Earthquake report review method and device based on domain knowledge enhanced large language model

Abstract

The present application relates to the technical field of large language model, in particular to a large language model anti-seismic report review method and device based on domain knowledge enhancement, which provides a large language model anti-seismic report review method based on domain knowledge enhancement, first establishes a large language model in the field of building anti-seismic specialty, then trains the large language model in the field of building anti-seismic specialty, then carries out data preprocessing on the anti-seismic review report, and finally reviews based on the trained large language model in the field of building anti-seismic specialty, so as to integrate the accumulated anti-seismic expert review experience and the large language model, build a professional LLM in the vertical field, strictly review against the standard and the file, effectively avoid the interference of human error and subjective factors, improve the accuracy and reliability of the review result, improve the review efficiency, provide a more scientific and accurate basis for expert review and internal pre-review, further improve the anti-seismic and disaster prevention ability of construction projects, and reduce the risk of earthquake disasters.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large language models, in particular to a large language model anti-seismic report review method and device based on domain knowledge enhancement. BACKGROUND

[0002] The seismic performance of buildings is directly related to people's life and property safety and social stability. In order to improve the anti-seismic ability of construction projects, reduce the risk of earthquake disasters, and protect people's life and property safety, corresponding building engineering anti-seismic fortification special review requirements have been introduced, especially in earthquake-prone areas.

[0003] For example, a province stipulates that "class A and medium-sized and above class B, and buildings outside the standard application range and not belonging to the out-of-limit building need to be subjected to anti-seismic fortification special review. At present, the anti-seismic special review is reported by the owner unit and the design unit, such as the anti-seismic special review report, and the housing and construction department organizes experts to review. The experts review by manually checking the submitted report, design drawings, calculation book and other large amount of data, and comparing with relevant standards and documents. There are the following problems:

[0004] (1) Low efficiency and large workload, which requires a lot of time and manpower, especially when facing large and complex projects;

[0005] (2) Easy to overlook, as the review data is complex, manual review is easy to overlook, and there are hidden dangers in the seismic safety performance of these important buildings;

[0006] (3) Limited experience and knowledge, the review project type is multiple, involves multiple technical contents, and there are cases beyond the experience and knowledge of the reviewer, which requires temporary review of a large amount of data, affecting the quality and efficiency of the review.

[0007] Recently, several open-source AI large language models (LLM) have been introduced at home and abroad. LLM has successfully applied in many fields such as education, medical treatment and finance, showing great application potential and providing new perspectives and methods for the construction industry. LLM can integrate a large amount of professional knowledge and data, relying on deep learning and natural language processing technology, to more accurately understand and analyze the information in the file. At the same time, the anti-seismic special review report is mainly in the form of text and table, which does not require a large amount of multi-modal calculation and has relatively low hardware requirements. The anti-seismic special review report is a long text that requires complex semantic segmentation and enhanced retrieval, which is the technical advantage of LLM, providing an opportunity to set up a large language model based on domain knowledge enhancement for anti-seismic report review. SUMMARY

[0008] The present application aims at the above-mentioned problems, and provides a large language model anti-seismic report review method and device based on domain knowledge enhancement.

[0009] The technical solution adopted by the present application is as follows: the large language model anti-seismic report review method based on domain knowledge enhancement comprises the following steps:

[0010] S1. Establish a building anti-seismic professional field large language model, set up a building anti-seismic professional field knowledge base, a review range library, a review integrity library and a review compliance library;

[0011] S2. Train the building anti-seismic professional field large language model to obtain sub-models for review tasks including structure classification model, design method recognition model, review range determination model, integrity determination model and compliance determination model;

[0012] S3. Data preprocessing is performed on the anti-seismic review report, and a project portrait is constructed;

[0013] S4. Review based on the trained building anti-seismic professional field large language model, output review results.

[0014] Further, the S1 comprises the following sub-steps:

[0015] S1.1. Collect national, industry, local anti-seismic design specifications, building type review cases and expert review opinions, and use Python natural language processing library to perform document analysis on text and table;

[0016] S1.2. Perform non-structured data processing on the parsed documents;

[0017] S1.3. Based on laws, regulations and building anti-seismic design standards, establish a review range library;

[0018] S1.4. Based on the design conditions and calculation results that must be included under different structure types and design methods in building anti-seismic, set up a review integrity library;

[0019] S1.5. Construct a review compliance library, based on the data after non-structured data processing, construct a four-tuple knowledge graph in the form of subject, relationship, sub-body and object;

[0020] S1.6. Convert the data in the building anti-seismic professional field knowledge base, the review range library, the review integrity library and the review compliance library into high-dimensional vectors through an embedding model, store them in a vector database for semantic retrieval.

