Archive room fire-fighting design method and device based on large language model and electronic equipment

Abstract

The invention provides an archive fire-fighting design method and device based on a large language model and electronic equipment, and relates to the technical field of fire-fighting engineering. The method comprises the following steps: analyzing an archive drawing to obtain structured node features; analyzing the standard articles to form a graph database; adopting a large language model to match the structured node features with regulation articles in a graph database to obtain target article semantics; obtaining a fire-fighting design scheme according to the target article semantics and the structured node features; performing compliance check on the fire-fighting design scheme; and if the fire-fighting design scheme is not compliant, modifying the fire-fighting design scheme by adopting a large language model, and skipping to the step of performing compliance check on the fire-fighting design scheme to continue execution until the fire-fighting design scheme is compliant. The fire-fighting design scheme is automatically generated by adopting the large language model, the design efficiency is high, the compliance accuracy is high, the knowledge dependence is low, and the method is suitable for complex scenes.

Description

Technical Field

[0001] This invention relates to the field of fire protection engineering technology, and in particular to a fire protection design method, device and electronic equipment for archives based on a large language model. Background Technology

[0002] As the core location for storing important documents, historical materials, and confidential files, archives contain documents that are often non-renewable—from crucial documents for business decisions to core experimental data from research institutions. If destroyed by fire, they will cause irreparable historical losses and economic costs. Therefore, the fire protection design of archives must not only meet basic fire prevention requirements but also achieve safety and regulatory standards far exceeding those of ordinary buildings, focusing on four core objectives: fire prevention, early warning, rapid fire control, and minimal damage. For example, it must strictly control fire source risks, ensure that the fire extinguishing system does not cause secondary damage to the archives, and guarantee absolutely unobstructed evacuation routes.

[0003] In current technology, fire protection design for archives mainly relies on manual review of fire protection regulations, with designers manually completing the design and review process based on experience. This method suffers from high knowledge dependence, low efficiency, and susceptibility to errors, easily leading to safety hazards due to human negligence. Furthermore, as archives expand in size and become more complex, the inefficiencies and quality risks of this traditional manual design model become increasingly prominent. The traditional manual design model is ill-suited to the demands of complex projects for professionalism, complexity, and efficiency. Summary of the Invention

[0004] This invention provides a fire protection design method, device, and electronic equipment for archives based on a large language model, in order to solve the problem that the traditional manual design mode in the prior art cannot meet the current fire protection design requirements.

[0005] In a first aspect, embodiments of the present invention provide a fire protection design method for archives based on a large language model, comprising: The archive drawings were analyzed to obtain structured node features; The standard provisions are parsed to form a graph database; A large language model is used to match structured node features with legal provisions in a graph database to obtain the semantics of the target provisions. Based on the semantics of the target clauses and the characteristics of the structured nodes, a fire protection design scheme is obtained; Conduct compliance checks on fire protection design plans; If the fire protection design scheme is not compliant, the large language model will be used to modify the fire protection design scheme, and the process will continue to the step of checking the compliance of the fire protection design scheme until the fire protection design scheme is compliant.

[0006] Secondly, embodiments of the present invention provide a fire protection design device for archives based on a large language model, comprising: The first data extraction module is used to parse the archive drawings and obtain structured node features; The second data extraction module is used to parse the standard provisions and form a graph database; The legal provision extraction module is used to match the structured node features with the legal provisions in the graph database using a large language model to obtain the semantics of the target provisions. The solution output module is used to obtain fire protection design solutions based on the semantics of the target clauses and the characteristics of structured nodes; The compliance check module is used to check the compliance of fire protection design schemes; The scheme optimization module is used to modify the fire protection design scheme using a large language model if the fire protection design scheme is not compliant, and then jumps to the step of checking the compliance of the fire protection design scheme to continue execution until the fire protection design scheme is compliant.

[0007] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the fire protection design method for archive rooms based on a large language model as described in the first aspect above or any possible implementation of the first aspect.

