10kv transformer substation design scheme generation method based on large language model fine tuning

Abstract

The invention provides a 10KV transformer substation design scheme generation method based on large language model fine tuning. The method comprises the following steps: constructing a 10KV transformer substation design related fine tuning data set; obtaining a basic large language model, and performing term understanding evaluation on the basic large language model; based on the basic large language model, utilizing the data set, and adopting a LoRA fine tuning method to perform fine tuning and evaluation on the basic large language model; and introducing a design logic reasoning strategy based on an attention mechanism into the fine-tuned large language model to enable the model to understand design condition input, and outputting a semantic design scheme by applying a design rule to generate a 10kv transformer substation design scheme. According to the method, intelligent generation of the 10kv transformer substation design scheme is realized through the field self-adaptive fine-tuning large language model, the design efficiency and quality are remarkably improved, and meanwhile, the design scheme is ensured to meet technical specifications and safety requirements.

Description

Technical Field

[0001] This invention relates to the field of auxiliary design technology for 10kV substations, specifically to a method for generating 10kV substation design schemes based on fine-tuning of a large language model; it also relates to a corresponding system, computer terminal, and computer-readable storage medium. Background Technology

[0002] The current 10kV substation design field faces numerous pain points. Traditional design processes heavily rely on manual experience, resulting in long design cycles, low efficiency, and insufficient standardization. Designers need to consult numerous technical specifications, equipment parameters, and historical case documents—a time-consuming, labor-intensive, and error-prone process. Especially in the context of renewable energy grid integration and smart grid construction, 10kV substation design requires comprehensive consideration of more factors, such as distributed power source integration, reactive power compensation, and harmonic mitigation. Traditional design methods struggle to quickly respond to these complex requirements. Furthermore, the design process suffers from severe knowledge silos, low collaboration efficiency between different disciplines, and difficulty in ensuring the quality and consistency of design solutions. With the advancement of "dual-carbon" goals and the demand for intelligent upgrades to the power system, there is an urgent need for an intelligent design method that can integrate multi-source data, understand design specifications, and automatically generate high-quality design solutions.

[0003] Currently, intelligent design technologies for 10kV substations mainly include rule-based systems, parametric design tools, and limited applications of artificial intelligence. While rule-based systems can ensure design compliance with specifications, they lack flexibility and struggle to handle complex scenarios and emerging demands. Parametric design tools improve design efficiency but lack semantic understanding and logical reasoning capabilities, making it impossible to autonomously generate innovative solutions. In recent years, some research has attempted to apply machine learning to 10kV substation design, but these methods are typically limited to specific sub-tasks, such as equipment selection or layout optimization, lacking overall design capabilities. Existing AI technologies face challenges such as insufficient training data, inadequate representation of domain knowledge, and limited model generalization ability, making it difficult to meet the needs of intelligent design throughout the entire 10kV substation process. In particular, traditional natural language processing technologies perform poorly in the task of understanding and generating design text, failing to accurately capture the complex relationships between technical terms and design intent. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a method for generating 10kV substation design schemes based on fine-tuning of a large language model; it also provides a corresponding system, computer terminal, and computer-readable storage medium.

[0005] According to one aspect of the present invention, a method for generating a 10kV substation design scheme based on fine-tuning of a large language model is provided, comprising: Construct a fine-tuning dataset related to the design of a 10kV substation; Obtain a basic large language model and perform term understanding evaluation on the basic large language model to obtain the selected basic large language model; Based on the selected basic large language model, the LoRA fine-tuning method is used to fine-tune the basic large language model using the dataset to obtain a fine-tuned large language model that can provide specific parameters for 10kV substations. An attention-based design logic reasoning strategy is introduced into the fine-tuned large language model, enabling the model to understand the design condition input and apply design rules to output semantic design schemes, thereby generating a 10kV substation design scheme.

[0006] According to another aspect of the present invention, a system for generating 10kV substation design schemes based on fine-tuning of a large language model is provided, comprising: The dataset construction module is used to build a fine-tuning dataset related to the design of a 10kV substation. The model selection module is used to obtain a basic large language model and perform term understanding evaluation on the basic large language model to obtain the selected basic large language model. The model training module, based on the selected basic large language model, uses the dataset and employs the LoRA fine-tuning method to fine-tune the basic large language model, resulting in a fine-tuned large language model that can provide specific parameters for 10kV substations. The model reasoning module is used to introduce an attention-based design logic reasoning strategy into the fine-tuned large language model, enabling the fine-tuned large language model to understand the design condition input and apply design rules to output semantic design schemes, thereby generating a 10kV substation design scheme.