[0021] Further, in the S1.2, the non-structured data processing on the parsed documents comprises setting metadata, experience pairs and a term library;

[0022] For the parsed specification articles of local seismic design specifications in the document, the articles are atomized and decomposed according to the chapter-clause-subitem, and the mandatory articles, applicable structure types are marked as metadata;

[0023] For the parsed review cases of building types in the document, the four elements of project parameters, violation points, corresponding articles and rectification suggestions are structured to form experience pairs for model learning;

[0024] The term library includes a mapping table related to building type professional terms.

[0025] Further, in S1.5, the data obtained based on S1.2 and the marked specification articles are stored in the form of four-tuple knowledge graph of subject, relationship, sub-body and object;

[0026] The subject can have multiple relationships, and each relationship can correspond to multiple sub-bodies, and finally point to a specific object.

[0027] Further, in S1.3, the review range library includes building classification, building purpose or building level, building scale, constraint condition;

[0028] The building classification includes disaster prevention and rescue buildings, public buildings and residential buildings;

[0029] The disaster prevention and rescue buildings include medical buildings, disease prevention and control center buildings, disaster prevention emergency command centers, emergency shelter buildings, fire truck garages and their duty rooms;

[0030] The public buildings and residential buildings include educational buildings, scientific experiment buildings, sports buildings, cultural and entertainment buildings, commercial buildings, office buildings, exhibition buildings, museum buildings, archives, information center buildings, old-age care institutions and child welfare institutions.

[0031] Further, in S1.4, the review integrity library is divided into two sub-libraries of design conditions and calculation results, and each review point R i The t i and m i are marked;

[0032] The t i represents the project structure category, and i is one of reinforced concrete frame structure, reinforced concrete frame-shear wall structure, reinforced concrete shear wall structure, reinforced concrete tube structure, steel frame structure, steel support structure, steel frame-support structure, steel frame-shear wall plate structure and steel tube structure;

[0033] The m i represents the project design method, and i is one of seismic design and seismic mitigation design.

[0034] Further, in S2, an open-source large language model is selected as the base, and the large language model in the building seismic professional field in S1 is trained, and the sub-models for the review task including the structure classification model, the design method recognition model, the review range determination model, the integrity determination model, and the compliance determination model are generated through instruction fine-tuning training;

[0035] The seismic review question and answer pair are injected into the large language model trained, which is used to impart field knowledge and reasoning mode to the large language model;

[0036] RAG must retrieve at least one specification provision or case from the vectorized database as a basis before generating any review results.

[0037] Further, in S3, the following sub-steps are included:

[0038] S3.1. Data preprocessing, using the natural language processing library of Python to clean the report text and table data to be reviewed, and the preprocessed data is uniformly converted into a format suitable for the input of the large language model;

[0039] S3.2. Through the structure classification model, retrieve project information, determine the structure category, and match the review point labeled t i in the review integrity library;

[0040] S3.3. Through the design method recognition model, retrieve project information, determine the design method, and match the provision labeled m i in the review integrity library;

[0041] S3.4. First-level review, through the review range determination model, retrieve project information, automatically extract the project information of the report to be reviewed, associate the review range library, and determine whether it belongs to the project within the review range;

[0042] S3.5. Second-level review, based on the labeled t i and the labeled m i , form a conditional trigger, call the integrity determination model, associate the review point R i labeled t i and m i in the review integrity library, and retrieve R i items in the report. If the relevant parameters are retrieved, enter the third-level review S3.6, which has the following formula:

[0043] R i =R 1i +R 2i ={R1∈(t 1i ∩m 1i )+R2∈(t 2i ∩m2i )};

[0044] In the above formula, Ri refers to the review points in the review integrity library;

[0045] R 1i refers to the review points in the review integrity library related to the design conditions;

[0046] R 2i refers to the review points in the review integrity library related to the calculation results;

[0047] t i refers to the review points in the review integrity library marked with the structure category;

[0048] t 1i refers to the review points in the review integrity library related to the design conditions and marked with the structure category;

[0049] t 2i refers to the review points in the review integrity library related to the calculation results and marked with the structure category;

[0050] m 1i refers to the review points in the review integrity library related to the design conditions and marked with the design method;

[0051] m 2i refers to the review points in the review integrity library related to the calculation results and marked with the design method;

[0052] If not retrieved, output the parameter missing in the review conclusion;

[0053] S3.6. Three-level review, establish the correspondence between the integrity judgment model and the compliance judgment model parameters, R i correspond to the four-tuple relationship in the compliance judgment model knowledge base, t i correspond to the four-tuple subject, m i and the remaining parameters correspond to the four-tuple subclass, and the quantitative parameters correspond to the four-tuple object.