[0008] This invention provides a method, apparatus, and electronic device for fire protection design of archives based on a large language model. The method includes: parsing archive drawings to obtain structured node features; parsing regulatory provisions to form a graph database; using a large language model to match the structured node features with the regulatory provisions in the graph database to obtain the target provision semantics; obtaining a fire protection design scheme based on the target provision semantics and structured node features; performing a compliance check on the fire protection design scheme; if the fire protection design scheme is non-compliant, modifying the fire protection design scheme using the large language model and then reverting to the compliance check step until the fire protection design scheme is compliant. This application performs structured data parsing of archive drawings and regulatory provisions, performs semantic matching to obtain the target provision semantics, and then dynamically optimizes the structured node features based on the target provision semantics to automatically generate a fire protection design scheme. This reduces knowledge dependence, greatly improves design efficiency, and achieves high compliance accuracy, adapting to complex scenarios and meeting current fire protection design requirements. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the implementation of a fire protection design method for archives based on a large language model, as provided in an embodiment of the present invention. Figure 2This is a structural schematic diagram of the fire protection design device for archives based on a large language model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0010] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] See Figure 1 The document illustrates a flowchart of an implementation of a fire protection design method for archives based on a large language model, as provided in an embodiment of the present invention, detailed below: The above-mentioned fire protection design method for archives based on large language models includes: S101: Analyze the archive drawings to obtain structured node features; Traditional drawings are unstructured and difficult to reuse. This application first analyzes the drawings in the archives to obtain structured node features.

[0012] In one possible implementation, S101 may include: S1011: Preprocess the drawings in the archives to obtain parsable vector graphic data; First, the drawings in the archives are preprocessed to convert the bitmap format drawings into parsable vector graphic data, preserving the geometric and topological relationships of the graphic elements, thus laying the foundation for subsequent analysis.

[0013] For example, the preprocessing specifically includes: format unification and layer filtering, converting the drawings into an intermediate format (AutoCAD DXF format) to obtain parsable vector graphics data, applying regular expression matching and layer naming rules to extract relevant layers from the archive, and removing auxiliary layers in the drawings that do not participate in structural analysis (such as title blocks, text annotations, dimension annotations, etc.).

[0014] S1012: Perform primitive recognition on parsable vector graphic data to obtain primitive recognition results; S1013: Based on the primitive recognition results, the geometric parameters of the drawing are extracted using a geometric analysis engine, and an adjacency graph is constructed; The system can combine rule-based matching with deep learning classification to obtain primitive recognition results (Line, Arc, Polyline, Insert). The geometric parsing engine extracts point coordinates, lengths, and angular relationships for different primitive types, and constructs an adjacency graph based on the spatial relationships between primitives. Primitives are used as nodes, and relationships such as "containment," "adjacent," and "connected" are used as edges to store the topological associations between nodes.

[0015] The set of point coordinates is represented as:

[0016] in, For the set of point coordinates, For the first Two-dimensional plane coordinates of a point This represents the total number of points contained in the primitive. The index of the current point. .

[0017] Length is expressed as:

[0018] in, The total length of the primitive. For the first The length of each primitive.

[0019] The angular relationship is expressed as:

[0020] in, For the first The orientation angle of each primitive.

[0021] The above features are encoded as primitive vectors, represented as follows:

[0022] in, For the first One-hot encoding of each primitive type.

[0023] Based on the above, an adjacency graph is constructed. , represented as:

[0024] in, For a set of primitive vectors, This represents the connection relationship between graphic elements.

[0025] S1014: A graph neural network is used to extract the global semantic structure of the adjacency graph to obtain node feature statements; The node feature statement is represented as:

[0026] in, For the first The first layer The characteristics of each node The weight matrix is ​​a learnable matrix. For the first The set of neighbors of a node. For the first The degree of each node, For the first The degree of each node, This is the ReLU activation function.

[0027] S1015: Perform unified encoding on node feature statements to obtain structured node features.

