: The column cross-section dimensions are 400mm × 400mm.

[0105] Confidence level: 0.95 (leader line is clear).

[0106] It should be noted that the first text label of the graphic element in alignment strategy 1 originates from the engineering drawing, and the graphic element of the engineering drawing is labeled with the text information contained in the engineering drawing.

[0107] S1026. Using the nearest matching principle, calculate the geometric center of the second text annotation, search for all graphic elements within the preset radius of the geometric center that are consistent with the annotation type of the second text annotation, and establish a correspondence between the second text annotation and the nearest searched graphic element; wherein, the second text annotation originates from the engineering drawing.

[0108] Alignment Strategy 2: Matching based on proximity.

[0109] Applicable scenarios: Labels without obvious guide lines.

[0110] Algorithm steps: 1. Calculate the geometric center of the text annotation.

[0111] 2. Search for all graphic elements within a radius R (R adapts to the drawing scale).

[0112] 3. Determine the annotation type according to engineering drawing specifications: Dimensioning: When located between dimension lines, it is associated with the graphic being dimensioned.

[0113] Material labeling: If it is located inside or beside a component, it is associated with the corresponding component.

[0114] Numbering label: Inside the circle / rectangle box, it is associated with the selected component.

[0115] Example: Text label: "C30", position (920, 1180).

[0116] Recent graphic: Column member, center (900, 1200), distance 28 pixels.

[0117] Label type determination: Material label (located next to the component).

[0118] Corresponding relationship: The concrete strength grade of this column is C30.

[0119] Confidence level: 0.88 (no guide line, based on proximity principle).

[0120] It should be noted that the second text annotation of the graphic elements in alignment strategy 2 originates from the engineering drawings, and uses the text information contained in the engineering drawings to annotate the graphic elements of the engineering drawings.

[0121] S1027. Input the graphic features of the engineering drawings, the text features of the text document, and the spatial relationship matrix. Using a cross-attention mechanism, use the graphic features as the query, the text features as the key, and the text semantics as the value. Output the third text annotation of the graphic elements and establish the corresponding relationship. The third text annotation originates from the text document, which includes construction contract documents, technical specification documents, design specifications, and historical project reference materials.

[0122] Alignment Strategy 3: Cross-modal semantic fusion.

[0123] Inputs to the cross-modal semantic fusion network: graphic features (also known as visual features): F_visual (from the hierarchical encoder); text features: F_text (from the OCR+BERT encoder); spatial relation matrix: R_spatial (alignment relation, N×M matrix).

[0124] Fusion mechanism: Cross-Modal Attention. Graphical features serve as the query: Query = F_visual; textual features serve as the key: Key = F_text; semantic text serves as the value: Value = F_text.

[0125] The output of the cross-modal semantic fusion network: each graphic element is accompanied by a complete semantic annotation, that is, the third text annotation of the graphic element.

[0126] It should be noted that the third text annotation of the graphic elements in alignment strategy 3 originates from the text document, and uses the text information of the text document to annotate the graphic elements of the engineering drawings.

[0127] The three alignment strategies in steps S1025-S1027 are adopted to achieve spatial alignment and semantic fusion of graphics and text, thereby integrating engineering drawings with text documents to enrich the content of the structured semantic representation of engineering drawings.

[0128] S1028. Identify view types based on preset view features; wherein, the view types include plan view, elevation view, and section view.

[0129] The view type is identified using step S1028. The view characteristics of plan views, elevation views, and section views are as follows: Floor plan features: include grid lines, wall projections, and room layout.

[0130] Elevation features: include the exterior facade, vertical window distribution, and building height.

[0131] Cross-sectional features: include floor height, interior space, and vertical structure.

[0132] S1029. Matching the same entity across views based on the consistency of axis position, numbering and labeling, and geometric dimensions.

[0133] For different entities in engineering drawings, set matching rules for the same entity in plan view, elevation view, and section view.

[0134] For example, the same column can be represented in different views as follows: Plan view: Rectangular cross section, 400mm×400mm, location of the intersection of axes.

[0135] Elevation view: Vertical straight lines, from the bottom floor to the top floor.

[0136] Sectional view: Rectangular section, height 3.6m (floor height).

[0137] The matching criteria for this column are: consistency of axis position (intersection of the same axis); consistency of numbering (e.g., column number "Z1"); and consistency of geometric dimensions (consistent cross-sectional dimensions).