[0054] Further, in the S4, after being reviewed based on the trained building anti-seismic professional field large language model, the word version of the review conclusion is output, and is arranged and presented, for the part of the review report that determines compliance, the compliance is marked, and is displayed in the background; for the part with problems, the location of the problem in the review conclusion, the type of the problem and the specific anti-seismic specification provisions violated are output, and if there are mandatory provisions, the mandatory provisions are output and displayed in bold.

[0055] The application also provides a large language model anti-seismic report review device based on field knowledge enhancement, which is suitable for executing the large language model anti-seismic report review method based on field knowledge enhancement.

[0056] The beneficial effects of the present application are;

[0057] A large language model anti-seismic report review method based on domain knowledge enhancement is provided, which integrates accumulated anti-seismic review experience and a large language model to build a professional LLM in a vertical field, can strictly review against standards and files, effectively avoids human error and subjective factor interference, improves the accuracy and reliability of the review results, improves the review efficiency, provides a more scientific and accurate basis for expert review and internal pre-review, further improves the anti-seismic and disaster prevention capability of construction projects, and reduces the risk of earthquake disasters. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 A large language model anti-seismic report review method based on domain knowledge enhancement is provided, which integrates accumulated anti-seismic review experience and a large language model to build a professional LLM in a vertical field, can strictly review against standards and files, effectively avoids human error and subjective factor interference, improves the accuracy and reliability of the review results, improves the review efficiency, provides a more scientific and accurate basis for expert review and internal pre-review, further improves the anti-seismic and disaster prevention capability of construction projects, and reduces the risk of earthquake disasters.

[0059] Figure 2 A main body, a relationship, a sub-body, and a relationship diagram are provided.

[0060] Figure 3 An embodiment diagram is provided.

[0061] Figure 4 A review integrity library structure diagram is provided.

[0062] Figure 5 A review integrity library and compliance determination model knowledge base matching diagram is provided.

[0063] Figure 6 A structure diagram of an electronic device is provided. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described in the drawings herein can be arranged and designed in various different configurations.

[0065] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0066] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0067] It should be noted that similar reference numerals and letters refer to similar items in the drawings, and therefore, once an item is defined in one drawing, it need not be further defined and explained in subsequent drawings.

[0068] As shown in the drawings, the method for reviewing an anti-seismic report based on a large language model enhanced by domain knowledge comprises the following steps: Figure 1

[0069] S1. Establish a large language model for the building anti-seismic professional field, and set up a building anti-seismic professional knowledge base, a review scope base, a review integrity base, and a review compliance base;

[0070] S2. Train the large language model for the building anti-seismic professional field to obtain sub-models for the review task, including a structure category classification model, a design method identification model, a review scope determination model, an integrity determination model, and a compliance determination model;

[0071] S3. Preprocess the data of the anti-seismic review report and construct a project portrait;

[0072] S4. Review based on the trained large language model for the building anti-seismic professional field and output the review results.

[0073] The purpose of such design is to provide a method for reviewing an anti-seismic report based on a large language model enhanced by domain knowledge, to integrate accumulated anti-seismic review experience with a large language model, to build a professional LLM for a vertical field, to strictly review against standards and documents, to effectively avoid human errors and subjective factors, to improve the accuracy and reliability of the review results, to improve the review efficiency, to provide a more scientific and accurate basis for expert review and internal pre-review, to further improve the anti-seismic and disaster prevention capability of construction projects, and to reduce the risk of earthquake disasters.

[0074] The method solves the problems of low efficiency, large review workload, and the need for a large amount of time and manpower in the existing anti-seismic report review, especially when facing large and complex projects, which is prone to omissions. Due to the complexity of the review materials, manual review is prone to omissions, which poses a risk to the anti-seismic safety performance of important buildings. The experience and knowledge are limited, the types of review projects are diverse, and the technical content involved is extensive, which exceeds the experience and knowledge range of the review personnel, and a large amount of data needs to be consulted temporarily, affecting the quality and efficiency of the review.

