Construction scheme generation method and system fusing large language model and knowledge graph

Abstract

The invention relates to the technical field of intelligent construction, and particularly discloses a construction scheme generation method and system fusing a large language model and a knowledge graph, and the method comprises key links such as intelligent directory generation, intelligent content generation and deep content quality analysis, achieves the intelligent generation of a building engineering construction scheme, and improves the construction efficiency. The technical blank in the field is effectively filled, the automation and intelligence process of the building industry is promoted, the advanced large language model technology and the knowledge graph technology are combined, a whole set of technical framework and intelligent processing system are constructed, the time and energy required in the manual compiling process are remarkably saved, the working efficiency is improved, and the construction cost is reduced. The project engineering information and the building standard specification are organically combined by utilizing the knowledge graph technology, so that the professionality and the standardization of intelligent generation of the construction scheme are ensured, and the accuracy and the reliability of the construction scheme are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction technology, and specifically to a construction plan generation method and system integrating a large language model and a knowledge graph. Background Art

[0002] Large language models (LLMs) are an advanced technology in the field of artificial intelligence. They are leading a revolution with their powerful language understanding and generation capabilities. These models can learn from large amounts of text data to achieve natural language understanding and generation. With the explosive growth of data and the improvement of computing power, deep learning technology has developed rapidly, promoting progress in the field of language models. Large language models are gradually penetrating into all aspects of our daily lives, from natural language processing to intelligent dialogue systems. Large language models are usually based on deep neural networks with Transformer architectures. This architecture consists of multiple encoders and decoders, each of which contains multiple self-attention mechanisms and feedforward neural networks. Through multi-layer stacking, Transformer can capture long-distance dependencies and contextual information. During training, the model continuously generates text through self-dialogue and fine-tunes it according to the target tasks (such as text classification, translation, and question-answering). With the increase in the amount of training data and the expansion of the model scale, large language models can better understand and generate natural language text. Large language models have a wide range of application scenarios, including natural language processing, intelligent customer service, machine translation, speech recognition, and dialogue systems. These models perform well in understanding and generating human language, and therefore have broad application potential in various situations that require processing and understanding natural language.

[0003] Knowledge graph is an important part of artificial intelligence and semantic web technology. Its core lies in organizing objects and concepts in the real world and the various relationships between them in a graphical way. This technology is not only a data structure, but also a way to express and store knowledge. It can provide rich and structured background knowledge for machine learning, thereby improving the understanding and reasoning ability of the algorithm. The principle of knowledge graph is based on graph database and semantic web technology. Through the structuring and semantic processing of knowledge, knowledge sharing, reuse and reasoning are realized. The application of knowledge graph is very wide. It can be used in intelligent question and answer, recommendation system, search engine, auxiliary writing and other fields. Through knowledge graph, we can better understand users' questions and needs and provide more accurate and intelligent answers and services. Knowledge graph technology uses big data and artificial intelligence technology to express Internet information in a form closer to the human cognitive world, providing strong support for Internet knowledge services. This technology enables computers to better understand and process the relationship between complex human language and the real world, thus playing an important role in many application fields. Traditional construction plan preparation has some limitations and challenges. The preparation method that mainly relies on manual experience requires a lot of time and human resources, which is especially time-consuming and inefficient when dealing with complex engineering or large projects. At the same time, manual preparation is prone to errors in construction plans due to human errors, which may cause construction delays or quality problems. In addition, manually prepared construction plans may lack standardization due to different personal habits and styles, which brings difficulties to construction management and quality control. Therefore, we propose a construction plan generation method and system that integrates large language models and knowledge graphs. Summary of the invention