[0138] S10210. Based on the preset consistency rule base, determine whether the representation of the same entity in different views conforms to the consistency rules. If it does not conform, determine the biased representation and mark it.

[0139] Configure a consistency rule base, for example: Rule 1: The column cross-sectional dimensions in the plan view are equal to the width of the corresponding column in the section view.

[0140] Rule 2: The height of a window in the elevation drawing is equal to the vertical dimension of the corresponding window in the sectional drawing.

[0141] Rule 3: The axis distance in the plan view is equal to the corresponding horizontal distance in the elevation view.

[0142] Rule 4: The elevation values ​​in each view should match each other.

[0143] If any rule in the consistency rule base is not met, an error correction mechanism is adopted to determine the biased expression. That is, when the information in the plan view, elevation view, and section view is inconsistent, the information with higher credibility is selected from the information contained in the three views as the biased expression, and this inconsistency is marked.

[0144] For example: Plan view annotation: Beam cross-section 200×500; Section view annotation: Beam cross-section 200×550. If the beam cross-section does not conform to the consistency rules, the automatic error correction process will be activated. 1. Calculate the actual measured dimensions of the graphic, including: Plan view measurement: Beam height cannot be measured (only width is displayed); Section view measurement: Actual beam height = 545mm.

[0145] 2. Determine which annotation is more reliable: The beam height annotation in the section view is more direct (shown in the elevation). Conclusion: Adopt the 550mm in the section view as a biased expression.

[0146] 3. Mark the plan view as potentially incorrect, confidence level: 0.87 (based on multi-view cross-validation).

[0147] Steps S1021 to S10210 simulate the human drawing reading process, perceiving engineering drawings in a hierarchical structure from macro to micro levels; three strategies are employed for spatial alignment of graphics and text to ensure accurate correspondence between graphics and text; and cross-validation using geometric constraints is used for automatic error correction, enabling multi-view processing. Figure 1 Consistent reasoning enables intelligent understanding of engineering drawings.

[0148] S103. The structured semantic representation is simultaneously input into the graph agent, text agent, specification review agent, and historical experience agent deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme.

[0149] Step S103: Multi-agent collaborative analysis.

[0150] Four types of intelligent agents are used to complete the processing task in step S103. The professional fields, core capabilities, and knowledge sources of the intelligent agents are as follows: Agent 1: Graph Agent (VA).

[0151] Area of ​​expertise: Visual spatial understanding.

[0152] Core competencies: analyzing geometric structures and spatial relationships, understanding the spatial layout of building components, and calculating engineering quantities and dimensional parameters.

[0153] Knowledge sources: architectural drawing standards and structural mechanics knowledge.

[0154] Agent 2: Text Agent (TA).

[0155] Area of ​​expertise: Text semantic understanding.

[0156] Core competencies: Understanding textual annotations and technical specifications, extracting constraints (project time, quality, cost), and analyzing contract terms and material specifications.

[0157] Knowledge sources: Engineering terminology dictionary, contract template library.

[0158] Agent 3: Specification Agent (SA).

[0159] Professional field: Standards and specifications verification.

[0160] Core capabilities: Retrieve applicable technical specifications, perform compliance checks, and identify conflicts between design and specifications.

[0161] Knowledge sources: national standards, industry norms, and local regulations.

[0162] Agent 4: Experience Agent (EA).

[0163] Area of ​​expertise: Experience and knowledge management.

[0164] Core capabilities: Search for similar historical projects, provide successful case studies for reference, and warn of potential risks.

[0165] Knowledge source: Interactive memory field (see the description of step S105, the interactive memory field includes a memory pool and a historical interactive memory bank).

[0166] In step S103, the multi-agent collaborative analysis process is as follows: S1031. Each agent independently analyzes the structured semantic representation and outputs preliminary conclusions.

[0167] Step S1031 is the parallel analysis phase. Each agent independently analyzes the same project and proposes preliminary insights. The structured semantic representation of the engineering drawings of the target project is distributed to VA, TA, SA, and EA. Then, VA, TA, SA, and EA simultaneously start the analysis process and output their respective analysis results, i.e., the preliminary conclusions of the construction plan, thus obtaining four preliminary conclusions.