[0075] In the present embodiment, the S1 comprises the following sub-steps:

[0076] S1.1. Collect national, industry, and local anti-seismic design specifications, building type review cases, and expert review opinions, and parse the documents through the natural language processing library of Python;

[0077] ​S1.2. Unstructured data processing on the parsed documents;

[0078] S1.3. Establishing a review scope library based on laws, regulations and building seismic design standards;

[0079] S1.4. Setting up a review integrity library based on design conditions and calculation results that must be included under different structural types and design methods in building seismic design;

[0080] S1.5. Building a review compliance library, based on the data after unstructured data processing, to build a quadruple knowledge graph in the form of subject, relationship, body and object;

[0081] S1.6. Converting the data in the building seismic professional knowledge base, review scope library, review integrity library and review compliance library into high-dimensional vectors through embedding models and storing them in a vector database for semantic retrieval.

[0082] In S1.1, local seismic design specifications, various building type review cases, and expert review opinions are used. Through the use of Python's natural language processing library NLTK, OCR+LayoutLM, the text and tables in the documents are parsed and preprocessed.

[0083] In S1.2, unstructured data processing on the parsed documents includes setting up metadata, experience pairs and a term library;

[0084] For the specification provisions of local seismic design specifications in the parsed documents, they are atomized and disassembled according to "chapter-clause-subitem", and the mandatory provisions and applicable structural types are marked as metadata;

[0085] For the review cases of building types in the parsed documents, the four elements of "project parameters-violation points-corresponding provisions-reform suggestions" are structured and extracted to form experience pairs for model learning;

[0086] The term library includes a mapping table of building type professional terms.

[0087] It should be noted that the mapping table of professional terms is used to solve the problems of synonymous different names, abbreviations, etc. such as "floor interlayer maximum displacement to floor height ratio" corresponding to "interlayer displacement angle", "Δu / h"; [Δu / h] corresponding to "Δu / h limit value", "interlayer displacement angle limit value".

[0088] Meanwhile, in S1.3, the review scope library includes building classification, building use or building level, building scale, and constraint conditions;

[0089] The building classification includes disaster prevention and rescue buildings, public buildings and residential buildings;

[0090] In S1.4, the integrity review library is divided into two sub-libraries: design conditions and calculation results, and each review point R... i Mark t i and m i ;

[0091] The t i This indicates the structural category of the project; i represents a type of reinforced concrete shear wall structure or steel frame structure.

[0092] The m i This indicates the project design method, where 'i' represents one of the following: seismic design or seismic isolation and damping design.

[0093] In S1.5, the normative clauses obtained from the data processing and annotation of S1.2 are stored in a knowledge graph format of four tuples in the form of subject, relation, sub-subject and object;

[0094] like Figure 2 As shown, the subject can have multiple relationships, and each relationship can correspond to multiple sub-sub ...

[0095] In practice, such as Figure 3 and Figure 6 As shown, according to building classification (including primary and secondary categories, purpose or level, building scale, and constraints such as building area or building height), the inpatient building area of ​​a tertiary hospital is >0m². 2 Or, if the height is greater than 0m, a special review is required. The inpatient building area of ​​a secondary hospital is greater than 5000m². 2 If the height is greater than 24m, a special review is required.

[0096] like Figure 3 As shown, based on the provisions of the "Technical Specification for Concrete Structures of High-Rise Buildings", the quaternion storage is as follows:

[0097] {“Reinforced concrete frame structure”, “Limit of elastic inter-story drift angle”, “Height not greater than 150m”, “1 / 550”};

[0098] {“Reinforced concrete frame structure”, “Limit of elastic inter-story drift angle”, “Height not less than 250m”, “1 / 500”}.

[0099] In the large language model training part, S2 uses an open-source large language model as the base to train the large language model in the field of seismic engineering in S1. Through fine-tuning training with instructions, sub-models for the review task are generated, including structural category classification model, design method identification model, review scope determination model, integrity determination model and compliance determination model.

[0100] LoRA is used to inject the large language model trained after the anti-seismic review question and answer pair into the large language model, which is used to impart domain knowledge and reasoning mode to the large language model.

[0101] RAG must retrieve at least one specification provision or case from the vectorized database as a basis before generating any review results.

[0102] It should be noted that the structural category classification model, the design method identification model, the review scope determination model, the integrity determination model and the compliance determination model are selected for the sub-models of the review task. This modular design reduces the training difficulty and improves the execution accuracy of each task.

[0103] At the same time, LoRA (Low-Rank Adaptation) technology is used to inject thousands of high-quality anti-seismic review question and answer pairs (QA pairs) into LLM, which efficiently imparts domain knowledge and reasoning mode to the model.

[0104] Format:

[0105] Input: report segment + review task;

[0106] Output: determination result + basis provision ID + reasoning process;

[0107] Comparison: The determination result is compared with the preset result.

