An engineering intelligent decision-making method and system driven by a large model and a dual-core engine.
By using a large model and a dual-core engine, the system automates the parsing and specification matching of engineering drawings, solving the problem of low efficiency in manual review. It enables intelligent compliance review and risk warning of engineering drawings, dynamically updates the experience base, and improves review efficiency and decision-making accuracy.
Patent Information
- Application Number
- CN202611415937.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-11
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, compliance review of architectural engineering drawings relies on manual processes, which is inefficient and makes it difficult to identify conflict points and subtle loopholes across regulations. Furthermore, general-purpose language models lack spatial logical understanding in the engineering field, leading to unreliable decisions.
We adopt a method driven by a large model and a dual-core engine. By acquiring multimodal data and performing cross-disciplinary semantic parsing, we construct a dynamic topological knowledge graph of the engineering domain. We then schedule a static specification engine and a dynamic experience engine in parallel to make collaborative decisions, generate decision results, and obtain human feedback data to update the experience base.
It has achieved automation and intelligence in the compliance review of engineering drawings, shortened the review time to minutes, reduced manpower input, ensured that the decision results are legal, compliant and in line with the actual situation of the project, dynamically identified potential risks, and broken through the limitations of traditional rules.
Smart Images

Figure CN123115486A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of building engineering technology, and in particular to an engineering intelligent decision-making method and system based on a large model and a dual-core engine. Background Technology
[0002] In the field of architecture and industrial engineering, from the compliance review of design drawings and the preparation of material and equipment lists to the prediction of potential risks during construction, there has long been a heavy reliance on manual experience or traditional rule-matching systems. This situation has led to many insurmountable technical bottlenecks.
[0003] Traditional manual review methods suffer from inherent problems such as inefficiency, inconsistent standards, and susceptibility to omissions. The number of drawing pages that can be processed daily and the depth of review are limited, and the varying experience levels and judgment standards among engineers lead to inconsistent review quality. For large and complex projects, the massive amount of drawings and documents makes manual review cycles lengthy, severely hindering project progress. More importantly, manual review struggles to cover all regulatory items, especially in the context of interwoven industry, local, and other multi-level standards, where cross-standard conflicts and subtle loopholes are easily overlooked.
[0004] While general-purpose large language models demonstrate powerful capabilities in natural language processing, they encounter serious illusion problems in specialized engineering fields. This is because these models lack the ability to perceive and understand the complex spatial geometric topology, physical assembly logic, and cross-disciplinary collision constraints within engineering drawings. When faced with highly dense engineering entity data with precise coordinates and attributes, the models are prone to making erroneous inferences that contradict physical laws and engineering logic, rendering their outputs completely unreliable and unusable in the engineering field. Engineering companies have accumulated massive amounts of hidden data over decades of operation, including historical tenders, design schemes, construction logs, change orders, and troubleshooting records. These unstructured text and graphic fragments contain valuable, practice-tested tacit knowledge and experience. However, due to a lack of effective technical means, these data assets have long remained fragmented and idle, unable to be systematically activated and reused. This leads to the repeated occurrence of the same design flaws and rework failures in new projects, resulting in significant resource waste and economic losses.
[0005] In summary, there is an urgent need for an intelligent system that can understand the spatial logic of engineering, integrate static specifications with dynamic experience, and possess proactive decision-making and self-evolution capabilities, in order to fundamentally change the backward situation in the engineering industry that relies on manual and mechanical rules. Summary of the Invention
[0006] This invention provides an engineering intelligent decision-making method and system based on a large model and a dual-core engine, which solves the problem that compliance review of engineering drawings in the prior art relies on manual labor and is inefficient.
[0007] This invention provides an engineering intelligent decision-making method based on a large model and a dual-core engine, comprising: S1, acquire multimodal data of the project to be processed, the multimodal data including multidisciplinary engineering drawings, standard specification documents and historical project data; S2, perform cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; S3. Based on the structured engineering entity data, construct a dynamic topologically related engineering domain knowledge graph; S4, input the engineering domain knowledge graph into the large model, and use the large model to schedule the static specification engine and the dynamic experience engine in parallel to make collaborative decisions and generate decision results; S5, obtain human feedback data on the decision results, associate the feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0008] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided, which performs cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data, including: Layer stripping is performed on the multi-disciplinary engineering drawings to identify and extract the equipment and pipeline elements corresponding to each discipline. Extract the attribute information of each graphic element, which includes at least geometric coordinates, device type, and spatial containment relationship; Unstructured text data is transformed into structured data to obtain text entities associated with primitives.