[0004] The purpose of the present invention is to provide a construction plan generation method and system that integrates a large language model and a knowledge graph, thereby solving the problems raised in the background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a construction plan generation method integrating a large language model and a knowledge graph, comprising S100 training a large language model for construction engineering, specifically comprising the following sub-steps, S101: collecting text data in the construction field; S102: Cleaning and preprocessing the text data in the architectural field; S103: Using the open source big language model of the Transformer architecture, the processed text data in the construction field is imported into the open source big language model of the Transformer architecture. Based on the open source big language model, the text data set of the text data in the construction field is used for fine-tuning to form a big language model for construction engineering; S104: Continuously evaluate the performance of the large language model for construction engineering during training, using validation and test sets to test the accuracy, robustness, and generalization of the model; S105: The trained large language model for construction engineering is deployed in actual applications and provides services in the form of questions and answers. The services include questions and answers on construction plan knowledge, legal and regulatory knowledge, engineering knowledge, legal and regulatory atlas knowledge, and online learning application scenarios. S200: after receiving the intelligent generation instruction of the construction plan of the building project issued by the user, query and obtain the basic information and design drawings of the building project, and query and obtain the engineering information required for the construction plan according to the type of the construction plan to be generated, so as to form a set of building project information; S300 generates a construction plan catalog, specifically including the following sub-steps, S301: constructing a first-level catalog question statement of the large language model of the building project according to the type of the construction plan and the required engineering information, inputting it into the large language model of the building project, and obtaining the first-level catalog of the construction plan and the first-level catalog structure priority; S302: for each first-level directory of the construction plan, construct a second-level directory question statement of the building engineering large language model according to the type of the construction plan and the required engineering information, input it into the building engineering large language model, and obtain the second-level directory of the construction plan and the priority of the second-level directory structure; S400 generates the initial plan content according to the directory structure, specifically including the following sub-steps S401: for each secondary directory in the construction plan, query the corresponding engineering information required for the secondary directory from the construction engineering information set; S402: for each secondary directory in the construction plan, and based on the corresponding engineering information, query the relevant engineering knowledge from the construction engineering knowledge graph to obtain the corresponding knowledge graph data; S403: For each secondary directory of the construction plan, and based on the corresponding engineering information and knowledge graph data, a large language model question sentence is formed; for each secondary directory of the construction plan, the formed large language model question sentence is formed; S404: inputting into the architectural language model, and obtaining the solution content of the secondary directory, forming the initial solution content of the secondary directory; S405: generating initial plan contents for all secondary directories of the construction plan according to step S404 to form the overall initial plan contents of the construction plan; S500 checks and optimizes the quality of the construction plan content, including the following sub-steps: S501: Review the quality of the initial content of each secondary directory in the construction plan, the construction engineering knowledge graph and the large language model. Query the relevant specifications and requirements of the building from the construction engineering knowledge graph. According to the relevant specifications and requirements, determine whether the initial content of the secondary directory is qualified. If qualified, continue to execute the subsequent step S502. If unqualified, input the unqualified information into the large language model question statement to form a new large language model question statement, regenerate the initial content of the directory, and re-examine the quality of the initial content. S502: After the initial plan contents of each secondary directory of the construction plan meet the quality review, the initial qualified construction plan contents are formed; S503: Analyze the consistency of the initial qualified contents between different catalogs for all the initial qualified contents of the construction plan; S504: If unqualified, combining the priorities of the first-level directory and the second-level directory, for the inconsistent initial qualified solution content, the low-priority initial qualified content is re-formed into a large language model question statement, and the high-priority initial qualified content is added to the question statement, the initial content of the low-priority directory is regenerated, and the initial content quality and initial qualified content consistency are re-examined; S505: After the initial qualified contents of each secondary directory of the construction plan meet the consistency review, the qualified construction plan contents are formed.

[0006] S600: Output the construction plan according to the construction plan output format requirements and using qualified construction plan content.