[0028] To improve the generality of the data, the node feature statements are uniformly encoded into intermediate semantic representations. The Qwen-VL multimodal large language model is used for semantic judgment and naming, and the results are output in JSON format, as follows: { Room attributes: { Area: 118.4 Layer: "ROOM_3", "Closedness": true, "Boundary shape": "rectangle", "Adjacent Rooms": ["Passage 1", "Control Room"], Export quantity: 2 }, Visual cue: ["Contains the text 'archives'"], Spatial Location: Central Region } This application features fully automated processing, which can automatically extract key data such as the geometric structure, file distribution, and fire compartmentation of the archives. This significantly reduces reliance on manual identification, lowers the identification error rate, improves the efficiency of data processing in the early stages of design, greatly reduces manpower input, and is adaptable to complex scenarios.

[0029] S102: Parse the standard provisions to form a graph database; Based on national / industry standards such as the "Code for Fire Protection Design of Buildings" and the "Code for Design of Archives Buildings," the RDF method is used to represent the triple relationship of "functional area - code clause - fire protection measures," for example (warehouse, requires automatic sprinkler system). The obtained triple relationship is converted into vectors using the Embedding model, and the triple vectors are stored in the database to form a graph database.

[0030] S103: Using a large language model, the structured node features are matched with the legal provisions in the graph database to obtain the semantics of the target provisions; To ensure accurate alignment between "design requirements" and "standard requirements" and avoid "mismatch" and "omission", it is necessary to match the structured node features with the legal provisions in the graph database and select the appropriate legal provisions for the current design scenario.

[0031] In one possible implementation, S103 may include: S1031: Construct standardized prompt word templates and generate prompt words based on structured node features; First, a standardized prompt word template is constructed based on prompt word engineering technology. The structured node features are then combined with the matching requirements, and the prompt words are obtained by filling the standardized prompt word template.

[0032] S1032: Input the prompt words into the large language model and retrieve at least one candidate legal provision from the graph database; The generated prompts are input into a large language model, which uses its natural language understanding capabilities to parse the search requirements and execute the query, resulting in at least one candidate legal provision. This process leverages the semantic understanding capabilities of the large language model to overcome the limitations of traditional keyword retrieval and can handle search requirements in complex scenarios such as "dense shelving storage area" and "multi-layer connected database area".

[0033] S1033: For any candidate legal provision, calculate the similarity between the candidate legal provision and the structured node features; if the similarity between the candidate legal provision and the structured node features is greater than a preset value, then the candidate legal provision is taken as the target legal provision semantics. Among them, the number of target clause semantics is at least one.

[0034] This application ensures a high degree of semantic fit between the target clause and the design scenario through similarity metric calculation, avoiding interference from clauses that "seem relevant but are actually inapplicable".

[0035] Specifically, the structured node features and candidate legal provisions can be converted into high-dimensional semantic vectors using the BERT encoder, and the similarity can be calculated using the SimCSE (Semantic Vector Matching) algorithm, as shown in the following formula:

[0036] in, This is a structured node feature representation vector. is the vector representing the candidate legal provisions.

[0037] For example, the preset value can be 0.85. For instance, if the similarity is 0.6, which is lower than the preset value, it indicates low similarity and will be discarded, retaining only applicable legal provisions with high similarity. The preset value is adjustable and can be dynamically adjusted according to actual application needs, taking into account the design requirements of different scenarios.

[0038] This application uses standardized prompt word templates to effectively avoid retrieval bias caused by "incomplete input information". The semantic understanding ability of the large language model solves the limitations of "keyword surface matching". Combined with similarity threshold filtering, it achieves accurate matching of the semantics of the final target text, and has higher retrieval efficiency and adaptability to complex scenarios.

[0039] It should be noted that the semantics of the target clause also need to be uniformly encoded, as shown in the following output: { "entity_id": "ROOM_003", "type": "warehouse", "area": ​​145.2, "norm_compliance": [ { "norm": "GB50016-2014 §8.4.1", "description": "The gas extinguishing fire protection area should be relatively sealed, and doors and windows must not leak." } ] } This application can intelligently recommend applicable standard provisions in different design scenarios and achieve semantic-level matching between standard provisions and drawing elements, effectively avoiding the problems of standard omissions or mismatches.