[0007] According to a third aspect of the present invention, a computer terminal is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can be used to perform the methods described above in the present invention, or to run the system described above in the present invention.

[0008] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, can be used to perform the methods described above in the present invention, or to run the system described above in the present invention.

[0009] By adopting the above technical solution, the present invention has at least one of the following beneficial effects compared with the prior art: The present invention provides a method and system for generating 10kV substation design schemes based on fine-tuning of a large language model. By fine-tuning a large language model using professional datasets, it breaks through the understanding bottleneck of traditional AI in the field of 10kV substation design. It can accurately analyze professional content such as equipment parameters, layout specifications and connection relationships, and solve the problem that existing intelligent design tools do not understand professional terms and design specifications in the power field. It realizes the leap from general semantic understanding to professional design decision-making and has the ability to deeply integrate domain knowledge.

[0010] The present invention provides a method and system for generating 10kV substation design schemes based on fine-tuning of a large language model. It innovatively combines a design logic reasoning framework with a large language model, which can not only handle structured design parameters, but also understand the engineering logic behind the design intent. It automatically generates complete schemes that meet technical specifications and have engineering rationality, overcoming the shortcomings of traditional parametric design tools that lack logical coherence and scheme innovation, and realizing the integration of intelligent reasoning and automated generation.

[0011] The present invention provides a method and system for generating 10kV substation design schemes based on large language model fine-tuning. It adopts a dynamic fine-tuning mechanism and a continuous learning architecture, which can continuously evolve with the updating of design specifications, the introduction of new equipment, and the accumulation of design experience. It solves the problems of rigidity and inflexibility of traditional intelligent design systems, which are difficult to adapt to the development of power grid technology. It ensures the forward-looking nature of the design method and its technological advancement throughout its life cycle, and has the characteristic of promoting adaptive optimization throughout the entire process.

[0012] The present invention provides a method and system for generating 10kV substation design schemes based on large language model fine-tuning, which significantly improves the efficiency and quality of 10kV substation design. Traditional 10kV substation design relies heavily on human experience, requiring designers to repeatedly consult numerous technical specifications and case studies, which is not only time-consuming and labor-intensive but also prone to design errors due to human negligence. This invention, through an intelligent design method based on a large language model, can automatically parse design requirements, understand technical specifications, and generate complete design schemes, shortening the design work that traditionally takes several days to within hours. Simultaneously, it ensures the design scheme's compliance with specifications and technical feasibility, significantly reducing the probability of human error.

[0013] This invention provides a method and system for generating 10kV substation design schemes based on large language model fine-tuning, realizing the automated acquisition and application of knowledge in the field of 10kV substation design. Traditional AI design tools are often limited to specific sub-tasks and lack the ability to understand the deep logic of design specifications. This invention, through domain-adaptive fine-tuning technology, enables the large language model to deeply master the professional knowledge system of 10kV substation design, including key elements such as equipment parameters, layout specifications, and connection relationships, and constructs a professional design logic reasoning framework. This allows the system not only to generate design schemes but also to understand the engineering logic behind the design intent, achieving a leapfrog development from simple parametric design to intelligent decision support.

[0014] This invention provides a method and system for generating 10kV substation design schemes based on large language model fine-tuning, offering a scalable technical solution for the digital transformation of the power industry. By employing a fine-tuning mechanism and a continuous learning architecture, the system can continuously evolve with updates to technical specifications, the introduction of new equipment, and the accumulation of design experience, ensuring long-term technological advancement. This invention is not only applicable to 10kV substation design but can also be extended to other voltage levels and new power system scenarios, providing a reusable technical framework for smart grid construction. Simultaneously, this invention reduces the reliance on specialized talent in design work, helping to address the industry's talent shortage problem and possessing significant economic and social value.

[0015] The present invention provides a method and system for generating 10kV substation design schemes based on large language model fine-tuning. It adopts intelligent design technology based on large language model fine-tuning and logical reasoning framework to realize the automated generation and optimization of 10kV substation design schemes, and solves the industry pain points of low efficiency, reliance on human experience and difficulty in ensuring compliance with design specifications in traditional design methods. Attached Figure Description

[0016] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the process of generating a 10kV substation design scheme based on fine-tuning of a large language model, according to a preferred embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the components of a 10kV substation design scheme generation system based on large language model fine-tuning in a preferred embodiment of the present invention.