[0009] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided, wherein the construction of a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data includes: The equipment, pipelines, and building components in the structured engineering entity data are used as nodes in the knowledge graph; Establish edges between nodes based on the physical connection relationships, logical constraint relationships, and spatial containment relationships between entities; Generate a dynamic topological knowledge graph that includes node attributes and edge attributes.
[0010] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided. The method involves inputting the engineering domain knowledge graph into a large model, and then using the large model to perform collaborative decision-making through parallel scheduling of a static specification engine and a dynamic experience engine to generate decision results. The static specification engine performs rule matching based on a pre-built structured specification library, while the dynamic experience engine performs graph retrieval matching based on a historical project experience library. The large model sends query requests to the static specification engine and the dynamic experience engine respectively through a function call mechanism, and receives the returned first decision information and second decision information.
[0011] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided, wherein the engineering domain knowledge graph is input into the large model, and the large model performs collaborative decision-making by scheduling a static specification engine and a dynamic experience engine in parallel to generate a decision result, and further includes: When the first decision information conflicts with the second decision information, the large model calls a preset conflict decision algorithm to calculate a comprehensive risk value based on engineering logic weighting factors, and determines the final decision result based on the comprehensive risk value.
[0012] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided, wherein the engineering logic weighting factor includes at least one of standard mandatory level weight, historical failure rate weight, and project scenario matching degree weight. The conflict decision algorithm determines whether to adopt the decision of the static normative engine, the decision of the dynamic experience engine, or a compromise between the two by comparing the weighted summation result of the weighted factors with a preset threshold.
[0013] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided. The method involves inputting the engineering domain knowledge graph into a large model, and then using the large model to perform collaborative decision-making through parallel scheduling of a static specification engine and a dynamic experience engine to generate a decision result. The method then includes: Obtain human feedback data on the correction of the decision results; The corrected feedback data is associated with the knowledge graph subgraph of the current project to generate new experience cases; The new experience cases are then injected back into the knowledge base of the dynamic experience engine to update the dynamic experience engine.
[0014] According to the present invention, an engineering intelligent decision-making method based on a large model and a dual-core engine is provided. The method involves extracting features from the new experience case using a graph embedding algorithm to generate a feature vector, and storing the feature vector and the new experience case together in the index library of the dynamic experience engine for subsequent graph retrieval and matching.
[0015] This invention also provides an engineering intelligent decision-making system based on a large model and a dual-core engine, comprising: An engineering intelligent decision-making system based on a large model and a dual-core engine, characterized in that the system includes: The data acquisition module is used to acquire multimodal data of the project to be processed, including multidisciplinary engineering drawings, standard and specification documents, and historical project data. The semantic parsing module is used to perform cross-disciplinary semantic parsing on the multimodal data and extract structured engineering entity data; The knowledge graph construction module is used to construct a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data. The dual-engine decision module is used to input the engineering domain knowledge graph into the large model, and through the large model, the static specification engine and the dynamic experience engine are scheduled in parallel to make collaborative decisions and generate decision results. The feedback learning module is used to obtain human correction feedback data on the decision results, associate the correction feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the engineering intelligent decision-making method based on a large model and a dual-core engine as described above.
[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the engineering intelligent decision-making method based on a large model and a dual-core engine as described above.
[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the engineering intelligent decision-making method based on a large model and a dual-core engine as described above.
[0019] This invention provides an engineering intelligent decision-making method and system driven by a large model and a dual-core engine. Through multimodal data acquisition, cross-disciplinary semantic parsing, knowledge graph construction, and collaborative decision-making by the large model and dual engines, it automates and intelligently reviews engineering drawings for compliance. It automatically completes the entire process of drawing parsing, specification matching, and risk warning, reducing manual review from several days to minutes, significantly lowering manpower and time costs. By constructing a knowledge graph reflecting the topology and logical constraints of the engineering space, it constrains the output of the large model, overcoming the illusion problem of general large models. Simultaneously, the collaborative decision-making of the static specification engine and the dynamic experience engine takes into account both specifications and historical experience, ensuring that the decision results are both legal and compliant and aligned with engineering realities, avoiding the one-sidedness of relying solely on rules or experience. Based on historical project data, the dynamic experience engine can identify potential risks in advance, breaking through the limitations of traditional rules and achieving a technological leap from passive inspection to proactive prevention. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0021] Figure 1 This is a flowchart illustrating the engineering intelligent decision-making method based on a large model and a dual-core engine provided by the present invention. Figure 2 This is a schematic diagram of the module connection of the engineering intelligent decision-making system based on a large model and a dual-core engine provided by the present invention. Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention.
