Transformer substation AI aided design method and system based on LLM and RAG

Abstract

The invention provides a transformer substation AI aided design method based on LLM and RAG, and the method comprises the steps: carrying out the demand analysis through a large language model LLM, and extracting key information; performing knowledge retrieval by using an RAG technology, wherein the knowledge retrieval comprises transformer substation aided design knowledge base construction, transformer substation aided design knowledge retrieval and knowledge reordering; according to the demand analysis and the knowledge retrieval result, the LLM is utilized to generate a preliminary design scheme of the transformer substation, and the design scheme comprises an electrical main wiring diagram, equipment type selection and layout. By adopting the combination of the LLM, RAG and AI aided design, the comprehensive improvement of the design efficiency, standardization and innovation of the transformer substation can be realized, and the problems of low efficiency, insufficient standardization and lack of innovation in the prior art are solved.

Description

Technical Field

[0001] The present application relates to the technical field of substation auxiliary design, and specifically, to a substation AI-assisted design method and system based on LLM and RAG. Background Art

[0002] At present, the auxiliary design method of substation mainly relies on traditional CAD software and some professional design tools. These tools usually include electrical design software, structural analysis software and 3D modeling software. Designers manually input parameters according to design specifications and experience to carry out electrical main wiring design, equipment selection, layout planning and structural design. These software can provide certain automation functions, such as automatic generation of drawings and equipment lists, but the overall design process is still highly dependent on the professional knowledge and experience of designers.

[0003] There are some limitations in existing auxiliary design methods. First, these methods often rely on the personal experience and skills of designers, resulting in uneven quality of design results. Second, since a lot of manual operations are required in the design process, the efficiency is relatively low, especially when facing complex or large-scale projects. In addition, these methods are insufficient in knowledge updating and design innovation, and it is difficult to integrate the latest design specifications and technological advances in real time, thus limiting the advancement and innovation of the design. Summary of the invention

[0004] In view of the defects in the prior art, the purpose of this application is to provide an AI-assisted design method for substations based on LLM and RAG. By combining large language models and retrieval enhancement generation technology, it not only improves the level of design automation, but also greatly improves the quality and efficiency of the design. It can update the knowledge base in real time and promote design innovation, while reducing human errors and improving the accuracy and reliability of the design. Through this interactive AI-assisted design, the substation design process becomes more efficient, intelligent and accurate.

[0005] In one aspect of the present application, a substation AI-assisted design method based on LLM and RAG is provided, including:

[0006] Use the large language model (LLM) to perform demand analysis and extract key information;

[0007] Use RAG technology for knowledge retrieval, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge re-ranking;

[0008] According to the demand analysis and the knowledge retrieval results, the LLM is used to generate a preliminary design scheme for the substation, the design scheme including an electrical main wiring diagram, equipment selection, and layout.

[0009] Furthermore, the use of the large language model LLM to perform demand analysis and extract key information includes:

[0010] Select the large language model LLM according to the design task book or user needs;

[0011] The selected LLM is used to analyze the design brief or user requirement document to extract key information.

[0012] Furthermore, the selected LLM is used to analyze the design task book or user requirement document to extract key information, including:

[0013] Extract all text content related to substation design based on the design task book or user requirement document, and eliminate non-text elements in the document;

[0014] Merge the related text contents scattered throughout the document;

[0015] Through structure recognition and keyword recognition, key information related to substations can be accurately located and extracted;

[0016] Performing content extraction on the extracted key information to convert the abstract demand description into specific design parameters;

[0017] The extracted and transformed information is input as context to the selected LLM, and the key information in the task book is given through the LLM.

[0018] Furthermore, the substation auxiliary design knowledge base is constructed, including:

[0019] Collect various documents and data related to substation design, including various design requirement task books, design specifications, standards, case libraries, and text descriptions or attribute data of these documents and data;

[0020] According to the various documents and data, select an embedding model that can maintain semantic information, has high-dimensional representation capabilities, and is scalable;

[0021] The various related documents and data are divided into blocks according to the dimensions of the selected embedding model, and the various related documents and data are converted into vector form using the selected embedding model.