[0018] Figure 3 This is a flowchart illustrating the design process of a method and system for generating 10kV substation design schemes based on fine-tuning of a large language model, as described in a specific application example of the present invention.

[0019] Figure 4This is a sample dataset from a specific application example of the present invention.

[0020] Figure 5 This is a schematic diagram of the qwen2.5 fine-tuning loss function in a specific application example of the present invention.

[0021] Figure 6 This is a sample response from a large model that has not been fine-tuned in a specific application example of the present invention.

[0022] Figure 7 This is an example of the response process of the large model after fine-tuning in a specific application example of the present invention.

[0023] Figure 8 This is a sample of the response results of the large model after fine-tuning in a specific application example of the present invention. Detailed Implementation

[0024] The embodiments of the present invention are described in detail below: These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

[0025] The current design field of 10kV substations faces numerous pain points. Traditional design processes heavily rely on human experience, resulting in long design cycles, low efficiency, and insufficient standardization. Existing AI technologies face challenges such as insufficient training data, inadequate representation of domain knowledge, and limited model generalization capabilities, making it difficult to meet the needs of intelligent design throughout the entire 10kV substation process. In particular, traditional natural language processing technologies perform poorly in tasks involving the understanding and generation of design text, failing to accurately capture the complex relationships between technical terms and design intent.

[0026] To address the aforementioned issues, one embodiment of this invention provides a method for generating 10kV substation design schemes based on fine-tuning of a large language model. This method aims to solve the industry pain points of traditional design methods, such as low efficiency, reliance on human experience, and difficulty in ensuring compliance with design specifications. It constructs a fine-tuning dataset for the 10kV substation design domain using data such as 10kV substation design technical specifications and historical design schemes; selects a suitable large language model from the market; fine-tunes the model using the LoRA method based on the large language model; and constructs a design logic reasoning framework to output the 10kV substation design scheme text. Through domain-adaptive fine-tuning of the large language model, intelligent generation of 10kV substation design schemes is achieved, significantly improving design efficiency and quality while ensuring that the design schemes comply with technical specifications and safety requirements.

[0027] Specifically, such as Figure 1As shown in the embodiment, the method for generating a 10kV substation design scheme based on fine-tuning of a large language model can include: S1, Construct a fine-tuning dataset related to the design of a 10kV substation; S2, obtain the basic large language model, and perform term understanding evaluation on the basic large language model to obtain the selected basic large language model; S3. Based on the selected basic large language model, the LoRA fine-tuning method is used to fine-tune the basic large language model using the dataset to obtain a fine-tuned large language model that can provide specific parameters for 10kV substations. S4 introduces an attention-based design logic reasoning strategy to the fine-tuned large language model, enabling the fine-tuned large language model to understand the design condition input and apply design rules to output semantic design schemes, generating a 10kV substation design scheme.

[0028] In some preferred embodiments, the above-mentioned S1, constructing a 10kV substation design-related fine-tuning dataset, may further include: S11, acquire relevant design data for the 10kV substation and extract it into a processable text format; S12, preprocess the extracted data and label it according to the Alpaca format to obtain the dataset; S13 divides the dataset into training, validation, and test sets according to a set ratio for fine-tuning the large language model.

[0029] In some preferred embodiments, S11, which involves acquiring design-related data for a 10kV substation and extracting it into a processable text format, may further include: The data related to the design of the 10kV substation includes: technical specifications, equipment manuals, and historical design schemes; document information is extracted from the technical specifications and equipment manuals using document text extraction methods; drawing information from historical design schemes is extracted using the Revit API and processed into JSON format.

[0030] In some preferred embodiments, S12 above, which involves preprocessing the extracted data and labeling it according to Alpaca format, may further include: The extracted data is cleaned and labeled according to the Alpaca format. The Alpaca format includes three fields: instruction, input, and output. The instruction field describes how the large language model completes the task and the output format, serving as the system prompt words for the large language model. The input field describes the design requirements and constraints. The output field represents the design scheme and design specifications.

[0031] In some preferred embodiments, the ratio of S13 is preferably set to 0.8, 0.1, or 0.1.