[0168] S1032. Each agent fills in the blind spot information by asking and answering questions with other agents. The preliminary conclusions output by each agent are verified from multiple perspectives by other agents. If the verification fails, the agent will state its position, present and demonstrate evidence, conduct third-party arbitration, update its position, and so on until a consensus solution is generated.

[0169] Step S1032 is the interactive analysis phase, which enables question-and-answer, verification, and semantic collaboration among agents. The first layer is question-and-answer information completion, where agents proactively ask questions to fill in their own knowledge gaps. The second layer is cross-validation, where each agent's output is verified from multiple perspectives by other agents. The third layer is consensus-based negotiation, where disagreements are resolved through structured negotiation. This includes: Round 1: Position statement; Round 2: Evidence presentation and argumentation; Round 3: Third-party arbitration; Round 4: Position update and consensus formation. Ultimately, a consensus scheme is generated that is agreed upon by graph agents, text agents, normative review agents, and historical experience agents.

[0170] S104. Based on the consensus scheme, decompose the tasks step by step and estimate the duration, establish task dependencies, optimize resource allocation, and generate the construction schedule plan for the target project.

[0171] Step S104: Construction plan generation.

[0172] The consensus scheme is decomposed into tasks. First, the first-level tasks are decomposed and their durations are estimated. Then, the first-level tasks are further decomposed to obtain second-level tasks and their durations are estimated. The second-level tasks are branches of the first-level tasks.

[0173] Generate a construction dependency graph, establish the dependencies between construction tasks, and determine the construction sequence of the tasks.

[0174] Resource allocation optimization, including human resources, equipment resources, and material procurement plans.

[0175] Develop a construction schedule: Use Gantt charts to schedule the implementation time of construction tasks, determine the critical path (i.e., the construction sequence), and determine the total construction period.

[0176] As one possible implementation method, embodiments of this application also include project execution tracking and feedback, see [link to relevant documentation]. Figure 2 As shown, Figure 2 A flowchart for project tracking and feedback provided in this application embodiment specifically includes: S1051. Store the project metadata, agent interaction records, decision nodes, and preliminary knowledge extraction results involved in generating the construction schedule into the memory pool.

[0177] S1052. When the target project analysis is completed, the construction schedule and its reasoning process are stored in the historical interactive memory bank; during the execution of the target project, the actual progress and problems of the project are stored in the historical interactive memory bank; after the target project is completed, the effect evaluation and lessons learned are stored in the historical interactive memory bank.

[0178] S1053. Based on the dual-layer memory architecture of the memory pool and the historical interaction memory bank, perform collective intelligence evolution of the intelligent agent, including: identifying the success pattern of the target project, extracting the failure lessons of the target project, and optimizing or generating the reasoning rules of the intelligent agent based on the success pattern and the failure lessons.

[0179] Steps S1051-S1053 are used to archive the current analysis memory pool (CAM) and the historical interaction memory bank. The data from the memory pool and the historical interaction memory bank are used to carry out swarm intelligence evolution, strengthen the existing rules of the agent (i.e. optimize the rules), or generate new rules for the agent.

[0180] Based on the same inventive concept, this application also provides an interactive graphic and text-based construction progress generation system corresponding to the interactive graphic and text-based intelligent evolution construction progress generation method. Since the principle of the system in this application is similar to the above-mentioned interactive graphic and text-based intelligent evolution construction progress generation method in this application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0181] See Figure 3 As shown, Figure 3 A schematic diagram of an interactive graphic-text combined intelligent evolution construction progress generation system provided in this application embodiment, the system comprising: The multimodal document collection generation module 301 is used to preprocess the engineering drawings, construction contract documents, technical specification documents, design specifications and historical project reference materials of the input target project to obtain a standardized multimodal document collection. The structured semantic representation generation module 302 is used to perform hierarchical reading of standardized engineering drawings based on a hierarchical structure perception mechanism. It identifies the macro layout, main drawing layer, and layer components from macro to micro and extracts the detailed features of the components. It establishes the correspondence between graphic elements and text annotations, wherein the text annotations are derived from engineering drawings and text documents. It uses the geometric constraint relationship between multiple views to cross-validate and correct cross-view entities and outputs the structured semantic representation of the engineering drawings. The consensus scheme generation module 303 is used to simultaneously input the structured semantic representation into the graph agent, text agent, specification review agent, and historical experience agent deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme. The construction schedule generation module 304 is used to decompose tasks step by step based on the consensus scheme, estimate the construction period, establish task dependencies, optimize resource allocation, and generate the construction schedule for the target project.

