A large language model training and fine-tuning system and method for power information communication

Abstract

The present invention relates to the field of model evaluation technology. More specifically, the present invention provides a system and method for training and fine-tuning a large language model for electric power information communication, integrating multiple modules such as data collection, preprocessing, model architecture selection, training, hyperparameter tuning, evaluation and optimization, and realizing efficient processing and analysis of text data in the field of electric power information communication. The system can accurately extract field-specific information. At the same time, the system also establishes a scientific evaluation system to comprehensively evaluate the trained model and perform targeted fine-tuning or retraining based on the evaluation results, thereby significantly improving the accuracy and generalization ability of the model when processing text in the field of electric power information communication, providing strong support for the intelligent operation, maintenance and management of the power system, and promoting the intelligent development process in the field of electric power information communication.

Description

Technical Field

[0001] The present invention relates to the technical field of model evaluation, and more specifically, to a system and method for training and fine-tuning a large language model for electric power information communication. Background Art

[0002] With the continuous development of power systems and the improvement of their intelligence level, the amount of data in the field of power information and communication has exploded. These data cover power equipment status, fault information, market transactions, network structure, dispatching instructions and other aspects, which are of great significance to the operation, maintenance and management of power systems. However, traditional data processing and analysis methods often seem to be unable to cope with such massive and complex data, and it is difficult to meet the requirements of real-time, accuracy and intelligence.

[0003] In recent years, large language models (LLM) have made significant progress in the field of natural language processing. Their powerful text generation and understanding capabilities provide new ideas for processing complex text data. However, directly applying large language models in general fields to the field of power information and communication often leads to poor model performance due to lack of domain knowledge and context mismatch, making it difficult to accurately understand and process specific information in the field of power information and communication.

[0004] For example, the current large language model in the field of power information and communication cannot accurately extract specific information in the field of power information and communication from massive text data, such as power equipment names, fault types, market transaction information, etc.; it is difficult to understand and process complex text data in the field; during the model training process, it is impossible to reasonably adjust hyperparameters such as learning rate, batch size, and number of iterations to optimize model performance; and it is impossible to establish an effective evaluation system to comprehensively evaluate the trained model and fine-tune or retrain the model based on the evaluation results.

[0005] To solve these problems, the present invention proposes a large language model training and fine-tuning system for power information communication and a method thereof. Summary of the invention

[0006] The purpose of the present invention is to provide a large language model training and fine-tuning system for power information communication and a method thereof to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a large language model training and fine-tuning system for power information communication, the system comprising: Data collection module, used to collect text data in the field of power information communication; A data preprocessing module is used to preprocess the text data, and the module further includes: The word segmentation submodule is used to divide text data into words; Denoising submodule, used to remove noise information from text data; Special character and symbol processing submodule, used to process special characters and symbols in text data; The case conversion submodule is used to convert the letters in the text data into uppercase or lowercase uniformly; an abbreviation processing submodule, for processing abbreviations, including establishing an abbreviation expansion table in the field of electric power information communication, and expanding abbreviations into complete words or phrases; The generalization processing submodule is used to improve the model's generalization ability for texts in the field of power information and communication, including a domain-specific vocabulary enhancer, a context understanding optimization submodule, and an unseen word processing submodule; Model architecture selection module, used to select a large language model architecture suitable for the power information communication field; Model training module, used to train the selected model architecture based on the preprocessed data; Hyperparameter tuning module, used to adjust hyperparameters during model training to optimize model performance; The model evaluation and optimization module is used to evaluate the performance of the trained model and optimize the model based on the evaluation results.

[0008] A further technical solution of the present application is as follows: the data preprocessing module also includes a domain-specific information extraction step for extracting specific information in the field of power information communication to provide rich features for subsequent model training.

[0009] A further technical solution of the present application is as follows: the model architecture selection module selects the Transformer model architecture or its variant as the basic architecture of the large language model according to the characteristics and needs of the power information communication field.

[0010] A further technical solution of the present application is: the hyperparameter tuning module adjusts the hyperparameters of learning rate, batch size, and number of iterations through grid search, random search, or Bayesian optimization to optimize model performance.

[0011] A further technical solution of the present application is: the model evaluation and optimization module uses evaluation indicators such as accuracy, recall rate, F1 score, etc. to evaluate the model, and fine-tunes or retrains the model based on the evaluation results to improve the accuracy and generalization ability of the model.

[0012] A further technical solution of the present application is: the domain-specific vocabulary enhancement submodule in the generalization processing submodule uses a domain dictionary or knowledge graph to perform vocabulary enhancement processing on text data, including adding synonyms and hypernyms to enhance the model's understanding of vocabulary specific to the power information communication field; The context understanding optimization submodule in the generalization processing submodule analyzes the text structure and sentence characteristics in the field of power information communication, performs semantic analysis and context reasoning on the text, and optimizes the model's ability to understand complex sentences and paragraphs; The unseen word processing submodule in the generalization processing submodule processes or replaces unseen words appearing in the text by using spelling correction, context-based guessing or external resource-based methods.