[0007] A construction plan generation system that integrates a large language model and knowledge graph, including a data collection and preprocessing module, which is used to collect text data in the construction field and perform data cleaning and preprocessing; The model architecture selection and design module is used to select an open source large language model with a Transformer architecture and design appropriate model parameters and configurations, such as the number of layers, number of attention heads, and hidden state size, based on task requirements and data characteristics. The training module is used to fine-tune the pre-trained large model using a text dataset of text data in the construction field to form a large language model for construction engineering; The evaluation and optimization module is used to continuously evaluate the performance of the model during the training process, using the validation set and test set to detect the accuracy, robustness, and generalization ability of the model; The deployment and application module is used to deploy the trained large language model of construction engineering into actual applications and provide services in the form of question and answer. The data collection and preprocessing module, model architecture selection and design module, training module, evaluation and optimization module, and deployment and application module all run in the computer system as software or algorithm components. The functions and processes of these modules are implemented and executed by software on the computer hardware platform. Information analysis and processing module: extract key information from basic engineering information and design drawings, such as engineering scale, structural characteristics, technical requirements, and construction difficulties, and analyze the core content and key links required to be included in the construction plan based on the type of construction plan and specific project requirements; Catalog structure planning module: Based on the results of information analysis and demand analysis, the catalog framework of the construction plan is designed, and the various parts in the catalog are prioritized according to the importance and implementation sequence of the construction plan; Review and optimization module: used for the initial plan content of each secondary directory in the construction plan, as well as the construction engineering knowledge graph and large language model to review the quality of the initial content; the functions and processes of these modules are implemented and executed through software on the computer hardware platform.

[0008] Furthermore, in step S101, text data collected in the construction field include construction laws and regulations, specifications and atlases, construction plans, industry reports, academic papers, professional books, technical manuals, and news reports. The data is of high quality and diversity to ensure that the model can learn the characteristics of the industry language and professional knowledge.

[0009] Furthermore, in step S102, the data is cleaned and preprocessed including removing noise, standardizing terms, segmenting words, and removing stop words to improve the usability of the data.

[0010] Compared with the prior art, the present invention has the following beneficial effects: 1. The technical solution of this application realizes the intelligent generation of construction plans for construction projects through advanced technical means, effectively filling the technical gap in this field, promoting the automation and intelligence process of the construction industry, and combining advanced large language model technology and knowledge graph technology to build a complete technical framework and intelligent processing system, which significantly saves the time and energy required in the manual compilation process and improves work efficiency.

[0011] 2. The technical solution of this application realizes the high efficiency and high quality of intelligent generation of construction plans by designing a comprehensive processing flow, including key links such as intelligent generation of catalogs, intelligent generation of content, and in-depth quality analysis of content, thereby greatly improving the efficiency and quality level of construction plan preparation.