[0032] In some preferred embodiments, S2, which involves obtaining a basic large language model and performing term understanding evaluation on the basic large language model to obtain the selected basic large language model, may further include: S21. Based on model size and performance indicators, a large language model is selected to obtain a basic large language model; S22, perform terminology comprehension evaluation on the selected basic large language model to obtain the selected basic large language model.

[0033] In some preferred embodiments, the basic large language model selected in S21 above should have the following characteristics: High level of understanding: The model should be able to accurately understand and analyze the design specifications or user requirements, ensuring that the extracted key information is comprehensive and accurate, including but not limited to scope information, equipment information, and other design constraints. The model's understanding ability is evaluated using natural language processing benchmarks. For example, GLUE / SuperGLUE tests grammar, semantics, and reasoning abilities (such as textual entailment and semantic dereference). The MMLU (Massive Multitask Language Understanding) dataset covers cross-domain knowledge understanding across 57 disciplines. Models with higher scores are considered optimal.

[0034] Powerful generation capabilities: LLM should be able to generate preliminary design schemes, including wall information, equipment dimensions, and coordinate information, based on the input design requirements and constraints, and the generated schemes should be logical and practical.

[0035] Flexibility and Adaptability: The model should be able to adapt to 10kV substation design projects of different sizes and types, handle complex design problems, and demonstrate good adaptability during the design process. Flexibility and adaptability are evaluated through small-sample learning. Without fine-tuning, the model is tested to handle unseen task types (such as suddenly switching from "generating operation tickets" to "explaining relay protection principles"). Task switching accuracy and instruction follow-up completeness are manually evaluated. Models with higher scores are selected.

[0036] Economy and applicability: The model parameters should be less than 10 bytes to run on consumer-grade graphics cards, increasing the method's versatility. The model's context processing capability should be at least 8k bytes to understand multiple specifications and design schemes simultaneously.

[0037] In some preferred embodiments, the above-mentioned S22, which evaluates the terminology understanding of the selected basic large language model, may further include: S221. Establish a terminology database for 10kV substations (such as "differential protection", "GIS equipment", "five-prevention system", etc.), including: equipment terminology, operation terminology and standard (IEC / IEEE) terminology; S222, based on a specialized terminology library, establishes hierarchical test cases, including: basic terminology test cases (such as circuit breakers), composite terminology test cases (such as oil-immersed transformers), and terminology relationship test cases (such as the relationship between "CT ratio" and "protection setting"). S223: The test cases are used as inputs to the selected basic large language model to obtain the model's response; the similarity between the model's response and the terminology database is quantified using indicators such as BLEU-4 and ROUGE-L, and the basic large language model with the highest score is selected.

[0038] In some preferred embodiments, S3 above, based on the selected basic large language model, uses the dataset and employs the LoRA fine-tuning method to fine-tune the basic large language model to obtain a fine-tuned large language model capable of providing parameters specific to 10kV substations, may further include: S31, configure training parameters, learning rate and training rounds, and construct training information; S32 manages multi-GPU training through the Accelerate framework, enabling gradient accumulation to build a training environment when GPU memory is insufficient. S33, based on training information and training environment, utilizes the dataset and trains the selected basic large language model using the LoRA fine-tuning method, updates the LoRA adapter layer parameters while freezing the main model weights, and saves the LoRA adapter weights for subsequent loading after training is completed, resulting in a fine-tuned large language model that can provide specific parameters for 10kV substations.

[0039] In some preferred embodiments, S4 above, which introduces an attention-based design logic reasoning strategy to fine-tune the large language model, enabling the fine-tune large language model to understand the design condition input and apply design rules to output a semantic design scheme, may further include: Input design requirements and constraints, construct prompts as context, input a large model and a low-rank adapter model. The model continuously calculates weights to predict the next word until a design solution is output. The inference framework is as follows: Figure 3 shown. Specifically: S41, In the input phase, design requirements and design constraints are concatenated into structured prompts to obtain contextual input to the large language model; among them, design requirements include: site scope and equipment size information of 10kV substation; design constraints include: the specification content in the design code of 20 kV and below substations, and are organized into text segments; S42, during the input encoding stage, a multi-head attention mechanism is used to automatically identify key descriptions (such as "10kv", "GB 50053") and secondary descriptions (such as "site leveling") in the context, and assign different attention weights to key and secondary descriptions (such as key description weight reaching 0.8+ and secondary description weight decreasing to below 0.2); the effective context is obtained based on the attention weights; The 10kV substation-specific parameters provided by the fine-tuned large language model are superimposed with the main model weights of the fine-tuned large language model to obtain hybrid weights. Utilizing effective context, during the autoregressive generation process, a causal attention mask is combined to ensure that the current word depends only on the preceding text. Based on the hybrid weights, the design scheme is output through iteration.