[0040] S104: Based on the semantics of the target clauses and the characteristics of structured nodes, a fire protection design scheme is obtained; In one possible implementation, S104 may include: S1041: Based on the semantics of the target clause and the characteristics of the structured nodes, determine whether there are any specification conflicts or spatial layout problems, and generate a problem diagnosis report and preliminary optimization strategies; Constrained by the semantics of the target clause, conflicts are identified in advance through problem diagnosis.

[0041] In one possible implementation, S1041 may include: 1. Conduct rule conflict determination, spatial constraint check and risk priority assessment in sequence to determine whether there are any regulatory conflicts or spatial layout issues.

[0042] Specifically, for example, in determining rule conflicts: "The fire compartment area of ​​a building is ≤1500m²". 2 "As a rule constraint, if the area field of the room area in the parsed drawing is greater than 1500m²..." 2 This triggers a "structural adjustment" decision; Space constraint check: If the number of exits is less than 2 or the space closure is false, the "exit optimization" strategy is triggered. Spatial constraint check: Construct a multi-objective evaluation function that comprehensively considers spatial risk coefficient, standard constraint weights and historical optimization data to determine whether to adjust the layout of ventilation, sprinkler and other systems.

[0043] S1042: Input the problem diagnosis report and preliminary optimization strategy into the large language model for design reasoning to generate optimized design parameters; Based on logical conditions and normative constraints, design reasoning is performed using a large language model.

[0044] In one possible implementation, S1042 may include: 1. Use a large language model to perform fire suppression system configuration reasoning, fire separation distance generation, and ventilation system design suggestions to obtain optimized design parameters.

[0045] Specifically, for example, fire suppression system configuration reasoning: if the room attribute is "warehouse" and important files are stored, then the suggestion "Set up a gas fire suppression system, preferably a heptafluoropropane system" will be generated; Fire separation distance generation: If the adjacent room is a "control room", the model will recommend setting up solid partition walls made of non-combustible materials and ensure that the fire separation time is ≥2h; Ventilation system design recommendations: In areas with high levels of enclosure, it is recommended to install both automatic smoke extraction systems and mechanical air supply systems. Ventilation calculations should use the following formula:

[0046] in, Air volume (m³) 3 / s), The heat release rate (kW) This refers to air density.

[0047] S1043: Integrate the optimized design parameters with the preliminary optimization strategy to obtain the fire protection design scheme.

[0048] The solutions are integrated to generate a fire protection design scheme that conforms to industry standards.

[0049] Furthermore, fire protection design schemes can be refined and professionalized, and industry terminology can be reconstructed using a large language model to achieve consistency and standardization.

[0050] This application significantly improves the compliance and feasibility of fire protection design schemes through a process of "problem-oriented - model reasoning - integrated output".

[0051] S105: Conduct a compliance check on the fire protection design plan; To verify the rationality of the plan and ensure that the "design parameters" are quantitatively aligned with the "standard requirements", a compliance check is conducted on the fire protection design plan to ensure that there are no compliance loopholes.

[0052] In one possible implementation, S105 may include: S1051: Construct a design fact graph based on the characteristics of structured nodes; Based on structured node features and fire protection design schemes, a set of design fact triples is formed using a large language model, and a design fact graph containing "entity-relationship-attribute" is constructed:

[0053] in, For physical nodes (such as rooms, equipment, etc.) This refers to attribute relationships (such as having an area of ​​, etc.). These are attribute values ​​(area, quantity, etc.).

[0054] S1052: Construct standard design maps based on the graph database; Based on the legal provisions in the graph database, the "standard entity-standard relationship-standard attribute" required by the standard are extracted to construct the standard design graph.

[0055] S1053: Compare the design fact diagrams with the standard design diagrams to determine whether the fire protection design scheme is compliant.

[0056] The design fact map is compared with the standard design map using prompt word engineering.

[0057] Specifically, in one possible implementation, S1053 may include: 1. Analyze the standard design diagram to obtain relational constraints, numerical constraints, and symbolic logic rules; 2. Determine whether the design fact graph satisfies relational constraints, numerical constraints, and symbolic logic rules; 3. If at least one of the requirements is not met, the fire protection design scheme is deemed non-compliant. 4. If all conditions are met, the fire protection design scheme is deemed compliant.