[0012] 3. The technical solution of this application organically combines project engineering information and construction standards and specifications by utilizing knowledge graph technology, ensuring the professionalism and standardization of the intelligent generation of construction plans, and greatly improving the accuracy and reliability of construction plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 A system composition diagram of the construction scheme of the present invention integrating a large language model and a knowledge graph; Figure 2 This is a processing flow chart of the construction plan generation method and system that integrates a large language model and a knowledge graph in the present invention. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0015] Embodiment 1, as Figure 1-2 As shown, the present invention provides a technical solution: a construction plan generation method integrating a large language model and a knowledge graph, comprising the following specific steps: Collecting text data in the field of construction requires comprehensive use of official channels, industry associations, academic institutions, library resources, news reports, social media, professional consulting companies, and participation in academic exchanges and conferences. Through these methods, rich text data such as construction laws and regulations, specifications and atlases, construction plans, industry reports, academic papers, professional books, technical manuals, and news reports can be systematically obtained; Before importing text data into the model, it is necessary to clean and preprocess the text data, including removing noise, standardizing terms, segmenting, and removing stop words to improve data usability. Then a series of preprocessing operations are required to ensure that the data format meets the requirements of the model input: the original text is segmented into tokens that the model can understand. This process requires the use of a specific tokenizer, which converts the text into a token ID sequence based on the vocabulary of the model's training data set, and converts the token ID sequence after segmentation into a tensor format that the model can process. This usually involves converting the token ID sequence into a fixed-length tensor and padding the insufficient length sequence to ensure that all input sequences have the same length; Before importing text data into the model, you need to ensure that you have prepared an open source large language model with a suitable Transformer architecture. Select a suitable Transformer model according to the task requirements, such as BERT, GPT, etc. These models have pre-trained weights and can be directly used for text processing tasks. Use a suitable library (such as Hugging Face's Transformers library) to load the pre-trained Transformer model. When loading the model, you need to load the word segmenter that matches the model. Convert the preprocessed text data into an input format that the model can accept, which involves passing the processed tensor to the model's tokenizer for final encoding processing to generate an input tensor that the model can process. After preparing the data and model, you can use the model to process the text data. If you have not yet created a model instance, you need to create one first, pass the prepared input tensor to the model, and call the model's prediction function (such as generate, forward, etc.) to get the output. According to the output type of the model (such as text, label, etc.), perform corresponding post-processing operations. For example, if the output is a text sequence, you may need to use the tokenizer's decoding function to convert the output tensor back to text format; Fine-tuning is the process of adapting a pre-trained model to a specific task or domain. When using the architectural text dataset for fine-tuning, the following strategies are adopted: Fine-tune all layers: All layers of the pre-trained model are fine-tuned to fully adapt to specific tasks in the architecture field; Fine-tuning the top layer: Fine-tuning only the top layer of the pre-trained model, while keeping the bottom layer parameters unchanged. This method can quickly adapt to new tasks while retaining most of the model's knowledge; Freeze the bottom layer: The bottom layer of the pre-trained model is fixed and only the top layer is fine-tuned. This method helps maintain the performance of the model on the original task while achieving some improvement on the new task. During fine-tuning, it is necessary to select appropriate fine-tuning strategies according to the characteristics of specific tasks and datasets, and set reasonable hyperparameters (such as learning rate, batch size, training rounds, etc.); After fine-tuning is completed, the model needs to be evaluated using an independent architectural test dataset. Evaluation indicators can include accuracy, recall, F1 score, etc. to comprehensively measure the performance of the model on architectural tasks. Based on the evaluation results, the model can be further adjusted and optimized to improve its performance; After full evaluation and adjustment, the fine-tuned large language model for construction engineering can be deployed to actual application scenarios. These application scenarios can include architectural design assistance, automatic generation of construction drawings, and building code query. Through actual application, the performance and practicality of the model can be further verified, and continuous improvement and optimization can be carried out based on feedback. During the training process, the performance of the large language model for construction engineering can be continuously evaluated, and the accuracy, robustness, and generalization ability of the model can be detected using validation sets and test sets. The trained large language model for construction engineering is deployed to actual applications and provides services to the outside in the form of questions and answers, including knowledge questions and answers on construction plans, knowledge questions and answers on laws and regulations, engineering knowledge questions and answers, knowledge questions and answers on regulatory atlases, and online learning application scenarios. The system receives the intelligent generation instruction of the construction plan of the building project issued by the user. The instruction should contain the basic information of the project and the type of construction plan to be generated. The system makes a preliminary analysis of the instruction to determine the scope and focus of the information to be queried, including the basic information of the building project, design drawings and the specific engineering information required for a specific type of construction plan. According to the requirements in the user's instruction, the type of construction plan to be generated is clarified, such as the foundation construction plan, the main structure construction plan, the decoration and renovation construction plan, etc. The queried basic information of the building project, the design drawings and the engineering information required for a specific type of construction plan are integrated to form a complete set of building project information. The integrated information is cleaned to remove duplicate, erroneous or invalid data, and is verified to ensure the accuracy and completeness of the information. The cleaned and verified information is stored in the system's database for subsequent construction plan generation and query use. According to the type of construction plan and the required engineering information, a first-level directory question statement of the large language model of the construction project is constructed and input into the large language model of the construction project to obtain the first-level directory of the construction plan and the priority of the first-level directory structure. For each first-level directory of the construction plan, according to the type of construction plan and the required engineering information, a second-level directory question statement of the large language model of the construction project is constructed and input into the large language model of the construction project to obtain the second-level directory of the construction plan and the priority of the second-level directory structure; For each secondary directory in the construction plan, the corresponding engineering information required for the secondary directory is queried from the construction engineering information set. For each secondary directory in the construction plan, and based on the corresponding engineering information, the relevant engineering knowledge is queried from the construction engineering knowledge graph to obtain the corresponding knowledge graph data. For each secondary directory in the construction plan, and based on the corresponding engineering information and knowledge graph data, a large language model question statement is formed. For each secondary directory in the construction plan, the formed large language model question statement is input into the building large language model, and the solution content of the secondary directory is obtained to form the initial solution content of the secondary directory. All secondary directories of the construction plan are generated with initial solution content according to step S404 to form the overall initial solution content of the construction plan. Finally, the quality of the construction plan content is checked and optimized, including the following sub-steps: the initial plan content of each secondary directory in the construction plan, as well as the construction engineering knowledge graph and the large language model are reviewed to review the quality of the initial content; from the construction engineering knowledge graph, the relevant specifications and requirements of the building are queried; based on the relevant specifications and requirements, whether the initial plan content of the secondary directory is qualified is judged; if qualified, the subsequent steps are continued; if unqualified, the unqualified information is input into the large language model question statement to form a new large language model question statement, the initial content of the directory is regenerated, and the initial content quality is reviewed again; after the initial plan content of each secondary directory of the construction plan meets the quality review, an initial qualified construction plan is formed. For all the initial qualified plan contents of the construction plan, analyze the consistency of the initial qualified contents between different directories. If they are unqualified, combine the first-level directory priority and the second-level directory priority. For the inconsistent initial qualified plan contents, re-form the low-priority initial qualified contents into the large language model question statement, and add the high-priority initial qualified contents to the question statement, regenerate the initial content of the low-priority directory, and re-examine the initial content quality and the consistency of the initial qualified content. For the initial qualified contents of each second-level directory of the construction plan, after they meet the consistency review, the qualified construction plan contents are formed. According to the construction plan output format requirements, the qualified construction plan contents are used to output the construction plan.

