500KV transformer substation automatic BIM modeling method and system based on LLM intelligent agent

Abstract

The invention provides a 500KV transformer substation automatic BIM modeling method based on an LLM intelligent agent, and the method comprises the steps: constructing the LLM intelligent agent based on a large language model LLM and an intelligent agent development technology, and building an overall architecture of transformer substation automatic BIM modeling; obtaining a transformer substation corpus database, training the LLM, and identifying information of the two-dimensional drawing through the trained LLM; selecting a family library model template according to the information of the identified drawing, adjusting the family library model template according to the specific design requirements of the 500KV transformer substation, and storing the family library model template as a 500KV family library; and obtaining model data in the 500KV family library, and performing automatic BIM modeling of the 500KV transformer substation according to the LLM intelligent agent. According to the method, the LLM and the intelligent agent are combined, the LLM recognizes the two-dimensional drawing, the intelligent agent participates in 500KV family library modeling, the intelligent agent is constructed and participates in three-dimensional model generation, automatic BIM modeling of the transformer substation is achieved, and the problems that in the prior art, modeling efficiency is low, precision is poor, manual intervention is much, coordination is insufficient, and full-life-cycle management is difficult to achieve are solved.

Description

Technical Field

[0001] The present application relates to the technical field of automatic BIM modeling of substations, and in particular, to an automatic BIM modeling method and system for a 500KV substation based on an LLM intelligent agent. Background Art

[0002] As the demand for digital technology in substation construction and operation continues to increase, related modeling technologies are also constantly developing.

[0003] Traditional substation engineering technology mostly uses CAD technology to complete computer-aided drawing. CAD technology meets the needs of complex component drawing to a certain extent, but it has many disadvantages. It cannot support the entire industrial chain. The correlation between different fields and links in the design process is not strong. It is impossible to intuitively obtain different parameters of each link in the project. It is difficult to effectively manage and monitor different facilities, which is not conducive to the collaborative work between different professions. Moreover, drawing is relatively inconvenient, fast and accurate, which may lead to an extension of the project cycle and affect work efficiency.

[0004] In recent years, BIM technology has gradually been applied in the field of substations. Some studies have used Bentley Substation software to establish the overall BIM model of the substation, which has made certain progress compared to CAD technology. It can more intuitively obtain different parameters of each link in the substation project, manage and monitor different facilities, and effectively ensure the collaborative work between different professions. At the same time, compared with traditional CAD technology, it is more convenient, fast and accurate, which can shorten the project cycle and improve work efficiency. It can achieve full life cycle coverage of substation construction, and can simultaneously realize two-dimensional and three-dimensional model display, including all-round information of different parts and facilities of the substation, and can transfer two-dimensional drawings and data at any time, as well as complete construction simulation, model improvement and optimization according to the construction progress.

[0005] The application number is: CN201510689250.5. The name: A multi-person online smart substation design system, which is equipped with a configuration management module, a drawing library module, a drawing management module, a retrieval module, a verification module, a version management module, a smart substation design drawing module and a database module. It is also equipped with a permission management module, a user management module, a project management module, a communication module and a function management module; the permission management module is used to authenticate the permission of the operation request issued by the user; the user management module is used to manage user permissions; the project management module is used for users to open or delete smart substation projects; the communication module is used to receive the permission authentication information sent by the permission management module and send it to the user, and is also used to receive the module conflict information sent by the function management module and send it to the user; the function management module is used to coordinate the operation and handle the conflicts generated when multiple users operate the same module at the same time, which relies on manual operation, is time-consuming and prone to errors. At the same time, model errors may be caused by inaccurate manual reading of drawings. Summary of the invention

[0006] In view of the defects in the prior art, the purpose of this application is to provide a 500KV substation automatic BIM modeling method and system based on LLM intelligent agent, which utilizes the technology combining large language model (LLM) and intelligent agent to realize three-dimensional BIM modeling of 500KV substation through the recognition of two-dimensional drawings and the calling of relevant tools by the intelligent agent.