[0022] Figure label: 110: Data Acquisition Module; 120: Semantic Parsing Module; 130: Knowledge Graph Construction Module; 140: Dual-Engine Decision Module; 150: Feedback Learning Module; 310: Processor; 320: Communication interface; 330: Memory; 340: Communication bus. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0024] The following is combined Figure 1 The present invention describes an engineering intelligent decision-making method based on a large model and a dual-core engine, comprising: S1, acquire multimodal data of the project to be processed, the multimodal data including multidisciplinary engineering drawings, standard specification documents and historical project data; S2, perform cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; S3. Based on the structured engineering entity data, construct a dynamic topologically related engineering domain knowledge graph; S4, input the engineering domain knowledge graph into the large model, and use the large model to schedule the static specification engine and the dynamic experience engine in parallel to make collaborative decisions and generate decision results; S5, obtain human feedback data on the decision results, associate the feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0025] In one embodiment, during data acquisition, the input file undergoes checks for format, completeness, layer / legend recognizability, coordinate reference, and object association integrity. For drawings, documents, or historical data that do not meet the automatic parsing conditions, the system outputs missing items, information to be supplemented, or prompts for manual confirmation, while retaining source evidence for the parsed portions. This input quality control does not limit the applicability of this invention to different engineering disciplines, different CAD formats, or different project data formats.
[0026] Specifically, in the process of acquiring multimodal data of the project to be processed, compressed packages uploaded by users are received through a unified API interface, including: Multidisciplinary engineering drawings, including low-voltage and intelligent building CAD drawings in DWG format, including integrated cabling system diagrams, security monitoring floor plans, building automation system diagrams, access control system diagrams, etc.; at the same time, the building structure base map and the integrated electromechanical pipeline diagram are obtained for subsequent cross-disciplinary clash analysis; Standard and specification documents include the national mandatory standard GB 50314-2015 Intelligent Building Design Standard, the industry standard JGJ 16-2008 Code for Electrical Design of Civil Buildings, and the local standard DB11 / 687-2015 Energy-Saving Design Standard for Public Buildings. Historical project data, selected from the company's internal knowledge base, includes data on commercial complexes of similar scale from the past five years, including low-voltage electrical design schemes, construction organization designs, equipment procurement lists, on-site construction logs, completion acceptance reports, and fault repair ledgers.
[0027] After acquiring the multimodal data, preprocessing is performed. For CAD drawings, the CAD parsing module can read DWG, DXF, or equivalent engineering drawing formats and parse layers, elements, block references, polylines, text annotations, and legend information into a set of structured objects. Then, coordinate normalization is performed based on the drawing frame, floor, coordinate datum, or project-defined reference points to form a traceable engineering entity input.
[0028] For PDF specification documents, methods such as text extraction, layout recognition, rule parsing, and named entity recognition can be used to structurally store clause numbers, main text, scope of application, and mandatory levels. For historical project data, Word, Excel, PDF, image OCR results, or other readable materials can be parsed, and fields such as project name, professional type, fault phenomenon, cause analysis, and solutions can be obtained through classification models, vector retrieval, or rule extraction, and stored in a searchable experience index library.
[0029] Cross-disciplinary semantic parsing is performed on the multimodal data to extract structured engineering entity data, including: Layer stripping is performed on the multi-disciplinary engineering drawings to identify and extract the equipment and pipeline elements corresponding to each discipline. Extract the attribute information of each graphic element, which includes at least geometric coordinates, device type, and spatial containment relationship; Unstructured text data is transformed into structured data to obtain text entities associated with primitives.