[0040] Based on the same inventive concept, an embodiment of the present invention also provides a 10kV substation design scheme generation system based on large language model fine-tuning.

[0041] Specifically, such as Figure 2 As shown in the embodiment, the 10kV substation design scheme generation system based on large language model fine-tuning provided may include: The dataset construction module is used to build a fine-tuning dataset related to the design of a 10kV substation. The model selection module is used to obtain the basic large language model, evaluate the terminology understanding of the basic large language model, and obtain the selected basic large language model. The model training module, based on the selected basic large language model, uses the dataset and employs the LoRA fine-tuning method to fine-tune the basic large language model, resulting in a fine-tuned large language model that can provide specific parameters for 10kV substations. The model reasoning module is used to introduce an attention-based design logic reasoning strategy into the fine-tuned large language model, enabling the fine-tuned large language model to understand the design condition input and apply design rules to output semantic design schemes, generating a 10kV substation design scheme.

[0042] In some preferred embodiments, the dataset construction module may further include: A data acquisition unit is used to collect domain documents and extract them into a processable text format; its specific implementation further includes: Collect relevant design data for 10kV substations, including technical specifications, equipment manuals, and historical design schemes. Extract document information from specifications and equipment manuals using text extraction methods such as OCR and the MinerU document parser. Historical design schemes primarily utilize existing design models; Revit API is used for secondary development to extract equipment dimensions, coordinates, wall extents, and other information from the drawings, which is then formatted into JSON.

[0043] A data annotation unit, used for preprocessing and annotating data; its specific implementation further includes: The data in the aforementioned data acquisition unit is cleaned, removing irrelevant content, incoherent sentences, grammatically incorrect sentences, or obscure statements, and annotated according to the ALPA format. The ALPA format generally contains three fields: instruction, input, and output. The instruction field primarily describes how the large model completes the task and the output format; it can be considered a system prompt in the large language model. The input field primarily describes the design requirements and constraints. The output field represents the design scheme. Therefore, the user's design requirements and constraints should be labeled as input, and the design scheme and design specifications as output. The instruction uses a fixed string: "You are now a senior architect. Using your design experience and talents, please think step by step to arrive at a design scheme. The design scheme is represented in standard Markdown style."

[0044] The data partitioning unit is used to divide the dataset into training, validation, and test sets at ratios of 0.8, 0.1, and 0.1, respectively, for use in fine-tuning large models.

[0045] In some preferred embodiments, the model selection module may further include: The model selection unit is used to select mainstream large-scale models; its specific implementation further includes: The selection of a Large Language Model (LLM) is a crucial step, directly impacting the effectiveness and results of AI-assisted design methods. The chosen LLM should possess the following characteristics: 1) High level of comprehension: The model should be able to accurately understand and analyze the design task book or user requirements, and ensure that the extracted key information is comprehensive and accurate, including but not limited to scope information, equipment information and other design constraints.

[0046] 2) Powerful generation capabilities: LLM should be able to generate preliminary design schemes, including wall information, equipment dimensions, and coordinate information, based on the input design requirements and constraints, and the generated schemes should be logical and practical.

[0047] 3) Flexibility and adaptability: The model should be able to adapt to 10kV substation design projects of different sizes and types, handle complex design problems, and demonstrate good adaptability during the design process.

[0048] 4) Economy and Applicability: The model parameters should be less than 10 bytes to run on consumer-grade graphics cards, increasing the method's versatility. The model's context processing capability should be at least 8k bytes to understand multiple specifications and design schemes simultaneously.

[0049] The model evaluation unit is used to evaluate the terminology comprehension of the model; its specific implementation includes: The terminology database construction unit is used to build a dedicated terminology database for 10kV substations (such as "differential protection", "GIS equipment", "five-prevention system", etc.), including: equipment terms, operation terms, and standard (IEC / IEEE) terms; The test case construction unit establishes hierarchical test cases based on a dedicated terminology library, including: basic terminology test cases (such as circuit breakers), composite terminology test cases (such as oil-immersed transformers), and terminology relationship test cases (such as the relationship between "CT ratio" and "protection setting"). The model evaluation unit takes test cases as input to the selected basic large language model to obtain the model's response; it uses metrics such as BLEU-4 and ROUGE-L to quantify the similarity between the model's response and the terminology database, and selects the basic large language model with the highest score.