[0108] As shown in the following example:

[0109] Input: Report segment: "Engineering summary (such as specific content: Chengdu XX Vocational Education School new office building, covering an area of 4000 square meters, a total of 6 layers, each layer 4.2m)", review task: "Judge whether the project needs anti-seismic special review"

[0110] Output: needs review, according to "Sichuan Province Housing Construction Engineering Anti-seismic Special Review Technical Points" Appendix 1, building classification is education building, use is office building, building height 6x4.2m=25.2m>24m, needs review.

[0111] Comparison: The determination result is: needs review, the preset result is: needs review. Comparison: Correct.

[0112] In addition, RAG is used for retrieval enhancement generation (RAG): the model must retrieve at least one specification provision or case from the vectorized knowledge base as a basis before generating any determination result. This significantly improves the reliability and traceability of the review conclusion, effectively suppressing the model "hallucination".

[0113] In specific implementation, in S3, Figure 4 and Figure 5As shown, including the following sub-steps:

[0114] S3.1. Data preprocessing, using Python natural language processing library to clean the report text and table data, and the preprocessed data is converted into a format suitable for large language model input;

[0115] S3.2. Retrieve project information by structure classification model, judge structure category, match the labeled t i of the review completeness library;

[0116] S3.3. Retrieve project information by design method recognition model, judge design method, match the labeled m i of the review completeness library;

[0117] S3.4. First-level review, through the review scope determination model, retrieve project information, automatically extract the project information of the report to be reviewed, associate the review scope library, and determine whether it belongs to the project within the review scope;

[0118] As in the "Engineering Summary" chapter of the "XX Hospital Inpatient Building Construction Project Seismic Special Review Report", the project introduction "This project is an expansion project of the outpatient department of XX Level 3 Hospital in XX City, Sichuan Province, with a construction address at XX Street. The project is 42.4m high and adopts a reinforced concrete frame structure."

[0119] The extracted information is { "Level 3 hospital", "outpatient department", "42.4m high"}, associated with S1.3, and the building classification is "medical building", the purpose is "Level 3 hospital outpatient department", and the constraint condition is "building area 5514m 2 " are all met, and it is determined to be within the scope of review, i.e. it needs to be reviewed.

[0120] S3.5. Second-level review, based on the labeled t i and the labeled m i , form condition trigger, call completeness judgment model, associate the review point R i of the labeled t i and the labeled m i in the review completeness library, search for R i in the report, and if the relevant parameters are found, proceed to the third-level review S3.6, which has the following formula:

[0121] R i =R 1i +R 2i ={R1∈(t 1i ∩m 1i )+R2∈(t 2i ∩m 2i )};

[0122] If not retrieved, output the parameter missing in the review conclusion;

[0123] In the above formula, Ri refers to the review point in the review integrity library;

[0124] R 1i Refers to the review point in the review integrity library related to the design condition;

[0125] R 2i Refers to the review point in the review integrity library related to the calculation result;

[0126] t i Refers to the review point in the review integrity library marked with the structure category;

[0127] t 1i Refers to the review point in the review integrity library related to the design condition and marked with the structure category;

[0128] t 2i Refers to the review point in the review integrity library related to the calculation result and marked with the structure category;

[0129] m 1i Refers to the review point in the review integrity library related to the design condition and marked with the design method;

[0130] m 2i Refers to the review point in the review integrity library related to the calculation result and marked with the design method.

[0131] At the same time, as the structure category judgment model and the design method identification model, output t i = reinforced concrete frame structure, m i = seismic design, then secondary review, R 1i = (structure design working life, seismic fortification level, safety level), R 2i = (period ratio, inter-story drift angle).

[0132] According to R i , the relevant text "structure design working life 50 years" "seismic fortification level two" "period ratio 0.89" "inter-story drift angle 1 / 550" is retrieved, compared with R 1i and R 2i .

[0133] Output: "safety level" missing.

[0134] S3.6. Third-level review, establish the correspondence between the integrity judgment model and the compliance judgment model parameters, R i corresponds to the four-tuple relationship in the compliance judgment model knowledge base, t i corresponds to the four-tuple subject, and m iAnd the remaining parameter corresponds to the four tuple subclass, the quantization parameter corresponds to the four tuple object.

[0135] As "period ratio 0.89" corresponds to:

[0136] R i = Period ratio correspondence;

[0137] t i = reinforced concrete frame structure corresponding object;

[0138] m i = seismic design + A-level height corresponding sub-object;

[0139] Quantization parameter = 0.89 corresponding four tuple object;

[0140] The processed four tuple is: {“reinforced concrete frame”, “A-level height”, “period ratio”, “0.89”}, compared with the four tuple {“A-level height high-rise reinforced concrete frame”, “period ratio limit”, “0.9”} in the “integrity library”, it is determined to be compliant.