[0016] The construction plan generation system that integrates the large language model and knowledge graph includes: The data collection and preprocessing module is used to collect text data in the field of architecture and perform data cleaning and preprocessing. A large amount of text data is collected, and the data should cover a wide range of topics and fields to ensure the generalization ability of the model. Data sources can include public data sets, web crawling, social media data, etc. The collected data is cleaned, deduplicated, segmented, tokenized, etc., so that the model can understand and process the data. In addition, the data needs to be formatted and encoded so that the model can learn effectively; The model architecture selection and design module is used to select an open source large language model with a Transformer architecture. According to task requirements and data characteristics, design appropriate model parameters and configurations, such as the number of layers, number of attention heads, and hidden state size. Select an appropriate model architecture, such as the Transformer-based architecture, which is the mainstream choice for current large language models. The Transformer architecture can capture long-range dependencies in text through self-attention mechanisms and multi-head attention mechanisms. According to task requirements and data characteristics, design appropriate model parameters and configurations, such as the number of layers, number of attention heads, and hidden state size. The selection of these parameters will directly affect the performance and training efficiency of the model; The training module is used to fine-tune the pre-trained large model using a text dataset of text data in the construction field to form a large language model for construction engineering, and to build a computing environment suitable for large model training, including hardware resources such as high-performance computing clusters, GPUs or TPUs. At the same time, it is also necessary to configure the corresponding software environment, such as deep learning frameworks (such as TensorFlow, PyTorch) and model training libraries (such as Hugging Face Transformers), and use pre-processed data to train the model. The training process usually includes multiple stages, such as pre-training and fine-tuning. The pre-training stage aims to allow the model to learn general language knowledge, while the fine-tuning stage uses data from specific fields to adjust the model's behavior to adapt it to specific tasks; The evaluation and optimization module is used to continuously evaluate the performance of the model during the training process. The validation set and test set are used to detect the accuracy, robustness and generalization ability of the model. The trained model is evaluated using the test data set to verify its performance and generalization ability. The evaluation indicators may include accuracy, recall rate, F1 score, etc. The model is optimized according to the evaluation results, including adjusting model parameters, improving model architecture, optimizing training algorithms, etc. In addition, regularization, dropout and other technologies can be used to prevent model overfitting; The deployment and application module is used to deploy the trained large language model of construction engineering to actual applications, provide services to the outside in the form of questions and answers, and deploy the trained model to actual application scenarios, such as intelligent customer service, text generation, question and answer systems, etc. The deployment process needs to consider issues such as model compatibility, performance, and security, and develop corresponding applications or interfaces based on actual needs so that users can easily use the model for tasks such as text generation and understanding; The data collection and preprocessing module, model architecture selection and design module, training module, evaluation and optimization module, and deployment and application module all run in the computer system as software or algorithm components. The functions and processes of these modules are implemented and executed by software on the computer hardware platform. Information analysis and processing module: extract key information from basic engineering information and design drawings, such as engineering scale, structural characteristics, technical requirements, and construction difficulties, and analyze the core content and key links required to be included in the construction plan based on the type of construction plan and specific project requirements; Catalog structure planning module: Based on the results of information analysis and demand analysis, the catalog framework of the construction plan is designed, and the various parts in the catalog are prioritized according to the importance and implementation sequence of the construction plan; Review and optimization module: used for the initial plan content of each secondary directory in the construction plan, as well as the construction engineering knowledge graph and large language model to review the quality of the initial content; the functions and processes of these modules are implemented and executed through software on the computer hardware platform.