[0050] In some preferred embodiments, the model training module described above, which uses a fine-tuning script to fine-tune the large language model, may further include: The parameter configuration unit is used to configure training parameters such as a small batch size (e.g., 4 or 8), an appropriate learning rate (e.g., 1e-4 to 3e-4), and the number of training epochs (e.g., 3 epochs).

[0051] The GPU training unit manages multi-GPU training through the accelerate framework, and gradient accumulation can be enabled if there is insufficient GPU memory.

[0052] The fine-tuning training unit is used to start the training process, where only the LoRA adapter layer parameters are updated while the main model weights are frozen. After completion, the LoRA adapter weights (not the complete model) are saved for subsequent loading and use.

[0053] In some preferred embodiments, the above-mentioned model inference module may further: The reasoning framework building unit, which is used to build the reasoning framework, is further implemented as follows: First, the design requirements and constraints are input, and prompt words are concatenated as context. A large model and a low-rank adapter model are then input. The models continuously calculate using weights to predict the next word until a complete design solution is output. The inference framework is as follows: Figure 3 As shown.

[0054] The technical solutions provided by the above embodiments of the present invention will be further described in detail below with reference to specific application examples.

[0055] This specific application example, taking the generation of a 10kV substation design scheme as an example, involves the following steps in the generation method: Step 1: Construct a fine-tuning dataset for the design of 10kV substations.

[0056] Relevant documents in the field of 10kV substation design were collected, and a dataset was constructed using data preprocessing techniques. The specific process is as follows: We collected the "Design Code for 20 kV and Below Substations (GB 50053-2013)" and extracted the document content using MinnerU. We also collected a 10 kV substation RVT model and extracted equipment dimensions and coordinates using Revit API. After data preprocessing, we obtained a fine-tuning data sample.

[0057] This step extracts the specification document into standardized text, obtains key information from the RVT model, and annotates it with Alpaca format renderings, such as... Figure 4 As shown.

[0058] Step 2: Selection and evaluation of large language models.

[0059] The mainstream large models are evaluated, and the most suitable base model is selected based on parameters such as model size and performance indicators. The terminology understanding of the model is also evaluated.

[0060] Step 3: Domain-adaptive fine-tuning.

[0061] Combining the dataset from step one and the basic model from step two, the LoRA method is used to efficiently fine-tune the model, followed by evaluation. The specific process is as follows: We selected qwen2.5-7B-instruction as the base model and used a fine-tuning script to fine-tune the base model using the dataset. We set batch_size to 1, learning_rate to 1.0e-4, epochs to 3.0, and specified the dataset as the dataset constructed above.

[0062] Through this step, the model is fine-tuned according to a predetermined script until the loss converges, as shown below. Figure 5 As shown, it outputs the adapter.

[0063] Step 4: Design and construct the logical reasoning framework, and output the semantic design scheme.

[0064] A design logic reasoning module based on an attention mechanism is constructed, enabling the model to understand design condition inputs and apply design rules to output semantic design solutions. The specific process is as follows: The design requirements and constraints are input into the large language model, and a preliminary design scheme is given through the reasoning architecture.

[0065] This step allows for the correct design scheme for a 10kV substation. Answers without fine-tuning the large model are as follows: Figure 6 As shown, the effect of the large model after fine-tuning is as follows: Figure 7 and Figure 8 As shown. The method and system for generating 10kV substation design schemes based on large language model fine-tuning provided in the above embodiments of the present invention, by introducing large language model fine-tuning technology, leverages the powerful semantic understanding and generation capabilities of a pre-trained large language model, combined with domain-specific fine-tuning strategies, to understand 10kV substation design specifications, equipment parameters, and engineering experience, generating semantically-based design schemes that meet the requirements. The method provided in the above embodiments of the present invention, through domain-adaptive fine-tuning, significantly improves the model's understanding and expression of 10kV substation professional knowledge; the method constructs a design logic reasoning framework, enabling the model to make reasonable inferences based on specifications and experience; the method achieves end-to-end intelligent design process, automating the entire process from requirements analysis to scheme generation. Practical applications show that this method can improve design efficiency by more than 60%, while ensuring the standard compliance and technical feasibility of the design scheme, providing strong technical support for the intelligent construction of 10kV substations.