[0058] This application uses a graph comparison method and constructs a global graph, which can be used for compliance verification in the collaborative design of multi-zone archives, with high verification coverage and accuracy.

[0059] S106: If the fire protection design scheme is not compliant, the fire protection design scheme shall be modified using a large language model, and the process shall continue to the step of checking the compliance of the fire protection design scheme until the fire protection design scheme is compliant.

[0060] For detected non-compliance items, the large language model calls the corresponding prompt word template to generate targeted correction suggestions, and the compliance review results are uniformly coded. The output results are shown below: { "entity_id": "ROOM_003", "entity_type": "archives storage", "compliance_result": [ { "item": "closed", "status": "Compliant", "reference": "GB50016-2014 §8.4.1", "description": "The enclosure meets the design requirements for gas extinguishing fire protection zones." } ] } Constructing a task graph model and initializing the scheduling process: Based on the compliance review results of the previous stage, a task graph model is constructed, formally represented as follows:

[0061] in, This is a set of nodes representing the agent's task (such as "drawing parsing", "semantic matching", etc.). Let be a set of edges, representing the execution dependencies between tasks (e.g., review → correction → re-review). The steps S105~S106 are executed repeatedly to generate and revise the fire protection design scheme. Each time, the dispatcher takes a task from the task queue and calls the corresponding method step to execute it, forming a closed-loop review process, which continues to iterate until all specification requirements are met.

[0062] During method execution, intermediate representations generated are recorded, and semantic information generated during execution is compressed and encoded into vector representations and stored in a vector database. Simultaneously, the initial state of the input design entity, the suggested modifications and execution content for each round, compliance status change records, and the criteria and chain of evidence for successful review are recorded. The memory structure is as follows: { "memory_id": "audit_loop_ROOM_003", "initial_design": {...}, "revision_history": [ { "round": 1, "suggestion": "Add an exit", "updated_by": "design_agent", "result": "Still non-compliant" }, { "round": 2, "suggestion": "Set up a two-way escape door", "updated_by": "design_agent", "result": "Compliance" } ], "final_status": "Fully compliant" } Once all designs have completed compliance review, the system generates the final revised draft and a complete process log, including: the final fire protection design scheme text, comparison results of each round of revisions, a call chain diagram for each step, and a list of review evidence chains and supporting clauses. The output is shown below: { "room_id": "ROOM_003", "final_status": "Compliance", "revision_steps": 2, "fix_summary": [ "Increase export volume to 2", "Equipped with automatic closing two-way escape doors" ], "compliance_basis": [ GB50016-2014 §5.5.10 GB50174-2017 §8.4.3 ] } This application automatically parses drawings and code provisions to generate fire protection design schemes, significantly reducing the workload of repeated manual review, drawing, and modification, and substantially improving overall design efficiency and consistency. Simultaneously, it performs semantic consistency comparison between the generated fire protection design schemes and drawings, identifying potential conflicts, omissions, or design non-compliance with codes, assisting designers in accurately locating the source of problems, improving the controllability and reliability of design outcomes, achieving high compliance accuracy, adapting to complex scenarios, and meeting current fire protection design requirements.

[0063] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0064] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0065] Figure 2 The diagram shows a structural schematic of a fire protection design device for an archive room based on a large language model, according to an embodiment of the present invention. For ease of explanation, only the parts relevant to the embodiment of the present invention are shown, and are described in detail below: like Figure 2 As shown, the fire protection design device for archives based on a large language model includes: The first data extraction module 21 is used to parse the archive drawings to obtain structured node features; The second data extraction module 22 is used to parse the standard clauses and form a graph database; The legal provision extraction module 23 is used to match the structured node features with the legal provisions in the graph database using a large language model to obtain the semantics of the target provisions. The solution output module 24 is used to obtain the fire protection design scheme based on the semantics of the target clauses and the characteristics of the structured nodes; Compliance inspection module 25 is used to conduct compliance inspections on fire protection design schemes; The scheme optimization module 26 is used to modify the fire protection design scheme using a large language model if the fire protection design scheme is not compliant, and then jump to the step of checking the compliance of the fire protection design scheme to continue execution until the fire protection design scheme is compliant.