[0007] In one aspect of the present application, a 500KV substation automatic BIM modeling method based on LLM agent is provided, comprising:

[0008] Based on the large language model (LLM) and agent development technology, an LLM agent is constructed to build the overall architecture of automatic BIM modeling of substations;

[0009] A substation corpus database is obtained to train the LLM, and the LLM agent recognizes information of the two-dimensional drawing through the trained LLM;

[0010] According to the information of the identified drawing, the LLM agent selects a family library model template, adjusts the family library model template according to the specific 500KV substation design requirements, and saves it as a 500KV family library;

[0011] The model data in the 500KV family library is obtained, and automatic BIM modeling of the 500KV substation is performed according to the constructed LLM agent.

[0012] Furthermore, the substation corpus database is obtained, and the LLM is trained, and the LLM agent recognizes the information of the two-dimensional drawing through the trained LLM, including:

[0013] Obtain a substation corpus, select a large language model LLM, and perform fine-tuning training on the LLM;

[0014] Obtaining the substation drawing data to be identified, converting the drawing data into a digital format, and generating a corresponding text description;

[0015] The text description is input into the fine-tuned LLM for text encoding and semantic understanding, and key information is output in a structured form.

[0016] Furthermore, the step of obtaining a substation corpus, selecting a large language model LLM, and fine-tuning the LLM includes:

[0017] Acquire the substation corpus database, and preprocess the corpus database to generate data that meets the LLM training requirements;

[0018] According to the preprocessed corpus database, the LLM is fine-tuned and trained to learn substation design methods and rules.

[0019] Further, the obtaining of the substation drawing data to be identified, converting the drawing data into a digital format, and generating a corresponding text description includes:

[0020] Acquire the substation drawing data to be identified, and convert the drawing data into a digital image format;

[0021] Using image recognition technology, the digital image is converted into a pixel value array, and key feature information in the drawing is extracted from the pixel value array;

[0022] Generate corresponding text description based on the extracted key feature information.

[0023] Furthermore, the text description is input into the fine-tuned LLM for text encoding and semantic understanding, and key information is output in a structured form, including:

[0024] According to the generated text description, input it into the fine-tuned LLM for text encoding; the text encoding is: converting the text description into a vector representation understandable by the model;

[0025] Using the natural language processing capability of the LLM, the text description is semantically understood, deep semantic information in the text description is extracted, and more in-depth analysis and reasoning are performed in combination with context information;

[0026] Key information is extracted from the text description of the substation drawing data, and output in a structured form based on the extracted key information.

[0027] Further, according to the information of the identified drawing, the LLM agent selects a family library model template, adjusts the family library model template according to the specific 500KV substation design requirements, and saves it as a 500KV family library, including:

[0028] Acquire the information of the drawing, and according to the information of the drawing, the LLM agent selects the corresponding equipment and structure family library model template;

[0029] By using the 3D modeling software Revit, parameterizing the family library model template according to the information of the drawing;

[0030] Testing the established family library model to check whether relevant parameters of the family library model meet the requirements of the drawings;

[0031] According to the test results, the family library model is adjusted and optimized;

[0032] The adjusted and optimized family library model and related model data are saved.

[0033] Further, the acquiring of the model data in the 500KV family library and the automatic BIM modeling of the 500KV substation according to the constructed LLM agent include:

[0034] According to the model data in the 500KV family library, data related to the target three-dimensional model is collected, and the related data is preprocessed; the related data includes: CAD drawings and their text descriptions or attribute data; the preprocessing includes: data cleaning, data labeling and data format conversion;

[0035] A large language model (LLM) is selected, training parameters are set to train the large language model (LLM), the trained model is used for reasoning, three-dimensional model coordinate points are generated, and the three-dimensional model is converted into a three-dimensional model.

[0036] Further, the selecting a large language model (LLM), setting training parameters to train the large language model (LLM), using the trained model for reasoning, generating three-dimensional model coordinate points, and converting into a three-dimensional model, includes:

[0037] Select a large language model (LLM) and set the training parameters according to the model architecture and the characteristics of the dataset;

[0038] The large model is trained according to the preprocessed data to generate a trained model; wherein during the training process, the intelligent agent can provide feedback according to the learned knowledge and experience and adjust the training parameters;

[0039] Get new target objects or scenes and prepare input data;

[0040] The input data is input into the trained model for reasoning to generate coordinate points, and the LLM agent calls a three-dimensional BIM modeling tool to convert the generated coordinate points into a three-dimensional model.