[0066] An embodiment of the present invention also provides a computer terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can be used to perform any of the methods described above in the present invention, or to run any of the systems described above in the present invention.

[0067] Optionally, the memory is used to store programs; the memory may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; the memory may also include non-volatile memory, such as flash memory. The memory is used to store computer programs (such as application programs and functional modules that implement the above methods), computer instructions, etc., and the aforementioned computer programs and computer instructions can be partitioned and stored in one or more memories. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by the processor.

[0068] A processor is used to execute computer programs stored in memory to implement the various steps of the methods or various modules of the systems involved in the above embodiments. For details, please refer to the relevant descriptions in the preceding method and system embodiments.

[0069] The processor and memory can be separate structures or integrated structures. When the processor and memory are separate structures, they can be coupled together via a bus.

[0070] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, can be used to perform the method of any of the above-described embodiments of the present invention, or to run the system of any of the above-described embodiments of the present invention.

[0071] Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of computer programs from one place to another. Storage media can be any available medium accessible to a general-purpose or special-purpose computer. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Alternatively, the ASIC can reside in a user device. Of course, the processor and storage medium can also exist as discrete components in a communication device.

[0072] The method and system for generating 10kV substation design schemes based on large language model fine-tuning provided in the above embodiments of the present invention significantly improve the efficiency and quality of 10kV substation design. Traditional 10kV substation design heavily relies on human experience, requiring designers to repeatedly consult numerous technical specifications and case studies, which is not only time-consuming and labor-intensive but also prone to design errors due to human negligence. The present invention, through an intelligent design method based on a large language model, can automatically parse design requirements, understand technical specifications, and generate complete design schemes, shortening the design work that traditionally takes several days to several hours. Simultaneously, it ensures the design scheme's compliance with specifications and technical feasibility, significantly reducing the probability of human error.

[0073] The method and system for generating 10kV substation design schemes based on large language model fine-tuning provided in the above embodiments of the present invention realize the automated acquisition and application of knowledge in the field of 10kV substation design. Traditional AI design tools are often limited to specific sub-tasks and lack the ability to understand the deep logic of design specifications. The present invention, through domain adaptive fine-tuning technology, enables the large language model to deeply master the professional knowledge system of 10kV substation design, including key elements such as equipment parameters, layout specifications, and connection relationships, and constructs a professional design logic reasoning framework. This allows the system not only to generate design schemes but also to understand the engineering logic behind the design intent, achieving a leapfrog development from simple parametric design to intelligent decision support.

[0074] The method and system for generating 10kV substation design schemes based on large language model fine-tuning provided in the above embodiments of the present invention offer a scalable technical solution for the digital transformation of the power industry. By employing a fine-tuning mechanism and a continuous learning architecture, the system can continuously evolve with updates to technical specifications, the introduction of new equipment, and the accumulation of design experience, ensuring long-term technological advancement. This invention is not only applicable to 10kV substation design but can also be extended to other voltage levels and new power system scenarios, providing a reusable technical framework for smart grid construction. Simultaneously, this invention reduces the reliance on specialized personnel in design work, helping to address the industry's talent shortage problem and possessing significant economic and social value.

[0075] Any matters not covered in the above embodiments of the present invention are well-known in the art.

[0076] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A method for generating 10kV substation design schemes based on fine-tuning of a large language model, characterized in that, include: Construct a fine-tuning dataset related to the design of a 10kV substation; Obtain a basic large language model and perform term understanding evaluation on the basic large language model to obtain the selected basic large language model; Based on the selected basic large language model, the LoRA fine-tuning method is used to fine-tune the basic large language model using the dataset to obtain a fine-tuned large language model that can provide specific parameters for 10kV substations. An attention-based design logic reasoning strategy is introduced into the fine-tuned large language model, enabling the model to understand the design condition input and apply design rules to output semantic design schemes, thereby generating a 10kV substation design scheme.

2. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 1, characterized in that, The construction of the 10kV substation design-related fine-tuning dataset includes: Obtain relevant design data for a 10kV substation and extract it into a processable text format; The extracted data is preprocessed and labeled according to the Alpaca format to obtain the dataset; The dataset is divided into training, validation, and test sets according to a set ratio for fine-tuning of the large language model.

3. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 2, characterized in that, The process of acquiring relevant design data for a 10kV substation and extracting it into a processable text format includes: The 10kV substation design-related data includes: technical specifications, equipment manuals, and historical design schemes; document information is extracted from the technical specifications and equipment manuals using document text extraction methods; drawing information is extracted from the historical design schemes using Revit API and processed into JSON format. The preprocessing of the extracted data, including annotation according to Alpaca format, includes: The extracted data is cleaned and labeled according to the Alpaca format. The Alpaca format includes three fields: instruction, input, and output. The instruction field describes how the large language model completes the task and the output format, serving as the system prompt words for the large language model. The input field describes the design requirements and constraints. The output field represents the design scheme and design specifications.

4. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 1, characterized in that, The process of obtaining a basic large language model and performing term understanding evaluation on the basic large language model to obtain a selected basic large language model includes: Based on model size and performance metrics, a basic large language model is selected. The selected basic large language model is evaluated for terminology comprehension to obtain the selected basic large language model.

5. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 4, characterized in that, The required model parameters for the basic large language model are less than 10 bytes. The large language model is required to have a context processing capability of 8k. The terminology comprehension evaluation of the selected basic large language model includes: Establish a dedicated terminology database for 10kV substations, including: equipment terminology, operational terminology, and standard terminology; Based on the aforementioned specialized terminology database, hierarchical test cases are established, including: basic terminology test cases, compound terminology test cases, and terminology relationship test cases; The test cases are used as input to the selected basic large language model to obtain the model's response; the similarity between the model's response and the terminology database is quantified using indicators, and the basic large language model with the highest score is selected.

6. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 1, characterized in that, Based on the selected basic large language model, the LoRA fine-tuning method is used with the dataset to fine-tune the basic large language model, resulting in a fine-tuned large language model capable of providing parameters specific to 10kV substations, including: Configure training parameters, learning rate, and training epochs to build training information; The Accelerate framework manages multi-GPU training and enables gradient accumulation when GPU memory is insufficient to build a training environment. Based on the training information and the training environment, the selected basic large language model is trained using the dataset through the LoRA fine-tuning method. The LoRA adapter layer parameters are updated while the main model weights are frozen. After training is completed, the LoRA adapter weights are saved for subsequent loading and use, resulting in a fine-tuned large language model that can provide parameters specific to 10kV substations.

7. The method for generating 10kV substation design schemes based on large language model fine-tuning according to claim 1, characterized in that, The introduction of an attention-based design logic reasoning strategy into the fine-tuned large language model enables the model to understand design condition inputs and apply design rules to output semantic design schemes, including: The design requirements and design constraints are concatenated into structured prompt words, and the resulting contextual input is fed into the large language model. The design requirements include: the site scope and equipment size information of the 10kV substation; the design constraints include: the specifications in the design code for 20kV and below substations, which are then organized into text segments. A multi-head attention mechanism is used to automatically identify key and secondary descriptions in the context, and different attention weights are assigned to the key and secondary descriptions; the effective context is obtained based on the attention weights. The 10kV substation-specific parameters provided by the fine-tuned large language model are superimposed with the main model weights of the fine-tuned large language model to obtain hybrid weights. Utilizing the effective context, during the autoregressive generation process, a causal attention mask is combined to ensure that the current word depends only on the preceding text. Based on the hybrid weights, the design scheme is output through iteration.

8. A system for generating design schemes for 10kV substations based on fine-tuning of a large language model, characterized in that, include: The dataset construction module is used to build a fine-tuning dataset related to the design of a 10kV substation. The model selection module is used to obtain a basic large language model and perform term understanding evaluation on the basic large language model to obtain the selected basic large language model. The model training module, based on the selected basic large language model, uses the dataset and employs the LoRA fine-tuning method to fine-tune the basic large language model, resulting in a fine-tuned large language model that can provide specific parameters for 10kV substations. The model reasoning module is used to introduce an attention-based design logic reasoning strategy into the fine-tuned large language model, enabling the fine-tuned large language model to understand the design condition input and apply design rules to output semantic design schemes, thereby generating a 10kV substation design scheme.

9. A computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it can be used to perform the method of any one of claims 1-7, or to run the system of claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program can be used to perform the method of any one of claims 1-7, or to run the system of claim 8.

Images