[0066] In one possible implementation, the first data extraction module 21 may include: The preprocessing unit is used to preprocess the drawings in the archives to obtain parsable vector graphic data; The primitive recognition unit is used to perform primitive recognition on parsable vector graphic data to obtain primitive recognition results; The adjacency graph construction unit is used to extract the geometric parameters of the drawing based on the primitive recognition results using a geometric analysis engine, and then construct the adjacency graph. The semantic structure extraction unit is used to extract the global semantic structure of the adjacency graph using a graph neural network to obtain node feature sentences; Standardized units are used to uniformly encode node feature statements to obtain structured node features.

[0067] In one possible implementation, the regulatory provision extraction module 23 may include: The prompt word generation unit is used to construct standardized prompt word templates and generate prompt words based on structured node features; The first clause selection unit is used to input prompt words into the large language model and retrieve at least one candidate legal clause from the graph database. The second clause selection unit is used to calculate the similarity between any candidate legal clause and the structured node features; if the similarity between the candidate legal clause and the structured node features is greater than a preset value, then the candidate legal clause is used as the target clause semantics. Among them, the number of target clause semantics is at least one.

[0068] In one possible implementation, the solution output module 24 may include: The first design unit is used to determine whether there are specification conflicts or spatial layout problems based on the semantics of the target clause and the characteristics of the structured nodes, and to generate a problem diagnosis report and preliminary optimization strategies. The second design unit is used to input the problem diagnosis report and preliminary optimization strategy into the large language model for design reasoning and to generate optimized design parameters. The third design unit is used to integrate the optimized design parameters with the preliminary optimization strategy to obtain the fire protection design scheme.

[0069] In one possible implementation, the first design unit can be used to: sequentially perform rule conflict determination, spatial constraint check and risk priority assessment to determine whether there are any regulatory conflicts or spatial layout problems.

[0070] In one possible implementation, the second design unit can be specifically used to: call a large language model to perform fire extinguishing system configuration reasoning, fire separation distance generation, and ventilation system design suggestions to obtain optimized design parameters.

[0071] In one possible implementation, the compliance check module 25 may include: The first graph construction unit is used to construct the design fact graph based on the structured node features; The second graph construction unit is used to construct a standard design graph based on the graph database. The diagram comparison unit is used to compare the design fact diagram with the standard design diagram to determine whether the fire protection design scheme is compliant.

[0072] In one possible implementation, the map comparison unit can be specifically used for: 1. Analyze the standard design diagram to obtain relational constraints, numerical constraints, and symbolic logic rules; 2. Determine whether the design fact graph satisfies relational constraints, numerical constraints, and symbolic logic rules; 3. If at least one of the requirements is not met, the fire protection design scheme is deemed non-compliant. 4. If all conditions are met, the fire protection design scheme is deemed compliant.

[0073] Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 3 As shown, the electronic device 3 of this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, it implements the steps in the various method embodiments described above. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module / unit in the various device embodiments described above.

[0074] For example, computer program 32 may be divided into one or more modules / units, which are stored in memory 31 and executed by processor 30 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 32 in electronic device 3.

[0075] Electronic device 3 may include, but is not limited to, processor 30 and memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 3 may also include input / output devices, network access devices, buses, etc.

[0076] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0077] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0078] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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, and should all be included within the protection scope of the present invention.

Claims

1. A fire protection design method for archives based on a large language model, characterized in that, include: The archive drawings were analyzed to obtain structured node features; The standard provisions are parsed to form a graph database; A large language model is used to match the structured node features with the legal provisions in the graph database to obtain the semantics of the target provisions; Based on the semantics of the target clauses and the structured node features, a fire protection design scheme is obtained; The fire protection design scheme shall be subject to a compliance check. If the fire protection design scheme is non-compliant, the fire protection design scheme will be modified using a large language model, and the process will jump to the step of checking the compliance of the fire protection design scheme until the fire protection design scheme is compliant.