[0041] Furthermore, after converting the generated coordinate points into a three-dimensional model, the method further includes:

[0042] The intelligent agent evaluates the quality of the generated three-dimensional model and performs iterative optimization according to the evaluation results.

[0043] The second aspect of the present application provides a 500KV substation automatic BIM modeling system based on LLM agent, including:

[0044] The architecture building module is used to build LLM agents based on the large language model (LLM) and agent development technology, and to build the overall architecture of automatic BIM modeling of substations;

[0045] A training module is used to obtain a substation corpus database and train the LLM, and the LLM agent recognizes the information of the two-dimensional drawing through the trained LLM;

[0046] A test adjustment module, which is used for the LLM agent to select a family library model template according to the information of the identified drawing, and adjust the family library model template according to the specific 500KV substation design requirements, and save it as a 500KV family library;

[0047] The output module is used to obtain the model data in the 500KV family library and perform automatic BIM modeling of the 500KV substation based on the constructed LLM agent.

[0048] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0049] 1. This application uses the technology of combining large language model (LLM) with intelligent agent, recognizes two-dimensional drawings and calls related tools (such as Revit) by intelligent agent to realize three-dimensional BIM modeling of 500KV substation. It not only inherits the advantages of BIM technology, but also improves the automation and accuracy of modeling through the application of LLM intelligent agent. When recognizing two-dimensional drawings, LLM can accurately understand and extract the information in the drawings, providing a more accurate data basis for subsequent modeling. At the same time, the intelligent agent can automatically call the Revit tool for modeling operations based on the information extracted by LLM, reducing manual intervention and improving modeling efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0051] Figure 1 This is a flow chart of an automatic BIM modeling method for a 500KV substation based on an LLM agent in one embodiment of the present application.

[0052] Figure 2 This is an overall architecture diagram of a 500KV substation automatic BIM modeling method based on LLM agent in one embodiment of the present application.

[0053] Figure 3 This is a flowchart of two-dimensional drawing recognition in one embodiment of the present application.

[0054] Figure 4 A three-dimensional model diagram is established for the intelligent agent in one embodiment of the present application. DETAILED DESCRIPTION

[0055] The present application is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any form. It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present application, several variations and improvements can also be made. These all belong to the protection scope of the present application. The parts not described in detail in the following embodiments can be implemented with reference to the prior art.

[0056] Reference Figure 1 As shown, an automatic BIM modeling method for a 500KV substation based on an LLM agent according to an embodiment of the present application includes:

[0057] S100, build LLM agent based on large language model LLM and agent development technology, and build the overall architecture of substation automatic BIM modeling;

[0058] S200, obtaining a substation corpus database, training the LLM, and allowing the LLM agent to recognize information on the two-dimensional drawing through the trained LLM;

[0059] S300, according to the information of the recognized drawing, the LLM agent selects a family library model template, adjusts the family library model template according to the specific 500KV substation design requirements, and saves it as a family library;

[0060] S400, obtain the model data in the 500KV family library, and perform automatic BIM modeling of the 500KV substation based on the constructed LLM intelligent agent.

[0061] This application uses the technology of combining large language model (LLM) with intelligent agent, recognizes two-dimensional drawings and calls related tools (such as Revit) by intelligent agent to realize three-dimensional BIM modeling of 500KV substation. It not only inherits the advantages of BIM technology, but also improves the automation and accuracy of modeling through the application of LLM intelligent agent. When recognizing two-dimensional drawings, LLM can accurately understand and extract the information in the drawings, providing a more accurate data basis for subsequent modeling. At the same time, LLM intelligent agent can automatically call Revit tool for modeling operation according to the information extracted by LLM, reducing manual intervention and improving modeling efficiency.

[0062] Specifically, firstly, an overall architecture for automatic BIM (Building Information Model) modeling of 500KV substation was constructed by combining LLM (Large Language Model) with intelligent agent; then, the corpus database related to the substation was obtained, and the LLM was trained with these data so that it could accurately identify the information in the two-dimensional drawings; then, according to the information of the identified drawings, the family library model template was selected, and the family library model template was adjusted according to the specific 500KV substation design requirements and saved as a family library; finally, the model data was extracted from the 500KV family library, and the three-dimensional BIM modeling of the 500KV substation was completed through the intelligent agent.