[0182] In one possible implementation, the structured semantic representation generation module 302, when performing hierarchical reading of standardized engineering drawings based on a hierarchical structure perception mechanism, sequentially identifying macro layout, main drawing layer, and layer components from macro to micro and extracting component detail features, includes: Identify the frame, title block, legend area, and main drawing area of ​​standardized engineering drawings; Based on the prior knowledge-guided attention mechanism of engineering drawing standards, the layer structure of the main drawing area is separated to obtain multiple layers; wherein, the layers include at least a grid separation layer, a wall layer, a column layer, a beam layer, and a label layer; A structure-aware target detection network is used to identify components and determine their spatial locations from each layer. Based on small target detection and a dedicated engineering symbol recognizer with an engineering symbol knowledge base, detailed features are extracted from the component detail area; the detailed features include at least material symbols, connection nodes, reinforcement details, and surface treatment markings.

[0183] In one possible implementation, the structured semantic representation generation module 302, when establishing the correspondence between graphic elements and text annotations, includes: The system employs intelligent guideline tracking to establish a correspondence between the graphic elements at the starting point of the guideline and the first text label at the ending point of the guideline; wherein, the first text label originates from the engineering drawings. Using the nearest matching principle, the geometric center of the second text annotation is calculated, and all graphic elements within a preset radius of the geometric center that match the annotation type of the second text annotation are searched. A correspondence is established between the second text annotation and the nearest searched graphic element; wherein, the second text annotation originates from the engineering drawing. Input the graphic features of engineering drawings, the text features of text documents, and the spatial relationship matrix. Using a cross-attention mechanism, the graphic features are used as queries, the text features as keys, and the text semantics as values. The third text annotations of the graphic elements are output and a corresponding relationship is established. The third text annotations are derived from the text documents, which include construction contract documents, technical specification documents, design specifications, and historical project reference materials.

[0184] In one possible implementation, the structured semantic representation generation module 302, when performing cross-validation and error correction on cross-view entities using the geometric constraints between multiple views, includes: Based on preset view features, the view type is identified; wherein, the view type includes plan view, elevation view, and section view; Matching criteria based on consistency of axis position, numbering and labeling, and geometric dimensions allow for cross-view matching of the same entity; Based on a pre-defined consistency rule base, it is determined whether the representation of the same entity in different views conforms to the consistency rules. If it does not conform, a biased representation is identified and marked.

[0185] In one possible implementation, the graph agent analyzes geometric structures and spatial relationships, understands the spatial layout of building components, and calculates engineering quantities and dimensional parameters based on architectural drawing standards and structural mechanics knowledge. The text agent, based on an engineering terminology dictionary and a contract template library, understands text annotations and technical specifications, extracts constraints on schedule, quality, and cost, and parses contract terms and material specifications. The specification review agent retrieves applicable technical specifications based on national standards, industry standards, and local regulations, performs compliance checks, and identifies conflicts between designs and specifications. The historical experience agent retrieves similar historical projects based on the interactive memory field, provides successful case references, and warns of potential risks.

[0186] In one possible implementation, the consensus scheme generation module 303, when multiple agents sequentially perform parallel analysis and interactive analysis on the structured semantic representation to generate a consensus scheme, includes: Each intelligent agent independently analyzes the structured semantic representation and outputs preliminary conclusions; Each agent fills in the blind spots by asking and answering questions with other agents; the preliminary conclusions output by each agent are verified from multiple perspectives by other agents; if the verification fails, the agent will state its position, present and demonstrate evidence, conduct third-party arbitration, update its position, and so on until a consensus solution is generated.

[0187] In one possible implementation, the system further includes: The memory pool storage module is used to store the project metadata involved in generating the construction schedule plan, as well as the interaction records, decision nodes, and preliminary knowledge extraction results of the intelligent agent into the memory pool. The historical interactive memory storage module is used to store the construction schedule plan and its reasoning process into the historical interactive memory when the target project analysis is completed; to store the actual progress and problems of the project into the historical interactive memory during the execution of the target project; and to store the effect evaluation and lessons learned into the historical interactive memory after the target project is completed. The agent reasoning rule optimization module is used to perform collective intelligence evolution of agents based on a two-layer memory architecture of the memory pool and the historical interaction memory bank. It includes: identifying the success patterns of the target project, extracting the lessons learned from the failures of the target project, and optimizing or generating the reasoning rules of the agent based on the success patterns and the lessons learned from the failures.