[0022] Furthermore, after converting the various related documents and data into vector form, the method further includes:

[0023] Select a vector database, convert the various related documents and data into vector data and store them in the vector database to build a substation auxiliary design knowledge base;

[0024] The selected vector database is a vector database based on fast retrieval and similarity query capabilities.

[0025] Furthermore, the substation auxiliary design knowledge retrieval includes:

[0026] Obtain the embedding model and vectorize the knowledge retrieval keywords;

[0027] Obtain the vectorized knowledge retrieval keyword, perform similarity matching with the vectors in the vector database, and select the top K elements with the highest correlation from the matching results to form a top_K vector;

[0028] Convert the top_K vectors one by one and restore them to the original text information.

[0029] Furthermore, the knowledge reordering includes:

[0030] Select a re-ranking model, and according to the selected re-ranking model, sort the text information converted from the top_K retrieved vectors according to the relevance between the knowledge and the user's question, to generate a knowledge retrieval result;

[0031] The reordering model is selected based on the specific requirements of the design, and has the ability to be sorted, trained and optimized.

[0032] Further, the generating a preliminary design scheme of a substation by using the LLM according to the demand analysis and the knowledge retrieval result includes:

[0033] Constructing the RAG prompt words of the LLM;

[0034] Obtain the knowledge search results, splice the information in the knowledge search results according to the prompts of the RAG prompt words, and input them to the LLM so that it can perform standard interpretation and solution generation according to the Chain-of-Thought reasoning method;

[0035] Output the final result.

[0036] Furthermore, the large language model parameter is less than 10B, and the context processing capability is at least 8k.

[0037] The second aspect of the present application provides an AI-assisted substation design system based on LLM and RAG, including:

[0038] Demand analysis module, used to use the large language model LLM to perform demand analysis and extract key information;

[0039] Knowledge retrieval module, used for knowledge retrieval using RAG technology, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge re-ranking;

[0040] The design scheme generating module is used to generate a preliminary design scheme of the substation using the LLM according to the demand analysis and the knowledge retrieval results, including the main electrical connection diagram, equipment selection, and layout.

[0041] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0042] 1. This application uses a large language model (LLM) to understand the design task book or user requirements and extract key information; uses retrieval-augmented generation (RAG) technology to retrieve relevant design specifications, standards, case libraries and other knowledge to provide reference for the design; based on the demand analysis and the retrieved knowledge, LLM is used to generate a preliminary design plan for the substation, including the main electrical connection diagram, equipment selection, layout, etc., and the design plan is output. At the same time, relevant design instructions and construction instructions are generated to realize the auxiliary design of the substation.

[0043] 2. This application can significantly improve design efficiency and solve the problem of low design efficiency in the existing technology by adopting LLM technology of automated demand analysis and intelligent design suggestions. LLM can quickly understand and process design tasks and automatically generate preliminary design plans, thereby greatly reducing the workload of designers, especially when dealing with complex or large-scale projects, effectively shortening the design cycle and improving the response speed of the project.

[0044] 3. This application can achieve standardization and regularization of design by adopting RAG technology for precise retrieval and optimized design, and solve the problem of insufficient design standardization in the prior art. RAG technology can accurately retrieve and follow design specifications and standards, ensure the consistency and compliance of the design scheme, reduce design deviations caused by personal habits and preferences, and provide guarantee for the reliable operation of the substation.

[0045] 4. This application can promote knowledge fusion and innovation by adopting AI-assisted technology for knowledge fusion and interactive design, and solve the problem of insufficient knowledge updating and design innovation in existing technologies. AI technology can effectively integrate cross-disciplinary and cross-domain professional knowledge, provide cutting-edge references for design, stimulate the innovative thinking of designers, enable design plans to absorb the latest scientific research results in a timely manner, and enhance the advancement and competitiveness of the design. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] 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:

[0047] Figure 1 This is a flow chart of an AI-assisted design method for a substation based on LLM and RAG in one embodiment of the present application.