[0030] Specifically, layer, legend, and professional context analysis are performed. After loading CAD drawings, the target professional objects are identified by combining layer names, legend text, block attributes, line types, colors, and spatial locations. Objects with clear attribution are automatically processed. Objects with mixed layers, missing legends, conflicting object names, broken topologies, or non-unique professional attributions retain source evidence and are marked as pending confirmation, rather than substituting speculative results for engineering facts. For cross-professional drawings, pipelines and equipment of different professions can be separated by combining spatial bounding boxes, line types, colors, legends, and object attributes.
[0031] For the activated low-voltage layer, the system performs entity recognition and attribute extraction, and performs the following operations: The system iterates through all block references in the graphical elements and compares them with the built-in device element library, which contains over 500 common low-voltage equipment blocks, such as network cameras, access points (APs), data sockets, and cabinets, performing shape matching and attribute comparison. Upon successful matching, it extracts each block's unique ID, type name, 3D coordinates (X, Y, Z) of the insertion point, rotation angle, and block attributes, such as camera model, IP address, and custom fields for installation height. Cable paths are extracted, multi-segment lines representing the cables are identified, vertex coordinate sequences are parsed, and line type names are recorded. Using a spatial nearest neighbor algorithm, cable endpoints are associated with the nearest device blocks to form connections. The Euclidean distance and azimuth angle between devices are calculated, and it is determined whether there is spatial overlap or insufficient spacing.
[0032] The structured text is processed to create a structured data set. For standard documents and historical data, a pre-defined large language model is used to fine-tune the extraction model, extracting triples (entity 1, relation, entity 2) from paragraphs. For example, from the instruction that the horizontal cable length from the information point to the patch panel should not exceed 90 meters, the following can be extracted: information point, distance constraint, patch panel, maximum length 90 meters. The system outputs a structured intermediate data set containing approximately 1500 entities and 4000 relations, with each entity carrying precise geometric coordinates and attribute labels.
[0033] Based on the structured engineering entity data, a dynamic topological association-based engineering domain knowledge graph is constructed, including: The equipment, pipelines, and building components in the structured engineering entity data are used as nodes in the knowledge graph; Establish edges between nodes based on the physical connection relationships, logical constraint relationships, and spatial containment relationships between entities; Generate a dynamic topological knowledge graph that includes node attributes and edge attributes.
[0034] Specifically, an engineering knowledge graph can be constructed using graph databases, relational databases with graph indexes, in-memory topology graphs, or combinations thereof. Equipment, pipelines, and building components can serve as nodes, with node attributes including ID, type, model, coordinates, floor, and subsystem. Building components can also store attributes such as material, thickness, and elevation.
[0035] The edge definition includes: physical connection edges, such as a network cable connecting a camera and a switch, with attributes including cable type, length, and routing path; logical constraint edges, such as the distance constraint between strong and weak current lines, with attributes including minimum distance value and consequences of violation; spatial inclusion edges, such as a camera mounted on a wall, with attributes including installation method and load-bearing information; and similarity association edges, which establish similarity edges between the current device and similar devices in historical failure cases based on historical project experience, with attributes including similarity score.
[0036] Once the topology map is constructed, the system automatically performs topology verification: checking whether devices have verifiable physical connection edges, whether cable lengths meet constraints, whether routes cross unreachable areas, and whether there are evidence conflicts between legends / devices / cables. For isolated nodes, excessively long cables, objects lacking input, or objects that cannot be uniquely identified, the system retains the original source, diagnostic reasons, and candidate results, and marks them as pending verification, supporting manual confirmation before continuing calculation or review.
[0037] The engineering domain knowledge graph is input into a large model, and the large model uses a static specification engine and a dynamic experience engine in parallel to make collaborative decisions, generating decision results, including: The static specification engine performs rule matching based on a pre-built structured specification library, while the dynamic experience engine performs graph retrieval matching based on a historical project experience library. The large model sends query requests to the static specification engine and the dynamic experience engine respectively through a function call mechanism, and receives the returned first decision information and second decision information.
[0038] When the first decision information conflicts with the second decision information, the large model calls a preset conflict decision algorithm to calculate a comprehensive risk value based on engineering logic weighting factors, and determines the final decision result based on the comprehensive risk value.
[0039] The engineering logic weighting factor includes at least one of the following: standard mandatory level weight, historical failure rate weight, and project scenario matching degree weight; The conflict decision algorithm determines whether to adopt the decision of the static specification engine or the decision of the dynamic experience engine by comparing the weighted summation result of the weighted factors with a preset threshold.