[0141] Finally, in S4, based on the trained building seismic professional field large language model review, output the review conclusion in word version, and arrange and present, for the part of the review report that is determined to be compliant, mark compliance and display in the background; for the part with problems, output the location of the problem, the type of the problem and the specific seismic code provisions violated, at the same time, if there are mandatory provisions, output the mandatory provisions and display them in bold.

[0142] In the example, the user uploads the “seismic special review report”, the background data is preprocessed, the trained professional field large language model is used to review the “seismic special review report” after data processing, and finally the large language model completes the review and outputs the review result.

[0143] Taking the review of “Sichuan Province XX City XX Level III Hospital Inpatient Building Construction Project Seismic Special Review Report” as an example, the implementation process of the present application is explained:

[0144] The system deployment part includes:

[0145] Knowledge base configuration: enter the current standard specifications, including more than 80 strong articles, more than 200 general articles, and 30 similar project review cases;

[0146] Model training: based on LLaMA3-70B model, using 300 question and answer pairs and review cases to fine-tune the base model, LoRA rank is set to 32, training epoch = 8, and validation set accuracy is 94.3%;

[0147] Hardware environment: 8x A100 GPU server is adopted.

[0148] Review execution steps, including:

[0149] 1. Report input: the user uploads a PDF format review report (more than 40 pages, including more than 10 tables);

[0150] 2. Preprocessing;

[0151] OCR+LayoutLM is adopted to process text and table data, and extract relevant data.

[0152] 3. Structure category classification:

[0153] The structure category classification model is called to extract the information "reinforced concrete frame" and output ti= reinforced concrete frame structure.

[0154] 4. Design method identification:

[0155] The design method identification model is called to extract the information {“tertiary hospital”, 8 degree area}, which belongs to “two area eight category buildings”, and output mi= seismic mitigation and isolation design.

[0156] 5. First-level review:

[0157] In the “project summary” chapter of the report, the extracted information is {“tertiary hospital”, “outpatient department”, “42.4m high”}, which is associated with S1.1.3, matches the knowledge base rules, and determines that it needs to be reviewed.

[0158] 6. Second-level review:

[0159] According to the labeled t i and m i , the R i is associated, and these R i items are searched in the report;

[0160] R 1i : structure design working age, durability age, seismic fortification category, safety level, foundation design level;

[0161] R 2i : period ratio, inter-story drift angle, period before and after seismic isolation under seismic action;

[0162] After searching, it is found that the foundation design level is missing, and the “missing foundation design level” is output to the “result output model”.

[0163] 7. Third-level review:

[0164] During the secondary review, the text "cycle ratio 0.91" was retrieved, and the corresponding quadruple was {"Grade A high-rise reinforced concrete frame", "cycle ratio", "0.91"}. Compared with the knowledge stored in the knowledge base, which is {"Grade A high-rise reinforced concrete frame", "cycle ratio limit", "0.9"}, it was determined to be non-compliant.

[0165] 8. Result Output:

[0166] Generate a review report. Report content:

[0167] ① Lack of foundation design grade;

[0168] ② The project's "cycle ratio" is in violation of regulations. According to the "Technical Specification for Concrete Structures of High-Rise Buildings", the torsion ratio of reinforced concrete buildings of Class A height shall not exceed 0.9.

[0169] like Figure 6 The diagram shown is a structural schematic of an electronic device according to an embodiment of this application. The electronic device includes a memory and a processor. The memory stores computer-readable instructions, and the processor executes the computer-readable instructions stored in the memory to implement the domain knowledge-enhanced large language model seismic report review method described in any of the above embodiments.

[0170] In one embodiment of this application, the electronic device further includes a bus and a computer program stored in the memory and executable on the processor, such as a domain knowledge-enhanced large language model seismic report review program.

[0171] The figure illustrates an electronic device with memory and a processor. Those skilled in the art will understand that the structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0172] It should be noted that the memory in the electronic device stores multiple computer-readable instructions to implement a large language model seismic report review method based on domain knowledge enhancement, and the processor can execute the multiple instructions to implement this method.

[0173] Specifically, the processor's implementation method for the above instructions can be found in the description of the relevant steps in the corresponding embodiment of the figure, and will not be repeated here.

[0174] Those skilled in the art will understand that the schematic diagram is merely an example of an electronic device and does not constitute a limitation on the electronic device. The electronic device may be a bus-type structure or a star-type structure. The electronic device may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, the electronic device may also include input / output devices, network access devices, etc.