[0048] Figure 2 This is a detailed flow chart of a substation AI-assisted design method based on LLM and RAG in one embodiment of the present application.

[0049] Figure 3 This is a rendering of the "Standardized Operation Instructions" knowledge base in one embodiment of the present application.

[0050] Figure 4 This is a diagram showing the effect of knowledge retrieval in one embodiment of the present application.

[0051] Figure 5 This is a rendering of the preliminary design results in one embodiment of the present application. DETAILED DESCRIPTION

[0052] 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, several variations and improvements can also be made without departing from the concept of the present application. These all belong to the protection scope of the present application.

[0053] Reference Figure 1 As shown, a substation AI-assisted design method based on LLM and RAG in an embodiment of the present application includes:

[0054] S100, use the large language model LLM to perform demand analysis and extract key information;

[0055] In this step, the large language model (LLM) can be used to deeply analyze the design requirements of the substation and extract key design elements and constraints from complex text information. The extracted key information includes substation capacity, geographical location, voltage level, etc.

[0056] S200, using RAG technology for knowledge retrieval, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge re-ranking;

[0057] In this step, RAG (retrieval-augmented generation) technology is used for knowledge retrieval to build a knowledge base containing substation design expertise, cases and best practices. Then, based on the results of the demand analysis, relevant design knowledge and experience are retrieved from the knowledge base. Finally, the retrieved knowledge is reordered to filter out the information that best meets the current design needs.

[0058] S300, based on the demand analysis and knowledge retrieval results, uses LLM to generate a preliminary design plan for the substation, which includes the main electrical connection diagram, equipment selection, and layout.

[0059] In this step, after S200 obtains sufficient design basis and reference, it again utilizes the generation capability of LLM, combined with the results of demand analysis and knowledge retrieval, to automatically or semi-automatically generate a preliminary design plan for the substation, which covers in detail the design of the main electrical connection diagram, the selection of equipment, and the overall layout planning.

[0060] This application implements an AI-assisted design method for substations based on LLM and RAG through the above S100-S300, which solves the problem of insufficient knowledge updating and design innovation in the prior art. Through the deep understanding of LLM and the extensive knowledge retrieval of RAG, it can effectively integrate cross-disciplinary and cross-domain professional knowledge to provide a comprehensive and cutting-edge reference for substation design; it not only promotes the integration of knowledge in the design process, but also stimulates the innovative thinking of designers, so that the design plan can absorb the latest scientific research results and technological trends in a timely manner.

[0061] In addition, AI-assisted design optimization and interactive design processes provide designers with an open and flexible innovation platform, allowing substation design to be constantly innovated while ensuring standardization and regularization, thereby improving the advancement and competitiveness of the design.

[0062] In some specific embodiments, a large language model LLM is used to perform demand analysis and extract key information, including selecting a large language model LLM according to a design task book or user requirements; and using the selected LLM to analyze the design task book or user requirement document to extract key information.

[0063] Specifically, in the process of using the large language model (LLM) for demand analysis, the LLM model will first be selected based on the characteristics of the design task book or user requirements. Then, the selected LLM model will be used to conduct an in-depth analysis of the design task book or user requirement document, and through natural language processing technology, key information directly related to substation design will be extracted, such as voltage level, capacity requirements, equipment selection preferences, layout constraints, etc., to provide accurate data support for subsequent design work.

[0064] In the above embodiment, the selection of a large language model (LLM) is a key step, which directly affects the efficiency and results of the AI-assisted design method. The selected LLM should have the following characteristics:

[0065] 1) High comprehension capability: The model should be able to accurately understand and analyze the design brief or user requirements, ensuring that the extracted key information is comprehensive and accurate, including but not limited to substation capacity, geographical location, voltage level, etc.