[0040] Specifically, the subgraphs in the knowledge graph relevant to the current review task are converted into executable queries, rule inputs, or structured contexts. The static specification engine maintains a structured specification library, transforming applicable clauses from national, industry, local, or enterprise standards into executable rules, constraints, or validation functions; each validation result is associated with the clause source, input entity, and triggering reason to return a traceable list of violations or items awaiting confirmation.
[0041] The current knowledge graph's subgraphs, including features such as device type, spatial layout, and project type, are embedded using the GraphSAGE algorithm to obtain a 128-dimensional feature vector. Then, an approximate nearest neighbor search is performed on the historical project experience database of the dynamic experience engine to find the top five similar historical project subgraphs. For each matching historical project, its associated fault cases, rework records, and optimization suggestions are extracted.
[0042] After receiving the outputs from both engines, the large model executes a conflict decision algorithm. It compares the outputs of the two engines to identify any contradictions. For example, the static engine might not detect any violations in camera installation height, but the dynamic engine might issue a high-risk warning. A comprehensive risk score is then calculated based on a pre-defined engineering logic weighted algorithm. Standard mandatory level weights: National Mandatory = 0.5, Industry Recommended = 0.3, Enterprise Experience = 0.2. Historical failure rate weights: Failure rate ≥ 80% = 0.9, 50%~80% = 0.6, < 50% = 0.3. Project scenario matching weights: Completely identical = 1.0, Highly similar = 0.8, Generally similar = 0.5.
[0043] In a specific example, the dynamic engine can generate risk factors based on historical failure rates, scenario matching, evidence completeness, and rule enforcement levels, and calculate a comprehensive risk score according to preset or configurable weights. When the comprehensive risk score reaches the threshold corresponding to the current project, the system outputs a high-risk warning; when evidence is insufficient or multiple candidate conclusions cannot be distinguished, the system outputs a conclusion requiring manual confirmation, rather than directly generating a deterministic engineering decision.
[0044] Based on the comprehensive risk score, the decision was made to adopt the dynamic engine's recommendations, generate a high-risk early warning decision, and include details of historical cases and recommended correction schemes.
[0045] After generating the decision result, obtain the correction feedback data of the decision result from human intervention; associate the correction feedback data with the knowledge graph subgraph of the current project to generate new experience cases; and inject the new experience cases back into the knowledge base of the dynamic experience engine to update the dynamic experience engine.
[0046] After receiving the risk alert, the project's chief engineer, following a site survey, determined that the space within the suspended ceiling was sufficient and decided not to modify the design. He then submitted his rejection comments and photographic evidence to the system. The system automatically captured this revised data and implemented self-learning through the following process: Data encapsulation involves packaging the current project's knowledge graph subgraph, triggering conditions (camera installation height 3.0 meters), the dynamic engine's original warnings, manual rejection reasons, and evidence links into a structured case object.
[0047] Feature extraction was performed on this case using the same GraphSAGE model, resulting in a 128-dimensional vector.
[0048] The inbound index stores the case objects and their embedding vectors into the rejected case partition of the dynamic experience base and creates an inverted index.
[0049] Weight Adjustment: The system automatically reduces the default risk weight of this trigger condition to avoid over-warning in similar scenarios in the future.
[0050] When the large model receives a compliance conclusion from the static engine and a high-risk warning from the dynamic engine, it triggers a conflict decision-making algorithm. The algorithm first extracts the conflict points, then queries the knowledge graph to obtain relevant attributes, and then requests the decision-making basis from both engines (static engine: no violation of regulations; dynamic engine: historical failure rate 80%, scenario matching degree 95%). The algorithm calculates the comprehensive risk value according to a preset weighted formula. If it exceeds the threshold, the decision is to adopt the dynamic experience, and a report including a conflict explanation, the basis for both sides, and the final recommendation is output.
[0051] Manual feedback corrections can include actions such as confirmation, rejection, supplementary information, adjustment of entity mapping, correction of rule scope, or marking experience cases as invalid. Before writing the feedback into the dynamic experience engine, the system records the feedback subject, project context, original evidence, version number, and applicable conditions, making experience updates traceable and reversible, and preventing unconfirmed model inferences from being directly solidified into engineering experience.