[0063] Among them, the innovation of this architecture is to combine LLM with intelligent agents to realize automatic BIM modeling of substations, which is different from the existing technology that relies solely on manual or traditional methods. By recognizing two-dimensional drawings and calling related tools (such as Revit) by intelligent agents, three-dimensional BIM modeling of 500KV substations is realized. The key difference from traditional modeling methods is that LLM provides powerful natural language processing capabilities for intelligent agents, enabling them to better adapt to and process drawing information.

[0064] The present invention is based on a unique architecture that combines LLM with intelligent agents. First, it is determined that the architecture must meet the needs of automatic BIM modeling of substations, and the natural language processing capabilities of LLM and the execution capabilities of intelligent agents are comprehensively considered. By organically combining the two, the system can recognize two-dimensional drawing information and use intelligent agents to call related tools (such as Revit) for three-dimensional modeling. Figure 2 shown.

[0065] In actual tests, the architecture has demonstrated good adaptability and efficiency. For example, for a complex substation project, it can quickly and accurately identify key elements in the drawings and start the intelligent agent to perform corresponding modeling operations, which greatly shortens the pre-modeling preparation time compared to traditional methods.

[0066] In some specific embodiments, a substation corpus database is obtained, an LLM is trained, and information on two-dimensional drawings is recognized through the trained LLM, including: obtaining a substation corpus, selecting a large language model LLM, and fine-tuning the LLM; obtaining substation drawing data to be recognized, converting the drawing data into a digital format, and generating a corresponding text description; inputting the text description into the fine-tuned LLM for text encoding and semantic understanding, and outputting key information in a structured form.

[0067] Among them, the substation corpus database is a database related to substation BIM modeling, including: substation design documents, engineering reports, annotations and instructions in substation drawings and other data.

[0068] By carefully selecting a suitable large language model (LLM) and combining it with fine-tuning training technology, the LLM's ability to recognize substation two-dimensional drawing information is improved. The drawing data is converted into a digital format and a text description is generated. The text description is then input into the fine-tuned LLM for processing. The key information in the drawing is extracted efficiently and accurately and output in a structured form, providing precise data support for subsequent substation BIM modeling and improving modeling efficiency and accuracy.

[0069] like Figure 3 As shown in the figure, based on the characteristics of the substation field, considering factors such as application scenarios, technical feasibility, resource constraints, performance requirements and long-term development plans, a suitable large language model (LLM) is selected. The LLM is fine-tuned and trained, and a large amount of substation-related corpus data, including design documents, engineering reports, etc., is collected and pre-processed to learn language rules and semantic information. Then fine-tuning is performed, and the model is further optimized using specific task data and instruction data so that it can better recognize substation drawing information.

[0070] Through actual operation verification, it is possible to accurately extract drawing information. For example, in the recognition of a substation primary wiring diagram, the connection relationship and parameter information of each device can be accurately identified, avoiding the errors that may occur in manual recognition.

[0071] Specifically, the substation corpus is obtained, and a suitable large language model (LLM) is selected according to the application scenario, technical feasibility, resource constraints, performance requirements, and long-term development plan. The LLM is then fine-tuned and trained, including: obtaining a corpus database related to substation drawings (such as annotations and instructions in the drawings), and preprocessing the corpus database (including cleaning and sorting) to generate data that meets the requirements for LLM training; based on the preprocessed corpus database, the LLM is fine-tuned and trained to learn substation design methods and rules (including learning the statistical rules, semantic information, and contextual relationships of the language, and performing vocabulary construction and text vectorization training, etc.), so that the LLM can better understand and execute relevant instructions.

[0072] By collecting specific task data related to substations (i.e., corpus data), including design drawings and technical specifications, meticulous data annotation and cleaning (i.e., preprocessing) are performed to ensure data quality. Subsequently, key information is extracted through feature engineering, and appropriate fine-tuning parameters such as learning rate and batch size are set to fine-tune the selected LLM. During the fine-tuning process, the validation set is used to monitor the model performance to prevent overfitting, and after the fine-tuning is completed, the final performance of the model is evaluated through the test set. Finally, the output results of the model are analyzed to ensure that the model can accurately perform the BIM modeling task of the substation.