[0066] 2) Powerful generation capability: LLM should be able to generate preliminary design solutions including electrical main wiring diagrams, equipment selection, layout, etc. based on demand analysis and retrieved knowledge, and the generated solutions should be logical and practical.

[0067] 3) Flexibility and adaptability: The model should be able to adapt to substation design projects of different sizes and types, be able to handle complex design problems, and show good adaptability in the design process.

[0068] 4) Economy and applicability: Model parameters should be less than 10B so that they can run on consumer-grade graphics cards and increase the universality of the method. The model context processing capability should reach at least 8k so that it can understand multiple specifications and design solutions at one time.

[0069] For example: BERT large language model based on Transformer structure, GPT and other large language models. Of course, this is just an example, and other large language models can also be selected.

[0070] In some specific embodiments, the selected LLM is used to analyze the design task book or user requirement document to extract key information, including: extracting all text content related to the substation design according to the design task book or user requirement document, and eliminating non-text elements in the document; merging relevant text content scattered throughout the document; accurately locating and extracting key information related to the substation through structure recognition and keyword recognition; performing content extraction on the extracted key information, and converting the abstract requirement description into specific design parameters; and inputting the extracted and converted information as context to the selected LLM, and providing the key information in the task book through the LLM.

[0071] Specifically, after the model selection is determined, the selected LLM is used to conduct an in-depth analysis of the design task book or user requirement document; by extracting the document content, eliminating non-text elements, and merging scattered information, to ensure that the model can understand the content of the task book in a continuous context; the document is carefully extracted, and key information such as substation capacity, geographical location, and voltage level are accurately located and extracted through structural recognition and keyword recognition, and abstract requirements are converted into specific design parameters. Finally, through output organization, the extracted information is presented in a clear and orderly manner, and is input as context to the large model, which gives key information in the task book, such as substation capacity, geographical location, voltage level, etc.

[0072] Among them, by analyzing the structure of the document, the key parts related to the substation design are identified. Using keyword recognition technology, key information such as substation capacity, geographical location, and voltage level can be accurately located and extracted, such as converting the geographical location description into specific longitude and latitude coordinates, and converting the voltage level description into a specific voltage value range.

[0073] For example: use LLM to extract information according to the text structure, such as first-level title and second-level title; combine the substation design demand analysis keywords, such as substation capacity, geographical location, voltage level, etc. to extract information, combine the two, and convert the extracted information into specific design parameters such as: {"substation capacity":"10MVA","voltage level":"10kv"...}.

[0074] Through the selected large language model LLM, the design task book or user requirement document is deeply analyzed, which can efficiently and accurately extract the key information closely related to the substation design, effectively eliminate the interference of non-text elements (including images and symbols), and integrate the scattered information into coherent content. At the same time, through structure recognition and keyword recognition technology, key information can be accurately located and extracted, and the abstract requirement description can be further converted into specific design parameters, which improves the design efficiency and accuracy.

[0075] In some specific embodiments, the construction of a substation auxiliary design knowledge base includes collecting various documents and data related to substation design; the various related documents and data include text descriptions or attribute data of various design requirement task books, design specifications, standards, and case libraries; selecting an embedding model based on the various related documents and data; dividing the various related documents and data into blocks according to the dimensions of the selected embedding model, and using the selected embedding model to convert the various related documents and data into vector form.

[0076] In the process of selecting the embedding model, it is necessary to ensure that the selected model can effectively convert the knowledge such as design specifications, standards, case libraries, etc. into vector form to facilitate subsequent reordering and retrieval. The selected embedding model should have the following characteristics:

[0077] 1) Semantic preservation: The model should be able to preserve the semantic information of the text and ensure that the information loss during the embedding process is minimized.

[0078] 2) High-dimensional representation capability: The model should be able to map text into a high-dimensional space so that similar texts are close to each other in the vector space.