[0052] When the engineer received the risk warning from the system, they deemed it overly conservative and decided not to modify the design. They submitted a rejection statement and reasons in the system: on-site surveys showed sufficient space within the ceiling, eliminating the need to lower the installation height. The system automatically captured this revised data and structured it along with the current project's knowledge graph subgraph and triggering conditions. Then, through a reverse injection interface, this case was written as a new rule of experience into the dynamic experience engine's knowledge base. This case was marked as a rejected case, along with the reasons for rejection. In the future, when similar projects trigger the same risk point again, the dynamic experience engine will simultaneously return the original high-risk case and this rejected case during retrieval, providing the larger model with a more comprehensive decision-making reference, thus achieving the dynamic evolution of system knowledge.
[0053] Based on the engineering intelligent decision-making method driven by a large model and a dual-core engine provided by this invention, the compliance review of engineering drawings is automated and intelligent through multimodal data acquisition, cross-disciplinary semantic parsing, knowledge graph construction, and collaborative decision-making by the dual engines of the large model. It automatically completes the entire process of drawing parsing, specification matching, and risk warning, reducing the manual review time from several days to minutes, significantly reducing manpower and time costs. By constructing a knowledge graph reflecting the topology and logical constraints of the engineering space, the output of the large model is constrained, overcoming the illusion problem of general large models. Simultaneously, the collaborative decision-making of the static specification engine and the dynamic experience engine takes into account both specifications and historical experience, ensuring that the decision results are both legal and compliant and aligned with engineering realities, avoiding the one-sidedness of relying solely on rules or experience. Based on historical project data, the dynamic experience engine can identify potential risks in advance, breaking through the limitations of traditional rules and achieving a technological leap from passive inspection to proactive prevention.
[0054] refer to Figure 2The present invention also discloses an engineering intelligent decision-making system based on a large model and a dual-core engine, the system comprising: The data acquisition module 110 is used to acquire multimodal data of the project to be processed, including multi-disciplinary engineering drawings, standard specification documents and historical project data; The semantic parsing module 120 is used to perform cross-disciplinary semantic parsing on the multimodal data and extract structured engineering entity data; The knowledge graph construction module 130 is used to construct a dynamic topologically related engineering domain knowledge graph based on the structured engineering entity data. The dual-engine decision module 140 is used to input the engineering domain knowledge graph into the large model, and to coordinate the decision-making of the static specification engine and the dynamic experience engine through the parallel scheduling of the large model to generate decision results. The feedback learning module 150 is used to obtain human correction feedback data on the decision results, associate the correction feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0055] In one implementation, the dual-engine decision module performs a structured comparison of the outputs of the static normative engine and the dynamic experience engine, and generates conflict decision results based on the rule enforcement level, historical case credibility, scenario matching degree, evidence completeness and manual confirmation status; when any key evidence is missing or there is an irresolvable conflict, the system generates items to be manually confirmed and retains the original evidence of both types of engines.
[0056] Specifically, cross-disciplinary semantic parsing is performed on the multimodal data to extract structured engineering entity data, including: Layer stripping is performed on the multi-disciplinary engineering drawings to identify and extract the equipment and pipeline elements corresponding to each discipline. Extract the attribute information of each graphic element, which includes at least geometric coordinates, device type, and spatial containment relationship; Unstructured text data is transformed into structured data to obtain text entities associated with primitives.
[0057] Based on the structured engineering entity data, a dynamic topological association-based engineering domain knowledge graph is constructed, including: The equipment, pipelines, and building components in the structured engineering entity data are used as nodes in the knowledge graph; Establish edges between nodes based on the physical connection relationships, logical constraint relationships, and spatial containment relationships between entities; Generate a dynamic topological knowledge graph that includes node attributes and edge attributes.
[0058] The engineering domain knowledge graph is input into a large model, and the large model uses a static specification engine and a dynamic experience engine in parallel to make collaborative decisions, generating decision results, including: The static specification engine performs rule matching based on a pre-built structured specification library, while the dynamic experience engine performs graph retrieval matching based on a historical project experience library. The large model sends query requests to the static specification engine and the dynamic experience engine respectively through a function call mechanism, and receives the returned first decision information and second decision information.