[0188] The interactive graphic-text joint intelligent evolution construction progress generation system provided in this application embodiment achieves intelligent understanding of engineering drawings through hierarchical drawing reading, text annotation corresponding to graphic elements, and cross-view entity verification and error correction. It also employs multi-agent parallel analysis and interactive analysis of the structured semantic representation of engineering drawings to automatically generate construction progress plans, effectively improving planning efficiency. Furthermore, it integrates information contained in the engineering drawings into the planning process, effectively enhancing the practicality, adaptability, and feasibility of the plan during construction. In addition, based on a two-layer memory architecture of a short-term memory pool and a long-term historical interactive memory bank, the intelligent agents extract reusable reasoning rules from project experience, continuously learning and self-optimizing from practice, making it an intelligent system with self-evolutionary capabilities.

[0189] See Figure 4 As shown, Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device 400 includes a processor 401, a memory 402, and a bus 403. The memory 402 stores machine-readable instructions that can be executed by the processor 401. When the electronic device is running, the processor 401 communicates with the memory 402 through the bus 403. The processor 401 executes the machine-readable instructions to perform the steps of the construction progress generation method of interactive graphic and text joint intelligent evolution as described above.

[0190] Specifically, the memory 402 and processor 401 mentioned above can be general-purpose memory and processor, without any specific limitations. When the processor 401 runs the computer program stored in the memory 402, it can execute the above-mentioned interactive graphic and textual joint intelligent evolution construction progress generation method.

[0191] Corresponding to the above-described interactive graphic and text-based intelligent evolution method for generating construction progress, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described interactive graphic and text-based intelligent evolution method for generating construction progress.

[0192] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces; the indirect coupling or communication connection of devices or modules may be electrical, mechanical, or other forms.

[0193] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0194] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0195] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0196] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A construction progress generation method based on interactive graphic and textual joint intelligent evolution, characterized in that, The method includes: The engineering drawings, construction contract documents, technical specifications, design specifications, and historical project references of the target project are preprocessed to obtain a standardized multimodal document set. Based on a hierarchical structure perception mechanism, standardized engineering drawings are viewed hierarchically, identifying macro layout, main drawing layer, and layer components from macro to micro perspectives and extracting detailed features of the components; establishing a correspondence between graphic elements and text annotations, wherein the text annotations originate from engineering drawings and text documents; utilizing the geometric constraints between multiple views, cross-validation and error correction are performed on cross-view entities, and a structured semantic representation of the engineering drawings is output. The structured semantic representation is simultaneously input into graph agents, text agents, specification review agents, and historical experience agents deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme. Based on the consensus scheme, tasks are decomposed step by step and the duration is estimated. Task dependencies are established, resource allocation is optimized, and a construction schedule plan for the target project is generated.

2. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 1, characterized in that, The hierarchical structure perception mechanism described above performs hierarchical reading of standardized engineering drawings, identifying macro layout, main drawing layer, and layer components sequentially from macro to micro, and extracting detailed features of the components, including: Identify the frame, title block, legend area, and main drawing area of ​​standardized engineering drawings; Based on the prior knowledge-guided attention mechanism of engineering drawing standards, the layer structure of the main drawing area is separated to obtain multiple layers; wherein, the layers include at least a grid separation layer, a wall layer, a column layer, a beam layer, and a label layer; A structure-aware target detection network is used to identify components and determine their spatial locations from each layer. Based on small target detection and a dedicated engineering symbol recognizer with an engineering symbol knowledge base, detailed features are extracted from the component detail area; the detailed features include at least material symbols, connection nodes, reinforcement details, and surface treatment markings.

3. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 1, characterized in that, The process of establishing the correspondence between graphic elements and text labels includes: The system employs intelligent guideline tracking to establish a correspondence between the graphic elements at the starting point of the guideline and the first text label at the ending point of the guideline; wherein, the first text label originates from the engineering drawings. Using the nearest matching principle, the geometric center of the second text annotation is calculated, and all graphic elements within a preset radius of the geometric center that match the annotation type of the second text annotation are searched. A correspondence is established between the second text annotation and the nearest searched graphic element; wherein, the second text annotation originates from the engineering drawing. Input the graphic features of engineering drawings, the text features of text documents, and the spatial relationship matrix. Using a cross-attention mechanism, the graphic features are used as queries, the text features as keys, and the text semantics as values. The third text annotations of the graphic elements are output and a corresponding relationship is established. The third text annotations are derived from the text documents, which include construction contract documents, technical specification documents, design specifications, and historical project reference materials.

4. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 1, characterized in that, The method of utilizing geometric constraints between multiple views to perform cross-validation and error correction on cross-view entities includes: Based on preset view features, the view type is identified; wherein, the view type includes plan view, elevation view, and section view; Matching criteria based on consistency of axis position, numbering and labeling, and geometric dimensions allow for cross-view matching of the same entity; Based on a pre-defined consistency rule base, it is determined whether the representation of the same entity in different views conforms to the consistency rules. If it does not conform, a biased representation is identified and marked.

5. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 1, characterized in that, The graph agent analyzes geometric structures and spatial relationships, understands the spatial layout of building components, and calculates engineering quantities and dimensional parameters based on architectural drawing standards and structural mechanics knowledge. The text agent, based on an engineering terminology dictionary and a contract template library, understands text annotations and technical specifications, extracts constraints on schedule, quality, and cost, and parses contract terms and material specifications. The specification review agent retrieves applicable technical specifications based on national standards, industry standards, and local regulations, performs compliance checks, and identifies conflicts between designs and specifications. The historical experience agent retrieves similar historical projects based on the interactive memory field, provides successful case references, and warns of potential risks.

6. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 5, characterized in that, The multi-agent system sequentially performs parallel and interactive analysis on the structured semantic representation to generate a consensus scheme, including: Each intelligent agent independently analyzes the structured semantic representation and outputs preliminary conclusions; Each agent fills in the blind spots by asking and answering questions with other agents; the preliminary conclusions output by each agent are verified from multiple perspectives by other agents; if the verification fails, the agent will state its position, present and demonstrate evidence, conduct third-party arbitration, update its position, and so on until a consensus solution is generated.

7. The interactive graphic and text-based intelligent evolution method for generating construction progress according to claim 1, characterized in that, The method further includes: The project metadata involved in generating the construction schedule plan, as well as the interaction records of the intelligent agent, decision nodes, and preliminary knowledge extraction results, are stored in the memory pool. When the target project analysis is completed, the construction schedule and its reasoning process are stored in the historical interactive memory; during the execution of the target project, the actual progress and problems of the project are stored in the historical interactive memory; after the target project is completed, the effect evaluation and lessons learned are stored in the historical interactive memory. Based on the two-layer memory architecture of the memory pool and the historical interaction memory bank, the collective intelligence evolution of the intelligent agent is carried out, including: identifying the success pattern of the target project, extracting the failure lessons of the target project, and optimizing or generating the reasoning rules of the intelligent agent based on the success pattern and the failure lessons.

8. An interactive, graphic-text-based, intelligently evolving construction progress generation system, characterized in that, The system includes: The multimodal document collection generation module is used to preprocess the input engineering drawings, construction contract documents, technical specification documents, design specifications and historical project reference materials of the target project to obtain a standardized multimodal document collection. The structured semantic representation generation module is used to perform hierarchical reading of standardized engineering drawings based on a hierarchical structure perception mechanism. It identifies the macro layout, main drawing layer, and layer components from macro to micro perspectives and extracts the detailed features of the components. It establishes the correspondence between graphic elements and text annotations, wherein the text annotations are derived from engineering drawings and text documents. It uses the geometric constraint relationship between multiple views to cross-validate and correct cross-view entities and outputs the structured semantic representation of the engineering drawings. The consensus scheme generation module is used to simultaneously input the structured semantic representation into graph agents, text agents, normative review agents, and historical experience agents deployed in the same shared environment. The multiple agents perform parallel analysis and interactive analysis on the structured semantic representation in turn to generate a consensus scheme. The construction schedule generation module is used to decompose tasks step by step based on the consensus scheme, estimate the construction period, establish task dependencies, optimize resource allocation, and generate the construction schedule for the target project.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the interactive graphic-text joint intelligent evolution construction progress generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the interactive graphic-text combined intelligent evolution construction progress generation method as described in any one of claims 1 to 7.

Images