[0079] 3) Computational efficiency: The model should have high computational efficiency and be able to quickly process large amounts of text data to meet real-time design requirements.

[0080] 4) Scalability: The model should be able to expand with the expansion of the knowledge base and maintain the consistency and stability of the embedding effect.

[0081] The embedding model in this application can retain semantic information, has high-dimensional representation capabilities, high computational efficiency, and strong scalability, such as the BERT (Bidirectional Encoder Representations from Transformers) model. Of course, this is just an example, and other embedding models can also be selected.

[0082] Among them, various documents and data related to substation design are not limited to the above documents and data, but also include equipment parameters and performance data, operation and maintenance records and fault data, etc.

[0083] Furthermore, after converting various related documents and data into vector form, the method further includes: selecting a vector database, converting various related documents and data into vector form data and storing them in the vector database, and constructing a substation auxiliary design knowledge base.

[0084] Among them, the vector database is selected as a vector database based on fast retrieval and similarity query capabilities.

[0085] Among them, in the AI-assisted design method, the selection of vector database is also crucial, which is responsible for storing and managing the vector representation of design specifications, standards, case libraries and other knowledge for fast retrieval and similarity query. The selected vector database should have the following characteristics:

[0086] 1) Efficient storage capacity: The database should be able to efficiently store large-scale vector data while maintaining data integrity and consistency.

[0087] 2) Fast retrieval performance: The database should support fast vector search and be able to quickly find the most similar knowledge vectors in large-scale data sets to meet the needs of real-time design.

[0088] 3) High scalability: The database should have good scalability and be able to expand storage and computing capabilities as the amount of data increases without affecting retrieval performance.

[0089] 4) Accurate similarity calculation: The database should provide an accurate similarity calculation method to ensure that the retrieved knowledge fragments are highly relevant to the design requirements.

[0090] 5) Easy to integrate and maintain: The database should be easy to integrate with existing AI models and design tools, have low maintenance costs, and facilitate long-term operation and updating.

[0091] 6) Support diversified queries: The database should support multiple query methods, including but not limited to nearest neighbor search, range query, combined query, etc., to adapt to different design scenarios.

[0092] 7) Data security and privacy protection: The database should have a complete data security mechanism to protect the security of design knowledge and user data and prevent unauthorized access and data leakage.

[0093] For example: chromadb vector database. Of course, this is just an example, and other databases can also be selected.

[0094] Specifically, when constructing the knowledge base of substation auxiliary design, we first collect various documents and data closely related to substation design, including but not limited to design requirement task books, design specifications, standards, rich case libraries, and other valuable text descriptions or attribute data. Then, we select an embedding model, divide the collected documents and data into reasonable blocks according to the dimensions of the selected embedding model, and use the embedding model to convert these documents and data into vector form. Finally, we store the vectors in the selected vector database.

[0095] In the above embodiment, a specific implementation method is to use the substation design work instruction to build a knowledge base, collect the "Substation Standardized Work Instructions", extract the document content through PyMuPDF4LLM, divide the document content into blocks and input the embedding model to generate vectors, store them in the vector database, form a knowledge base of the substation standardized work instruction, store the documents (156 documents are used for the test) in the vector database, and form a knowledge base. The effect diagram is as follows: Figure 3 shown.

[0096] In some specific embodiments, substation auxiliary design knowledge retrieval includes: obtaining an embedding model and vectorizing knowledge retrieval keywords; obtaining the vectorized knowledge retrieval keywords, performing similarity matching with vectors in a vector database, and selecting the top K elements with the highest correlation from the matching results to form a top_K vector; and converting the top_K vectors in turn to restore them to the original text information.

[0097] Specifically, the knowledge search keywords entered by the user are first converted into vector form, and then similarity matching is performed in the vector database containing a large amount of substation design-related knowledge, and the top K vectors (i.e., top_K vectors) with the highest correlation with the search keywords are quickly screened out. Finally, the top_K vectors are converted in turn and restored to the original text information, which is convenient for users to directly read and understand the use steps. By converting both keywords and database content into vector form for matching, the search speed is accelerated, and the semantic connection between keywords and knowledge content is captured more accurately, ensuring that users can quickly obtain the most relevant and valuable design knowledge.