[0059] When the first decision information conflicts with the second decision information, the large model calls a preset conflict decision algorithm to calculate a comprehensive risk value based on engineering logic weighting factors, and determines the final decision result based on the comprehensive risk value.
[0060] The engineering logic weighting factor includes at least one of the following: standard mandatory level weight, historical failure rate weight, and project scenario matching degree weight; The conflict decision algorithm determines whether to adopt the decision of the static specification engine or the decision of the dynamic experience engine by comparing the weighted summation result of the weighted factors with a preset threshold.
[0061] The engineering domain knowledge graph is input into a large model, and the large model uses a static specification engine and a dynamic experience engine in parallel to perform collaborative decision-making, generating decision results, which then include: Obtain human feedback data on the correction of the decision results; The corrected feedback data is associated with the knowledge graph subgraph of the current project to generate new experience cases; The new experience cases are then injected back into the knowledge base of the dynamic experience engine to update the dynamic experience engine.
[0062] The new experience case is subjected to feature extraction using a graph embedding algorithm to generate a feature vector. The feature vector and the new experience case are then stored together in the index of the dynamic experience engine for subsequent graph retrieval and matching.
[0063] Based on the engineering intelligent decision-making system driven by a large model and a dual-core engine provided by this invention, the system automates and intelligently reviews engineering drawings for compliance through multimodal data acquisition, cross-disciplinary semantic parsing, knowledge graph construction, and collaborative decision-making by the large model and dual engines. It automatically completes the entire process of drawing parsing, specification matching, and risk warning, reducing manual review from several days to minutes, significantly lowering manpower and time costs. By constructing a knowledge graph reflecting the topology and logical constraints of the engineering space, the system constrains the output of the large model, overcoming the illusion problem of general large models. Simultaneously, the collaborative decision-making of the static specification engine and the dynamic experience engine takes into account both specifications and historical experience, ensuring that the decision results are both legal and compliant and aligned with engineering realities, avoiding the one-sidedness of relying solely on rules or experience. The dynamic experience engine, based on historical project data, can identify potential risks in advance, breaking through the limitations of traditional rules and achieving a technological leap from passive inspection to proactive prevention.
[0064] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute an engineering intelligent decision-making method driven by a large model and a dual-core engine. This method includes: acquiring multimodal data of the engineering project to be processed, including multi-disciplinary engineering drawings, standard specification documents, and historical project data; performing cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; constructing a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data; inputting the engineering domain knowledge graph into the large model, and using the large model to schedule a static specification engine and a dynamic experience engine in parallel for collaborative decision-making to generate decision results; acquiring manual feedback data on the decision results; associating the feedback data with the current project's engineering knowledge graph subgraph, original evidence, version information, and applicable conditions to generate traceable experience cases; and updating the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0065] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part 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 the present invention. 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.
[0066] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the engineering intelligent decision-making method based on a large model and a dual-core engine provided by the above methods. The method includes: acquiring multimodal data of the engineering project to be processed, the multimodal data including multi-disciplinary engineering drawings, standard specification documents, and historical project data; performing cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; constructing a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data; inputting the engineering domain knowledge graph into a large model, and performing collaborative decision-making through parallel scheduling of a static specification engine and a dynamic experience engine by the large model to generate decision results; acquiring human correction feedback data on the decision results, associating the correction feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information, and applicable conditions to generate traceable experience cases, and updating the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0067] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the engineering intelligent decision-making method based on a large model and a dual-core engine provided by the above methods. The method includes: acquiring multimodal data of an engineering project to be processed, the multimodal data including multi-disciplinary engineering drawings, standard specification documents, and historical project data; performing cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; constructing a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data; inputting the engineering domain knowledge graph into a large model, and performing collaborative decision-making through parallel scheduling of a static specification engine and a dynamic experience engine by the large model to generate decision results; acquiring manual correction feedback data on the decision results, associating the correction feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information, and applicable conditions to generate traceable experience cases, and updating the confirmed experience cases to the knowledge base of the dynamic experience engine.
[0068] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0069] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications 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 the present invention.