2. The fire protection design method for archives based on a large language model according to claim 1, characterized in that, The process of parsing the archive drawings to obtain structured node features includes: The archive drawings are preprocessed to obtain parsable vector graphic data; The parsable vector graphic data is subjected to primitive recognition to obtain primitive recognition results; Based on the primitive recognition results, the geometric parameters of the drawing are extracted using a geometric analysis engine, and an adjacency graph is constructed. A graph neural network is used to extract the global semantic structure of the adjacency graph to obtain node feature sentences; The node feature statements are uniformly encoded to obtain the structured node features.

3. The fire protection design method for archives based on a large language model according to claim 1, characterized in that, The method employs a large language model to match the structured node features with the legal provisions in the graph database to obtain the semantics of the target provisions, including: Construct standardized prompt word templates and generate prompt words based on the structured node features; The prompt words are input into a large language model, and at least one candidate legal provision is retrieved from the graph database. For any candidate legal provision, calculate the similarity between the candidate legal provision and the structured node features; if the similarity between the candidate legal provision and the structured node features is greater than a preset value, then the candidate legal provision is taken as the target legal provision semantics. The number of target clause semantics is at least one.

4. The fire protection design method for archives based on a large language model according to claim 1, characterized in that, The process of obtaining a fire protection design scheme based on the semantics of the target clause and the structured node features includes: Based on the semantics of the target clauses and the characteristics of the structured nodes, determine whether there are any specification conflicts or spatial layout issues, and generate a problem diagnosis report and preliminary optimization strategies; The problem diagnosis report and the preliminary optimization strategy are input into the large language model for design reasoning to generate optimized design parameters. The optimized design parameters are integrated with the preliminary optimization strategy to obtain the fire protection design scheme.

5. The fire protection design method for archives based on a large language model according to claim 4, characterized in that, Determining whether there are regulatory conflicts or spatial layout issues includes: The process involves sequentially determining rule conflicts, checking spatial constraints, and assessing risk priorities to identify any regulatory conflicts or spatial layout issues.

6. The fire protection design method for archives based on a large language model according to claim 4, characterized in that, The step of inputting the problem diagnosis report and the preliminary optimization strategy into a large language model for design reasoning to generate optimized design parameters includes: The large language model is invoked to perform fire suppression system configuration reasoning, fire separation distance generation, and ventilation system design suggestions to obtain the optimized design parameters.

7. The fire protection design method for archives based on a large language model according to any one of claims 1 to 6, characterized in that, The compliance check of the fire protection design scheme includes: Based on the structured node characteristics, a design fact graph is constructed; Based on the graph database, construct a standard design graph; The design fact map and the standard design map are compared to determine whether the fire protection design scheme is compliant.

8. The fire protection design method for archives based on a large language model according to claim 7, characterized in that, The comparison of the design fact map and the standard design map to determine whether the fire protection design scheme is compliant includes: The standard design diagram is analyzed to obtain relational constraints, numerical constraints, and symbolic logic rules; Determine whether the design fact graph satisfies the relational constraints, the numerical constraints, and the symbolic logic rules; If at least one of the requirements is not met, the fire protection design scheme is determined to be non-compliant. If all conditions are met, then the fire protection design scheme is deemed compliant.

9. A fire protection design device for archives based on a large language model, characterized in that, include: The first data extraction module is used to parse the archive drawings and obtain structured node features; The second data extraction module is used to parse the standard provisions and form a graph database; The legal provision extraction module is used to match the structured node features with the legal provisions in the graph database using a large language model to obtain the semantics of the target provisions. The scheme output module is used to obtain the fire protection design scheme based on the semantics of the target clauses and the structured node features; The compliance check module is used to check the compliance of the fire protection design scheme. The scheme optimization module is used to modify the fire protection design scheme using a large language model if the fire protection design scheme is not compliant, and then jump to the step of checking the compliance of the fire protection design scheme to continue execution until the fire protection design scheme is compliant.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the fire protection design method for archive rooms based on a large language model as described in any one of claims 1 to 8.

Images