[0098] Among them, the form stored in the vector library is a numerical vector (digital form), and the vector needs to be restored to text, namely, substation auxiliary design knowledge (text form). The purpose of restoring the numerical form to text is to construct the context of auxiliary design so that the large model can refer to this knowledge when performing scheme design.

[0099] In the above embodiment, one implementation method is to use RAG technology to perform knowledge retrieval.

[0100] Process: The test uses PyMuPDF4LLM to extract the content of the document "GB-50229-2019 Fire Protection Standard for Design of Thermal Power Plants and Substations", divides the content into blocks of 500 characters, inputs the embedding model to generate vectors, and then stores them in the chromadb vector database. Enter the knowledge query "What requirements should the main plant building of a coal-fired power plant meet for safe evacuation" to search, guide the large model to answer, correctly retrieve relevant content, and the large language model can correctly answer the queried knowledge. The effect diagram is as follows: Figure 4 shown.

[0101] Through the AI-assisted design method of substations based on LLM and RAG, the problem of insufficient design standardization and regularization in existing technologies is overcome. Through LLM's in-depth understanding of the design task book and RAG's accurate retrieval of design specifications and standards, it is ensured that every step in the design process strictly follows industry standards and best practices, reducing inconsistencies caused by personal design habits and preferences, and ensuring the consistency and repeatability of substation design solutions. In addition, the AI-assisted design review step further ensures the accuracy and compliance of the design solution, thereby improving the overall design standardization level and providing a solid foundation for the reliable operation and maintenance of substations.

[0102] In some specific embodiments, knowledge re-ranking includes: selecting a re-ranking model, and according to the selected re-ranking model, sorting the text information converted from the retrieved top_K vectors according to the relevance of the knowledge to the user's question to generate knowledge retrieval results; the re-ranking model, based on the specific needs of the design, selects a re-ranking model with sorting capabilities, which can be trained and optimized.

[0103] Specifically, using the re-ranking model, the top_K retrieved text information is sorted according to the relevance of the knowledge to the user's question, with the most relevant ones at the front and the least relevant ones at the end.

[0104] Among them, the role of the re-ranking model is to sort the retrieved knowledge fragments based on the preliminary retrieval results and according to the specific needs of the design, so as to improve the accuracy and practicality of the relevant knowledge.

[0105] The selected reordering model should have the following characteristics:

[0106] 1) Accurate ranking capability: The model should be able to accurately evaluate the relevance of retrieval results and effectively sort the results to ensure that the most relevant knowledge is ranked first.

[0107] 2) Adaptability: The model should be able to adapt to different design scenarios and requirements and provide customized sorting results.

[0108] 3) Real-time: The model should have the ability to respond quickly and be able to complete the sorting task in a short time to meet the real-time requirements of the design process.

[0109] 4) Easy to train and optimize: The model should be easy to train, be able to quickly achieve good sorting results with a small amount of labeled data, and be easy to optimize and adjust later.

[0110] For example, the Deep Listwise Context Model (DLCM) model uses the GRU model to learn the information of the Top N documents, forms a local ranking context, and then uses it for re-ranking.

[0111] Through the deep understanding of LLM and the extensive knowledge retrieval of RAG, it is possible to effectively integrate interdisciplinary and cross-domain professional knowledge to provide a comprehensive and cutting-edge reference for substation design. It not only promotes the integration of knowledge in the design process, but also stimulates the innovative thinking of designers, so that the design scheme can absorb the latest scientific research results and technological trends in a timely manner. In addition, AI-assisted design optimization and interactive design links provide the design team with an open and flexible innovation platform, so that substation design can not only ensure standardization and normalization, but also continuously innovate and improve the advancement and competitiveness of the design.