Claims
1. An engineering intelligent decision-making method based on a large model and a dual-core engine, characterized in that, include: S1, acquire multimodal data of the project to be processed, the multimodal data including multidisciplinary engineering drawings, standard specification documents and historical project data; S2, perform cross-disciplinary semantic parsing on the multimodal data to extract structured engineering entity data; S3. Based on the structured engineering entity data, construct a dynamic topologically related engineering domain knowledge graph; S4, input the engineering domain knowledge graph into the large model, and use the large model to schedule the static specification engine and the dynamic experience engine in parallel to make collaborative decisions and generate decision results; S5, obtain human feedback data on the decision results, associate the feedback data with the current project's engineering knowledge graph subgraph, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
2. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 1, characterized in that, Cross-disciplinary semantic parsing is performed on the multimodal data to extract structured engineering entity data, including: Layer stripping is performed on the multi-disciplinary engineering drawings to identify and extract the equipment and pipeline elements corresponding to each discipline. Extract the attribute information of each graphic element, which includes at least geometric coordinates, device type, and spatial containment relationship; Unstructured text data is transformed into structured data to obtain text entities associated with primitives.
3. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 1, characterized in that, The construction of a dynamic topologically related engineering domain knowledge graph based on the structured engineering entity data includes: The equipment, pipelines, and building components in the structured engineering entity data are used as nodes in the knowledge graph; Establish edges between nodes based on the physical connection relationships, logical constraint relationships, and spatial containment relationships between entities; Generate a dynamic topological knowledge graph that includes node attributes and edge attributes.
4. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 1, characterized in that, The process involves inputting the engineering domain knowledge graph into a large model, and then using the large model to schedule a static specification engine and a dynamic experience engine in parallel for collaborative decision-making to generate decision results, including: The static specification engine performs rule matching based on a pre-built structured specification library, while the dynamic experience engine performs graph retrieval matching based on a historical project experience library. The large model sends query requests to the static specification engine and the dynamic experience engine respectively through a function call mechanism, and receives the returned first decision information and second decision information.
5. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 4, characterized in that, The step of inputting the engineering domain knowledge graph into a large model, and using the large model to schedule the static specification engine and the dynamic experience engine in parallel for collaborative decision-making to generate decision results, also includes: When the first decision information conflicts with the second decision information, the large model calls a preset conflict decision algorithm to calculate a comprehensive risk value based on engineering logic weighting factors, and determines the final decision result based on the comprehensive risk value.
6. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 5, characterized in that, The engineering logic weighting factor includes at least one of the following: standard mandatory level weight, historical failure rate weight, and project scenario matching degree weight; The conflict decision algorithm determines whether to adopt the decision of the static specification engine or the decision of the dynamic experience engine by comparing the weighted summation result of the weighted factors with a preset threshold.
7. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 1, characterized in that, The process involves inputting the engineering domain knowledge graph into a large model, using the large model to schedule a static specification engine and a dynamic experience engine in parallel for collaborative decision-making, generating decision results, and then including: Obtain human feedback data on the correction of the decision results; The corrected feedback data is associated with the knowledge graph subgraph of the current project to generate new experience cases; The new experience cases are then injected back into the knowledge base of the dynamic experience engine to update the dynamic experience engine.
8. The engineering intelligent decision-making method based on a large model and a dual-core engine as described in claim 7, characterized in that, The new experience case is subjected to feature extraction using a graph embedding algorithm to generate a feature vector, and the feature vector and the new experience case are stored together in the index library of the dynamic experience engine for subsequent graph retrieval and matching.
9. An engineering intelligent decision-making system based on a large model and a dual-core engine, characterized in that, The system includes: The data acquisition module is used to acquire multimodal data of the project to be processed, including multidisciplinary engineering drawings, standard and specification documents, and historical project data. The semantic parsing module is used to perform cross-disciplinary semantic parsing on the multimodal data and extract structured engineering entity data; The knowledge graph construction module is used to construct a dynamically topologically related engineering domain knowledge graph based on the structured engineering entity data. The dual-engine decision module is used to input the engineering domain knowledge graph into the large model, and through the large model, the static specification engine and the dynamic experience engine are scheduled in parallel to make collaborative decisions and generate decision results. The feedback learning module is used to obtain human correction feedback data on the decision results, associate the correction feedback data with the engineering knowledge graph subgraph of the current project, original evidence, version information and applicable conditions, generate traceable experience cases, and update the confirmed experience cases to the knowledge base of the dynamic experience engine.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the engineering intelligent decision-making method based on a large model and a dual-core engine as described in any one of claims 1 to 8.