[0073] Specifically, in the working process, the large language model (LLM) is selected according to factors such as application scenarios (such as substation drawing recognition), technical feasibility (such as whether the complexity of the model matches the available computing resources), resource constraints (such as budget and time), performance requirements (such as recognition accuracy and speed), and long-term development plans (such as the scalability and adaptability of the model). Different from the limitations of model selection in traditional methods, it pays more attention to the practicality and foresight of the model.

[0074] Fine-tune the LLM and further fine-tune it using data specific to the substation drawing recognition task. This includes collecting instruction data such as annotations and instructions in the drawings and using this data to fine-tune the LLM so that it can more accurately understand and execute specific instructions and operations related to substation BIM modeling.

[0075] Through the above process, compared with the traditional substation BIM modeling method that relies on manual operation, is time-consuming and prone to errors, this application aims to use LLM intelligent agent technology to reduce manual intervention and improve modeling efficiency through automatic recognition of two-dimensional drawings and the call of related tools by the intelligent agent. For example, in traditional modeling, manual recognition of drawing information and input into the modeling tool is a tedious and time-consuming process, while this application can realize automated processing and greatly shorten the modeling time.

[0076] Traditional modeling may lead to model errors due to inaccurate manual reading of drawings. This application uses LLM's ability to accurately extract and understand drawing information to provide an accurate data basis for 3D modeling, ensuring that the model can accurately reflect the actual structure, equipment configuration, and line connection details of the substation, and achieve accurate extraction and understanding of drawing information, solving the problem of model errors caused by inaccurate manual reading of drawings.

[0077] Specifically, obtaining the substation drawing data to be identified, converting the drawing data into a digital format, and generating a corresponding text description includes: obtaining the substation drawing data to be identified, converting the drawing data into a digital image format; using image recognition technology to convert the digital image into a pixel value array, and extracting key feature information in the drawing from the pixel value array; generating a corresponding text description based on the extracted key feature information.

[0078] During operation, the substation drawings are converted into a digital format that can be processed by LLM, and the images are converted into pixel value arrays using convolutional neural networks (CNNs) or other image recognition technologies. Key feature information is extracted and corresponding text descriptions are generated, converting image information into text information for easy processing by LLM.

[0079] This application improves work efficiency by converting substation drawing data into digital format and generating corresponding text descriptions; by extracting key feature information through image recognition technology, it can accurately capture the information of the drawing, avoid information loss or misunderstanding, and achieve accurate extraction and understanding of the drawing information, solve the problem of model errors caused by inaccurate traditional manual reading of drawings, and achieve the purpose of making the design, construction and maintenance process of substations more intelligent and efficient.

[0080] In some specific embodiments, the text description is input into a fine-tuned LLM for text encoding and semantic understanding, and key information is output in a structured form, including: according to the generated text description, input into the fine-tuned LLM for text encoding; the text encoding is: the LLM converts the text description into a vector representation understandable by the model so that the model can further analyze and process the information; the natural language processing capability of the LLM is used to semantically understand the text description, extract deep semantic information in the text description, and combine context information for deeper analysis and reasoning; extract key information from the text description of the substation drawing data, and output in a structured form based on the extracted key information.

[0081] The fine-tuned LLM performs text encoding and semantic understanding of the text description and outputs key information in a structured form, thus avoiding misunderstanding or omission of information, making key information clearer and more organized, facilitating automated modeling and data analysis, and improving the efficiency and accuracy of substation drawing information processing.

[0082] In some specific embodiments, according to the information of the identified drawings, a suitable related family library model template is selected, and the family library model template is adjusted according to the specific 500KV substation design requirements and saved as a family library, including: obtaining the information of the drawings, and selecting the corresponding equipment and structure family library model template according to the information of the drawings; parametrically adjusting the family library model template according to the information of the drawings through the three-dimensional modeling software Revit; testing the established family library model to check whether the relevant parameters of the family library model meet the requirements of the drawings; adjusting and optimizing the family library model according to the test results; and saving the adjusted and optimized family library model and related model data.

[0083] By accurately identifying the drawing information and selecting the appropriate family library model template in the 500KV family library, combined with the 3D modeling software Revit for parametric adjustment, a substation model that is highly consistent with the drawing can be quickly established; and after testing and adjustment, the model parameters are ensured to be accurate, which improves work efficiency and project quality. The use of intelligent agents to participate in the 500KV family library modeling process can realize parametric modeling and intelligent decision-making, solving the problem of low efficiency and poor flexibility of traditional reliance on manual input parameters.