[0112] In some specific embodiments, according to the demand analysis and the knowledge retrieval results, LLM is used to generate a preliminary design plan for the substation, including: constructing the RAG prompt words of LLM; obtaining the knowledge retrieval results, splicing the information in the knowledge retrieval results according to the prompts of the RAG prompt words, and inputting it to LLM so that it can interpret the specifications and generate plans according to the Chain-of-Thought reasoning method; and outputting the final result.

[0113] Specifically, construct a large language model RAG prompt word "You are a senior architectural designer. Use your design experience and your design talent to think step by step to get a design plan, and combine the context {context} to answer the user's question. The design plan is expressed in the standard markdown json format and placed at the end of the paragraph." Combined with the prompt words, interact with the large language model, splice the re-ordered context after retrieval according to the prompt, and input it into the large language model, so that it can perform standardized interpretation and plan generation according to the Chain-of-Thought reasoning method, and output the final result.

[0114] Among them, the prompt refers to the prompt word of the large language model, for example: "You are an experienced architectural designer. Use your design experience and your design talent to think step by step to get a design plan, and combine the context {context} to answer the user's question {question}. The design plan is expressed in the standard markdown json format and placed at the end of the paragraph." When using it, you need to splice the retrieved knowledge as the value of the variable context, and the user's question as the value of the variable question.

[0115] By automating steps such as demand analysis, knowledge retrieval, preliminary design, solution optimization, design review, interactive design, and output, the manual operations and time investment of designers are greatly reduced. Especially when dealing with complex or large-scale projects, the intervention of AI can quickly process large amounts of data and provide intelligent design suggestions, thereby accelerating the design process and shortening the project cycle. In addition, by updating the knowledge base in real time and integrating the latest research results, this application ensures that the design process is always at the forefront of technology, further improving design efficiency and achieving rapid response and iteration of substation design.

[0116] In the above embodiments, one implementation method is to use a large language model for preliminary result design. The process is to input the design requirements into the large language model, and the model consults the relevant specifications and constraints to give a preliminary design plan. The model can correctly give a preliminary design plan, and the effect is as follows: Figure 5 shown.

[0117] The second aspect of the present application provides an AI-assisted design system for substations based on LLM and RAG, including: a demand analysis module for using a large language model (LLM) to perform demand analysis and extract key information; a knowledge retrieval module for using RAG technology to perform knowledge retrieval, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge reordering; a design scheme generation module for using the LLM to generate a preliminary design scheme for the substation based on the demand analysis and the knowledge retrieval results, including an electrical main connection diagram, equipment selection, and layout.

[0118] This application first uses the large language model LLM through the demand analysis module to deeply analyze the design task book or user needs and accurately extract key information; then, the knowledge retrieval module uses RAG technology to perform efficient retrieval in the constructed substation auxiliary design knowledge base, and intelligently reorders the retrieval results to screen out the most relevant information; finally, the design solution generation module combines the results of demand analysis and knowledge retrieval, and again uses the powerful generation capability of LLM to automatically generate a preliminary design solution including the main electrical connection diagram, equipment selection and layout. Through precise demand analysis and efficient knowledge retrieval, the accuracy and innovation of the design solution are ensured, providing solid technical support for the safe, reliable and economical operation of the substation.

[0119] The modules in the above-mentioned LLM and RAG-based substation AI assisted design system embodiment of the present application correspond to the steps in the above-mentioned LLM and RAG-based substation AI assisted design method embodiment. The module implementation technology can refer to the implementation technology of the steps in the method embodiment, which will not be repeated here.

[0120] By adopting the combination of the above-mentioned LLM, RAG and AI-assisted design, this application can achieve a comprehensive improvement in substation design efficiency, standardization and innovation, overcoming the problems of low efficiency, insufficient standardization and lack of innovation in the existing technology.