[0175] It should be noted that the electronic device is only an example, and other existing or future electronic products can also be adapted to the present application and should be included in the protection scope of the present application.

[0176] The memory includes at least one type of readable storage medium, which can be non-volatile or volatile. The readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory can be an internal storage unit of the electronic device in some embodiments, such as a mobile hard disk of the electronic device. The memory can also be an external storage device of the electronic device in other embodiments, such as a plug-in mobile hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the electronic device. The memory can be used not only to store application software and various data installed on the electronic device, such as the code of the large language model anti-seismic report review program enhanced based on domain knowledge, but also to temporarily store data that has been or will be output.

[0177] The processor can be composed of integrated circuits in some embodiments, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same function or different functions, including one or more combinations of central processing units (CPU), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor is the control core of the electronic device, which connects all components of the electronic device through various interfaces and lines, executes programs or modules stored in the memory (such as the customer group pushing program based on intellectual property change data), and calls data stored in the memory to perform various functions and process data of the electronic device.

[0178] The processor executes the operating system of the electronic device and various installed application programs. The processor executes the application programs to implement the steps in each of the above large language model anti-seismic report review method embodiments enhanced based on domain knowledge, such as the steps shown in the figure.

[0179] Illustratively, the computer program can be segmented into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units can be a series of computer-readable instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the electronic device. For example, the computer program can be segmented into a receiving module, a preprocessing module, a projection module, and a determination module.

[0180] The integrated units implemented in the form of software function modules described above can be stored in a computer-readable storage medium. The software function modules described above are stored in a storage medium and include a number of instructions for causing a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor (Processor) to execute part of the anti-seismic report review method based on the field knowledge enhanced large language model described in various embodiments of the present application.

[0181] The modules / units integrated in the electronic device, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program to instruct related hardware devices, and the computer program can be stored in a computer-readable storage medium. When the processor executes the computer program, the steps of the above-mentioned various method embodiments can be implemented.

[0182] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some of the technical features, as long as they are within the spirit and principles of the present application. Any modification, equivalent replacement, improvement, etc. made shall be included in the protection scope of the present application.

Claims

1. A method for reviewing seismic resistance reports based on a large language model with enhanced domain knowledge, characterized in that: Includes the following steps: S1. Establish a comprehensive language model for the field of seismic resistance in buildings, and set up a knowledge base, review scope base, review completeness base, and review compliance base for the field of seismic resistance in buildings, including the following sub-steps: S1.

1. Collect national, industry, and local seismic design codes, building type review cases, and expert review opinions, and use Python's natural language processing library to parse the text and tables. S1.

2. Perform unstructured data processing on the parsed document; S1.

3. Establish a review scope database based on laws, regulations, and building seismic design standards; S1.

4. A review integrity database is established based on the design conditions and calculation results that must be included under different structural types and design methods in seismic engineering. The review integrity database is divided into two sub-databases: design conditions and calculation results. Each review point R... i Mark t i and m i ; The t i Indicates the structural category of the project, where i represents one of the following: reinforced concrete frame structure, reinforced concrete frame-shear wall structure, reinforced concrete shear wall structure, reinforced concrete tube structure, steel frame structure, steel braced structure, steel frame-braced structure, steel frame-shear wall panel structure, and steel tube structure. The m i This indicates the project design method, where 'i' represents one of the following: seismic design or seismic isolation and damping design. S1.

5. Construct a review compliance library, based on unstructured data after processing, and build a four-tuple knowledge graph in the form of subjects, relations, sub-subjects and objects; S1.

6. Transform the data in the knowledge base, review scope base, review integrity base, and review compliance base of the building seismic resistance professional field into high-dimensional vectors through an embedding model, and store them in a vector database for semantic retrieval; S2. Train the large language model in the field of seismic resistance of buildings to obtain sub-models for review tasks, including structural category classification model, design method identification model, review scope determination model, integrity determination model and compliance determination model; S3. Perform data preprocessing on the seismic review report and construct a project profile, including the following sub-steps: S3.

1. Data preprocessing: Python's natural language processing library is used to clean the text and table data of the review report. After preprocessing, the data is uniformly converted into a format suitable for input to a large language model. S3.

2. Using the structural category classification model, retrieve project information, determine the structural category, and match the annotations t in the integrity review database. i The review points; S3.

3. Identify the model through design methods, retrieve project information, determine the design method, and match it with the m marked in the integrity review library. i The provisions; S3.

4. Level 1 review: Through the review scope determination model, retrieve project information, automatically extract project information of the report to be reviewed, associate with the review scope database, and determine whether the project falls within the review scope. S3.