[0084] Reference Figure 4 As shown, in the above embodiment, the family type is determined to collect parameters, a suitable template is selected in Revit to create a new family, and geometric, electrical and connection parameters are added. The intelligent agent assists in decision-making based on the information extracted by LLM, uses modeling tools to create the 3D shape of the equipment, adds electrical connection points, realizes parametric modeling, and finally saves after testing and adjustment. The equipment model is created efficiently and accurately with good parametric characteristics, and parametric modeling makes later modification and adjustment more convenient, effectively improving work efficiency.

[0085] In some specific embodiments, model data in the 500KV family library is obtained, and three-dimensional BIM modeling of the 500KV substation is performed according to the constructed LLM intelligent agent, including: according to the model data in the 500KV family library, data related to the target three-dimensional model is collected, and the related data is preprocessed; the related data includes: CAD drawings and their text descriptions or attribute data; the preprocessing includes: data cleaning, data annotation and data format conversion; selecting a large language model (LLM), setting training parameters to train the large language model (LLM), using the trained model for reasoning, generating three-dimensional model coordinate points, and converting them into a three-dimensional model.

[0086] By adopting the above processing, model data can be efficiently obtained and used from the 500KV family library, combined with the pre-processing capabilities of the intelligent agent, ensuring the accuracy and availability of the data. The selection of the large language model (LLM) and training improves the efficiency and accuracy of 3D BIM modeling. At the same time, using the trained model for reasoning, the coordinate points of the 3D model can be automatically generated and converted into an intuitive 3D model, which greatly simplifies the modeling process, reduces the cost of manual intervention, and improves the degree of automation of modeling.

[0087] Specifically, a large language model (LLM) is selected, and training parameters are set to train the large language model (LLM), and the trained model is used for reasoning to generate three-dimensional model coordinate points, and converted into a three-dimensional model, including: selecting a large language model (LLM), and setting training parameters (including learning rate, batch size, number of training rounds, optimization algorithm, loss function, regularization parameter, etc.) according to the characteristics of the model architecture and the data set; training the large model according to the preprocessed data to generate a trained model (the trained model is a machine learning or deep learning model for reasoning and generating three-dimensional coordinate points); acquiring a new target object or scene and preparing input data; inputting the input data into the trained model for reasoning to generate coordinate points, and converting the generated coordinate points into a three-dimensional model.

[0088] Specifically, during work, various data related to the target 3D model are collected, including substation CAD drawings and other related text descriptions or attribute data; the collected data are preprocessed, including data cleaning, data annotation and data format conversion. During the data preprocessing process, the intelligent agent can perform auxiliary operations based on existing knowledge and rules to improve data quality.

[0089] Next, select the Large Language Model (LLM) and set the training parameters to ensure that the model achieves good performance on the training set without overfitting. During the training process, the agent can provide feedback based on the learned knowledge and experience, adjust the training parameters, and further optimize the model performance. Adaptive optimization algorithms such as Adam, Adagrad, RMSprop, etc. are used to adjust the learning rate (i.e., step size) of the training parameters. The step size is dynamically adjusted based on the historical information of the gradient so that different parameters can converge at different speeds.

[0090] Furthermore, use the preprocessed data for training and regularly evaluate the performance indicators of the model on the training set and validation set. If it is found that the model performs well on the training set but the performance deteriorates on the validation set, take corresponding measures to adjust it.

[0091] Finally, prepare the input data for the target object or scene, enter the text description after ensuring the format and preprocessing meet the requirements, and then input the data into the trained large model for reasoning to generate coordinate points. The intelligent agent calls the 3D modeling software to convert the coordinate points into a 3D model.

[0092] When processing new data, the input data for the target object or scene is first prepared, including collecting relevant data and ensuring that its format and preprocessing steps are consistent with those used when training the model. Preprocessing may include data cleaning, normalization, and encoding conversion to ensure data consistency and valid input to the model. Next, the preprocessed data, especially the text description, is input into the trained large model. The model then performs inference and generates coordinate points that define the object or scene in three-dimensional space based on the patterns and features it has learned. Finally, these coordinate points are used to build or render an accurate three-dimensional model.