[0017] In summary, the present invention innovatively proposes a construction plan intelligent generation method and system, which realizes the intelligent generation of construction plans for building projects through advanced technical means, effectively fills the technical gap in this field, and promotes the automation and intelligentization process of the construction industry.

Claims

1. A construction plan generation method and system integrating a large language model and a knowledge graph, characterized by: The steps include: S100 trains a large language model for construction engineering, specifically including the following sub-steps: S101: collect text data in the construction field; S102: Cleaning and preprocessing the text data in the architectural field; S103: Using the open source big language model of the Transformer architecture, the processed text data in the construction field is imported into the open source big language model of the Transformer architecture. Based on the open source big language model, the text data set of the text data in the construction field is used for fine-tuning to form a big language model for construction engineering; S200: after receiving the intelligent generation instruction of the construction plan of the building project issued by the user, query and obtain the basic information and design drawings of the building project, and query and obtain the engineering information required for the construction plan according to the type of the construction plan to be generated, so as to form a set of building project information; S300 generates a construction plan catalog, specifically including the following sub-steps, S301: constructing a first-level catalog question statement of the large language model of the building project according to the type of the construction plan and the required engineering information, inputting it into the large language model of the building project, and obtaining the first-level catalog of the construction plan and the first-level catalog structure priority; S302: for each first-level directory of the construction plan, construct a second-level directory question statement of the building engineering large language model according to the type of the construction plan and the required engineering information, input it into the building engineering large language model, and obtain the second-level directory of the construction plan and the priority of the second-level directory structure; S400 generates the initial plan content according to the directory structure, specifically including the following sub-steps S401: for each secondary directory in the construction plan, query the corresponding engineering information required for the secondary directory from the construction engineering information set; S402: for each secondary directory in the construction plan, and based on the corresponding engineering information, query the relevant engineering knowledge from the construction engineering knowledge graph to obtain the corresponding knowledge graph data; S403: For each secondary directory of the construction plan, and based on the corresponding engineering information and knowledge graph data, a large language model question sentence is formed; for each secondary directory of the construction plan, the formed large language model question sentence is formed; S404: inputting into the architectural language model, and obtaining the solution content of the secondary directory, forming the initial solution content of the secondary directory; S405: generating initial plan contents for all secondary directories of the construction plan according to step S404 to form the overall initial plan contents of the construction plan; S500 checks and optimizes the quality of the content of the construction plan, specifically including the following sub-steps, S501: reviewing the quality of the initial content of each secondary directory in the construction plan, as well as the construction engineering knowledge graph and the large language model; S502: After the initial plan contents of each secondary directory of the construction plan meet the quality review, the initial qualified construction plan contents are formed; S503: Analyze the consistency of the initial qualified contents between different catalogs for all the initial qualified contents of the construction plan; S504: If unqualified, combining the priorities of the first-level directory and the second-level directory, for the inconsistent initial qualified solution content, the low-priority initial qualified content is re-formed into a large language model question statement, and the high-priority initial qualified content is added to the question statement, the initial content of the low-priority directory is regenerated, and the initial content quality and initial qualified content consistency are re-examined; S505: After the initial qualified contents of each secondary directory of the construction plan meet the consistency review, the qualified construction plan contents are formed; S600: Output the construction plan according to the construction plan output format requirements and using qualified construction plan content.