5. Secondary review, based on annotation t i and the annotation m i This triggers a condition, invokes the integrity determination model, and associates the integrity data marked with t in the integrity review library. i and the annotation m i Review point R i Search for R in the report i If relevant parameters are found, the process proceeds to Level 3 review (S3.6), as shown in the following formula: R i =R 1i +R 2i ={R1∈(t 1i ∩ m 1i ) + R2∈(t 2i ∩ m 2i )}; In the above formula, Ri refers to the review point in the review integrity database; R 1i This refers to the review points related to the integrity of the review library and design conditions; R 2i This refers to the review points related to the integrity of the database and the calculation results; t i This refers to the review points in the integrity review library that are marked with structural categories; t 1i This refers to review points in the integrity database that are relevant to the design conditions and have been marked with structural categories; t 2i This refers to review points that are related to the integrity of the review database and the calculation results, and that are marked with structural categories; m 1i This refers to review points in the integrity library that are related to the design conditions and have been marked with the design methodology. m 2i This refers to review points in the integrity database that are relevant to the calculation results and have been marked with the design method. If not retrieved, the review conclusion will output that this parameter is missing. S3.

6. Three-level review: Establish the correspondence between the parameters of the integrity assessment model and the compliance assessment model, and incorporate R... i The corresponding four-tuple relationship in the compliance judgment model knowledge base will be t i Corresponding to the main body of the quadruple, m i The remaining parameters correspond to the quadruple subclasses, and the quantized parameters correspond to the quadruple objects; S4. Review based on the trained large language model of building seismic resistance and output the review results.

2. The method for reviewing seismic resistance reports based on a large language model with domain knowledge enhancement according to claim 1, characterized in that, In step S1.2, the unstructured data processing of the parsed document includes setting metadata, experience pairs, and a terminology database; For the code provisions of local seismic design codes in the parsed documents, they are atomized and decomposed according to "chapter-clause-sub-item", and mandatory provisions and applicable structural types are marked as metadata; For the review cases of building types in the parsed documents, the four elements of "project parameters - violation points - corresponding clauses - rectification suggestions" are extracted in a structured manner to form experience pairs for the model to learn from; The terminology database includes a mapping table of professional terms related to building types.

3. The method for reviewing seismic resistance reports based on a large language model with domain knowledge enhancement according to claim 2, characterized in that, In S1.5, the normative clauses obtained from S1.2 after data processing and annotation are stored in a knowledge graph format of four tuples in the form of subject, relation, sub-subject and object; The subject can have multiple relationships, and each relationship can correspond to multiple sub-sub ...

4. The method for reviewing seismic resistance reports based on a large language model with domain knowledge enhancement according to claim 1, characterized in that, In S1.3, the scope of review includes building classification, building use or building level, building size, and constraints; The building classification includes disaster prevention and relief buildings, public buildings, and residential buildings; The disaster prevention and relief buildings include medical buildings, disease prevention and control center buildings, disaster prevention and emergency command centers, emergency shelter buildings, fire truck garages and their duty rooms; The public and residential buildings mentioned include educational buildings, scientific research buildings, sports buildings, cultural and entertainment buildings, commercial buildings, office buildings, convention and exhibition buildings, museum buildings, archives, information center buildings, elderly care institutions, and children's welfare homes.

5. The method for reviewing seismic resistance reports based on a large language model with domain knowledge enhancement according to claim 1, characterized in that, In S2, an open-source large language model is selected as the base to train the large language model in the field of seismic resistance of buildings in S1. Through fine-tuning training with instructions, sub-models for the review task are generated, including structural category classification model, design method identification model, review scope determination model, integrity determination model and compliance determination model. LoRA was used to inject seismic review question-and-answer pairs into the trained large language model, which was used to teach the large language model domain knowledge and reasoning patterns. Before generating any review results, RAG must first retrieve at least one normative clause or case from the vectorized database as a basis.

6. The method for reviewing seismic resistance reports based on a large language model with enhanced domain knowledge as described in claim 1, characterized in that, In S4, after reviewing based on the trained large language model of building seismic resistance, a Word version of the review conclusion is output, organized, and presented. For the compliant parts in the review report, compliance is marked and displayed in the background. For the problematic parts, the review conclusion outputs the location of the problem, the type of problem, and the specific seismic code clauses violated. At the same time, if there are mandatory clauses marked, the mandatory clauses are output and displayed in bold.

7. A large language model-based seismic report review device based on domain knowledge enhancement, characterized in that: The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the domain knowledge-enhanced large language model seismic report review method as described in any one of claims 1 to 6.

Images