[0093] In some specific embodiments, after the generated coordinate points are converted into a three-dimensional model, the method further includes: the intelligent agent performs a quality assessment on the generated three-dimensional model and performs iterative optimization based on the assessment results.

[0094] By conducting quality assessment and iterative optimization of the three-dimensional model, efficient and accurate automatic BIM modeling of the substation and optimized design and full life cycle management of related systems are achieved, overcoming the problems of low modeling efficiency, poor accuracy, frequent manual intervention, insufficient system coordination and difficulty in achieving full life cycle management in existing technologies.

[0095] In this application, specifically, LLM is used to deeply analyze the two-dimensional drawings, and the intelligent agent is responsible for calling professional modeling tools (such as Revit). The two work together to complete the three-dimensional BIM modeling task of the 500KV substation, and the two-dimensional drawings are recognized by using the trained LLM model. LLM can accurately parse the text descriptions, symbols and graphic elements in the drawings and convert them into a structured data format. After recognizing the two-dimensional drawings, by building an intelligent agent that can perform complex tasks and interact with BIM software (such as Revit), it can also automatically call the corresponding modeling commands and parameters according to the drawing information analyzed by LLM to generate a three-dimensional BIM model.

[0096] The second aspect of the present application provides a 500KV substation automatic BIM modeling system based on the LLM agent, including: an architecture building module, which is used to build an LLM agent based on a large language model LLM and agent development technology to build an overall architecture for automatic BIM modeling of the substation; a training module, which is used to obtain a substation corpus database and train the LLM, and the LLM agent recognizes the information of two-dimensional drawings through the trained LLM; a test adjustment module, which is used for the LLM agent to select a family library model template according to the information of the recognized drawing, and adjust the family library model template according to the specific 500KV substation design requirements, and save it as a family library; an output module, which is used to obtain model data in the 500KV family library, and perform automatic BIM modeling of the 500KV substation through the constructed LLM agent.

[0097] The overall framework is constructed through the architecture building module, and the training module is used to obtain the substation corpus data to train the LLM so that it can recognize the two-dimensional drawing information; then, the test and adjustment module establishes the family library model in the 500KV family library according to the recognition information, and performs tests and adjustments to ensure the accuracy of the model; finally, the output module obtains the model data from the family library and completes the three-dimensional BIM modeling through the intelligent agent.

[0098] In this application, LLM can extract best practices and solutions to common problems from a large amount of historical project data, which helps designers avoid repeated errors in the design phase of new projects and adopt proven effective methods. LLM can also automatically explore multiple possible solutions based on the input design parameters, and recommend the optimal layout solution, or predict the possible impact of certain design choices, such as energy consumption, construction costs, etc., thereby guiding designers to make more scientific and reasonable decisions, improving the automation level of BIM modeling of substations, reducing manual intervention, and speeding up modeling. By using powerful learning capabilities and pattern recognition functions, it overcomes the problems of low modeling efficiency, poor accuracy, frequent manual intervention, insufficient system coordination, and difficulty in achieving full life cycle management in existing technologies.

[0099] The above describes the specific embodiments of the present application. It should be understood that the present application is not limited to the above specific embodiments, and those skilled in the art can make various modifications or variations within the scope of the claims, which does not affect the substantive content of the present application. The above preferred features can be used in any combination without conflicting with each other.

Claims

1. A 500KV substation automatic BIM modeling method based on LLM agent, characterized in that: include: Based on the large language model (LLM) and agent development technology, an LLM agent is constructed to build the overall architecture of automatic BIM modeling of substations; A substation corpus database is obtained to train the LLM, and the LLM agent recognizes information of the two-dimensional drawing through the trained LLM; According to the information of the identified drawing, the LLM agent selects a family library model template, adjusts the family library model template according to the specific 500KV substation design requirements, and saves it as a 500KV family library; The model data in the 500KV family library is obtained, and automatic BIM modeling of the 500KV substation is performed according to the constructed LLM agent.

2. According to the LLM agent-based 500KV substation automatic BIM modeling method according to claim 1, it is characterized in that: The step of acquiring a substation corpus database and training the LLM, wherein the LLM agent recognizes information of a two-dimensional drawing through the trained LLM, includes: Obtain a substation corpus, select a large language model LLM, and perform fine-tuning training on the LLM; Obtaining the substation drawing data to be identified, converting the drawing data into a digital format, and generating a corresponding text description; The text description is input into the fine-tuned LLM for text encoding and semantic understanding, and key information is output in a structured form.

3. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 2 is characterized in that: The step of obtaining a substation corpus, selecting a large language model LLM, and fine-tuning the LLM includes: Acquire the substation corpus database, and preprocess the corpus database to generate data that meets the LLM training requirements; According to the preprocessed corpus database, the LLM is fine-tuned and trained to learn substation design methods and rules.

4. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 3 is characterized in that: The step of obtaining the substation drawing data to be identified, converting the drawing data into a digital format, and generating a corresponding text description includes: Acquire the substation drawing data to be identified, and convert the drawing data into a digital image format; Using image recognition technology, the digital image is converted into a pixel value array, and key feature information in the drawing is extracted from the pixel value array; Generate corresponding text description based on the extracted key feature information.

5. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 4 is characterized in that: The step of inputting the text description into the fine-tuned LLM for text encoding and semantic understanding, and outputting key information in a structured form, includes: According to the generated text description, input it into the fine-tuned LLM for text encoding; the text encoding is: converting the text description into a vector representation understandable by the model; Using the natural language processing capability of the LLM, the text description is semantically understood, deep semantic information in the text description is extracted, and more in-depth analysis and reasoning are performed in combination with context information; Key information is extracted from the text description of the substation drawing data, and output in a structured form based on the extracted key information.

6. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 1 is characterized in that: According to the information of the identified drawing, the LLM agent selects a family library model template, adjusts the family library model template according to the specific 500KV substation design requirements, and saves it as a 500KV family library, including: Acquire the information of the drawing, and according to the information of the drawing, the LLM agent selects the corresponding equipment and structure family library model template; By using the 3D modeling software Revit, parameterizing the family library model template according to the information of the drawing; Testing the established family library model to check whether relevant parameters of the family library model meet the requirements of the drawings; According to the test results, the family library model is adjusted and optimized; The adjusted and optimized family library model and related model data are saved.

7. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 1 is characterized in that: The step of acquiring the model data in the 500KV family library and performing automatic BIM modeling of the 500KV substation according to the constructed LLM agent includes: According to the model data in the 500KV family library, data related to the target three-dimensional model is collected, and the related data is preprocessed; the related data includes: CAD drawings and their text descriptions or attribute data; the preprocessing includes: data cleaning, data labeling and data format conversion; A large language model (LLM) is selected, training parameters are set to train the large language model (LLM), the trained model is used for reasoning, three-dimensional model coordinate points are generated, and the three-dimensional model is converted into a three-dimensional model.

8. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 7 is characterized in that: The selecting of a large language model (LLM), setting training parameters to train the large language model (LLM), using the trained model for reasoning, generating three-dimensional model coordinate points and converting them into a three-dimensional model, includes: Select a large language model (LLM) and set the training parameters according to the model architecture and the characteristics of the dataset; The large model is trained according to the preprocessed data to generate a trained model; wherein during the training process, the intelligent agent can provide feedback according to the learned knowledge and experience and adjust the training parameters; Get new target objects or scenes and prepare input data; The input data is input into the trained model for reasoning to generate coordinate points, and the LLM agent calls a three-dimensional BIM modeling tool to convert the generated coordinate points into a three-dimensional model.

9. The 500KV substation automatic BIM modeling method based on LLM agent according to claim 8 is characterized in that: After converting the generated coordinate points into a three-dimensional model, the method further includes: The intelligent agent evaluates the quality of the generated three-dimensional model and performs iterative optimization according to the evaluation results.

10. A 500KV substation automatic BIM modeling system based on LLM agent, characterized in that: include: The architecture building module is used to build LLM agents based on the large language model (LLM) and agent development technology, and to build the overall architecture of automatic BIM modeling of substations; A training module is used to obtain a substation corpus database and train the LLM, and the LLM agent recognizes the information of the two-dimensional drawing through the trained LLM; A test adjustment module, which is used for the LLM agent to select a family library model template according to the information of the identified drawing, and adjust the family library model template according to the specific 500KV substation design requirements, and save it as a 500KV family library; The output module is used to obtain the model data in the 500KV family library and perform automatic BIM modeling of the 500KV substation based on the constructed LLM agent.

Images