[0121] 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 substation AI-assisted design method based on LLM and RAG, characterized in that: include: Use the large language model (LLM) to perform demand analysis and extract key information; Use RAG technology for knowledge retrieval, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge re-ranking; According to the demand analysis and the knowledge retrieval results, the LLM is used to generate a preliminary design scheme for the substation, the design scheme including an electrical main wiring diagram, equipment selection, and layout.

2. According to claim 1, a substation AI-assisted design method based on LLM and RAG is characterized in that: The large language model LLM is used to perform demand analysis and extract key information, including: Select the large language model LLM according to the design task book or user needs; The selected LLM is used to analyze the design task book or user requirement document to extract key information.

3. According to claim 2, a substation AI-assisted design method based on LLM and RAG is characterized in that: The selected LLM is used to analyze the design task book or user requirement document to extract key information, including: Extract all text content related to substation design based on the design task book or user requirement document, and eliminate non-text elements in the document; Merge the related text contents scattered throughout the document; Through structure recognition and keyword recognition, key information related to substations can be accurately located and extracted; Performing content extraction on the extracted key information to convert the abstract demand description into specific design parameters; The extracted and transformed information is input as context to the selected LLM, and the key information in the task book is given through the LLM.

4. According to claim 1, a substation AI-assisted design method based on LLM and RAG is characterized in that: The substation auxiliary design knowledge base is constructed, including: Collect various documents and data related to substation design, including text descriptions or attribute data of various design requirement task books, design specifications, standards, and case libraries; Select an embedding model based on the various documents and data involved; The various related documents and data are divided into blocks according to the dimensions of the selected embedding model, and the various related documents and data are converted into vector form using the selected embedding model.

5. According to claim 4, a substation AI-assisted design method based on LLM and RAG is characterized in that: After converting the various related documents and data into vector form, the following steps are also included: Select a vector database, convert the various related documents and data into vector data and store them in the vector database to build a substation auxiliary design knowledge base; The selected vector database is a vector database based on fast retrieval and similarity query capabilities.

6. According to claim 5, a substation AI-assisted design method based on LLM and RAG is characterized in that: The substation auxiliary design knowledge retrieval includes: Obtain the embedding model and vectorize the knowledge retrieval keywords; Obtain the vectorized knowledge retrieval keyword, perform similarity matching with the vectors in the vector database, and select the top K elements with the highest correlation from the matching results to form a top_K vector; Convert the top_K vectors one by one and restore them to the original text information.

7. The AI-assisted design method for substation based on LLM and RAG according to claim 6 is characterized in that: The knowledge reordering includes: Select a re-ranking model, and according to the selected re-ranking model, sort the text information converted from the top_K retrieved vectors according to the relevance between the knowledge and the user's question, to generate a knowledge retrieval result; The reordering model is selected based on the specific requirements of the design, and has the ability to be sorted, trained and optimized.

8. The AI-assisted design method for substation based on LLM and RAG according to claim 1 is characterized in that: The step of generating a preliminary design scheme of a substation by using the LLM according to the demand analysis and the knowledge retrieval result includes: Constructing the RAG prompt words of the LLM; Obtain the knowledge search results, splice the information in the knowledge search results according to the prompts of the RAG prompt words, and input them to the LLM so that it can perform standard interpretation and solution generation according to the Chain-of-Thought reasoning method; Output the final result.

9. The AI-assisted design method for substation based on LLM and RAG according to claim 2 is characterized in that: The large language model has parameters less than 10B and a context processing capability of at least 8k.

10. A substation AI-assisted design system based on LLM and RAG, characterized in that: include: Demand analysis module, used to use the large language model LLM to perform demand analysis and extract key information; Knowledge retrieval module, used for knowledge retrieval using RAG technology, including substation auxiliary design knowledge base construction, substation auxiliary design knowledge retrieval and knowledge re-ranking; The design scheme generating module is used to generate a preliminary design scheme of the substation using the LLM according to the demand analysis and the knowledge retrieval results, including the main electrical connection diagram, equipment selection, and layout.

Images