2. The construction plan generation method integrating a large language model and a knowledge graph according to claim 1 is characterized in that: The step S100 also includes step S104: continuously evaluating the performance of the large language model for construction engineering during the training process, and using a validation set and a test set to detect the accuracy, robustness and generalization ability of the model.

3. The construction plan generation method integrating a large language model and a knowledge graph according to claim 2 is characterized in that: The step S100 also includes step S105: deploying the trained large language model of construction engineering into practical applications, and providing services to the outside in the form of questions and answers, including questions and answers on construction plan knowledge, questions and answers on laws and regulations, questions and answers on engineering knowledge, questions and answers on legal atlas knowledge, and online learning application scenarios.

4. The construction plan generation method integrating a large language model and a knowledge graph according to claim 1 is characterized in that: The step S501 also includes: querying relevant specifications and requirements of the building from the construction engineering knowledge graph, and judging whether the initial plan content of the secondary directory is qualified based on the relevant specifications and requirements; if qualified, continuing to execute the subsequent step S502; if unqualified, inputting the unqualified information into the large language model question statement to form a new large language model question statement, regenerating the initial content of the directory, and re-examining the quality of the initial content.

5. The construction plan generation method integrating a large language model and a knowledge graph according to claim 1 is characterized in that: The step S503 also includes querying related contents between different directories in all initial qualified solution contents, such as the construction process chapter contains the materials used, and the material preparation chapter also contains the materials, to form chapter pairs with related contents, such as construction materials, construction process chapter, and material preparation chapter. For all chapter pairs with related contents, it is determined whether the contents of different chapters are consistent. For example, if the construction process chapter contains material A, and the material preparation chapter also contains material A, then the contents are determined to be consistent. Otherwise, they are inconsistent. According to the above steps, it is determined whether the consistency review of the initial qualified content is qualified. If qualified, execute the subsequent sub-step S505. If unqualified, execute the subsequent sub-step S504.

6. The construction plan generation method integrating a large language model and a knowledge graph according to claim 1 is characterized in that: The text data collected in the construction field in step S101 includes construction laws and regulations, specifications and atlases, construction plans, industry reports, academic papers, professional books, technical manuals, and news reports. The data is of high quality and diversity to ensure that the model can learn the characteristics of the industry language and professional knowledge.

7. The construction plan generation method integrating a large language model and a knowledge graph according to claim 1 is characterized in that: The data cleaning and preprocessing in step S102 include noise removal, term standardization, word segmentation, and stop word removal to improve the usability of the data.

8. A construction plan generation system integrating a large language model and a knowledge graph, characterized by: It includes data collection and preprocessing modules, which are used to collect text data in the field of architecture and perform data cleaning and preprocessing; The model architecture selection and design module is used to select an open source large language model with a Transformer architecture and design appropriate model parameters and configurations, such as the number of layers, number of attention heads, and hidden state size, based on task requirements and data characteristics. The training module is used to fine-tune the pre-trained large model using a text dataset of text data in the construction field to form a large language model for construction engineering; The evaluation and optimization module is used to continuously evaluate the performance of the model during the training process, using the validation set and test set to detect the accuracy, robustness, and generalization ability of the model; The deployment and application module is used to deploy the trained large language model of construction engineering into actual applications and provide services in the form of question and answer. The data collection and preprocessing module, model architecture selection and design module, training module, evaluation and optimization module, and deployment and application module all run as software or algorithm components in the computer system. The functions and processes of these modules are implemented and executed through software on the computer's hardware platform.

9. The construction plan generation system integrating a large language model and a knowledge graph according to claim 8 is characterized in that: Also includes Information analysis and processing module: extract key information from basic engineering information and design drawings, such as engineering scale, structural characteristics, technical requirements, and construction difficulties, and analyze the core content and key links required to be included in the construction plan based on the type of construction plan and specific project requirements; Catalog structure planning module: Based on the results of information analysis and demand analysis, the catalog framework of the construction plan is designed, and the various parts in the catalog are prioritized according to the importance and implementation sequence of the construction plan; Review and optimization module: used for the initial plan content of each secondary directory in the construction plan, as well as the construction engineering knowledge graph and large language model to review the quality of the initial content; the functions and processes of these modules are implemented and executed through software on the computer hardware platform.

Images