[0013] A further technical solution of the present application is: the specific information includes the name of the power equipment, the type of power fault, the power market transaction information, the power network structure information, the power dispatching instructions, the power load data, the power safety information, the power communication network information and the power equipment operating status information.

[0014] The present application also provides a method for training and fine-tuning a large language model for electric power information communication, the method comprising the following steps: S1. Collect text data in the field of power information and communication using a data collection module; S2, preprocessing the collected text data through the data preprocessing module, including word segmentation, denoising, processing special characters and symbols, case conversion, processing abbreviations, and generalization; S3. According to the characteristics and needs of the power information communication field, a suitable large language model architecture is selected through the model architecture selection module; S4. Based on the preprocessed data, train the selected model architecture through the model training module; S5. During the model training process, the hyperparameters of learning rate, batch size, and number of iterations are adjusted through the hyperparameter tuning module to optimize the model performance; S6. Use the model evaluation and optimization module to evaluate the trained model, and fine-tune or retrain the model based on the evaluation results until the model reaches the expected accuracy and generalization ability.

[0015] A further technical solution of the present application is as follows: in the step S2, it further includes a domain-specific information extraction sub-step, which utilizes the domain-specific information extraction step in the data preprocessing module to extract specific information in the field of power information communication, thereby providing rich features for subsequent model training.

[0016] A further technical solution of the present application is as follows: in step S5, the model is comprehensively evaluated using evaluation indicators such as accuracy, recall, and F1 score, and based on the evaluation results, the model is fine-tuned or retrained in a targeted manner in combination with domain-specific vocabulary enhancement, context understanding optimization, and unseen word processing methods to further improve the accuracy and generalization ability of the model in the field of power information and communication.

[0017] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: The present invention integrates multiple modules including data collection, preprocessing, model architecture selection, training, hyperparameter tuning, evaluation and optimization, and realizes efficient processing and analysis of text data in the field of power information and communication. The system can accurately extract field-specific information, such as the name of power equipment and fault type, optimize the selection and adapt the Transformer advanced model architecture, improve the model's ability to understand complex texts through the generalization processing submodule, and use the hyperparameter tuning module to fine-tune key parameters such as learning rate and batch size to optimize model performance. At the same time, the system has also established a scientific evaluation system to comprehensively evaluate the trained model and perform targeted fine-tuning or retraining based on the evaluation results, thereby significantly improving the accuracy and generalization ability of the model in processing texts in the field of power information and communication, providing strong support for the intelligent operation, maintenance and management of the power system, and promoting the intelligent development process in the field of power information and communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a system block diagram of the present invention; Figure 2 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. The present invention is further described below in combination with the embodiments.

[0020] See also Figure 1 and Figure 2 In an embodiment of the present application, a large language model training and fine-tuning system for electric power information communication is provided, the system comprising: Data collection module, used to collect text data in the field of power information communication; A data preprocessing module is used to preprocess the text data, and the module further includes: The word segmentation submodule is used to divide text data into words; Denoising submodule, used to remove noise information from text data; Special character and symbol processing submodule, used to process special characters and symbols in text data; The case conversion submodule is used to convert the letters in the text data into uppercase or lowercase uniformly; an abbreviation processing submodule, for processing abbreviations, including establishing an abbreviation expansion table in the field of electric power information communication, and expanding abbreviations into complete words or phrases; The generalization processing submodule is used to improve the model's generalization ability for texts in the field of power information and communication, including a domain-specific vocabulary enhancer, a context understanding optimization submodule, and an unseen word processing submodule; Model architecture selection module, used to select a large language model architecture suitable for the power information communication field; Model training module, used to train the selected model architecture based on the preprocessed data; Hyperparameter tuning module, used to adjust hyperparameters during model training to optimize model performance; The model evaluation and optimization module is used to evaluate the performance of the trained model and optimize the model based on the evaluation results.

[0021] Specifically, in this embodiment, the data collection module is responsible for collecting text data from various power information communication related resources. These data can come from internal documents of power companies, public reports of the power industry, power-related academic papers, power news websites, etc. In order to ensure the diversity and comprehensiveness of data, the data collection module needs to regularly update the data source and automatically crawl and store new text data; the data preprocessing module performs a series of processing on the collected text data to ensure the quality and consistency of the data.

[0022] The model architecture selection module is responsible for selecting a large language model architecture suitable for the power information and communication field. When selecting a model architecture, factors such as model performance, training speed, and resource consumption need to be considered. The following are some possible model architecture choices: Transformer-based model architectures, such as BERT, GPT, etc., which perform well in natural language processing tasks and have good generalization capabilities; model architectures customized for the power information and communication field, such as models that are improved by combining specific knowledge and rules in the power field.

[0023] The model training module uses the preprocessed data to train the selected model architecture. During the training process, appropriate training parameters and training strategies need to be set to ensure that the model can learn the characteristics and patterns of text data.

[0024] Here are some training strategies: Use large-scale datasets for training to improve the generalization ability of the model; Use distributed training or parallel training techniques to speed up training; use pre-trained models for transfer learning to save training time and resources.

[0025] The hyperparameter tuning module is responsible for adjusting the hyperparameters during model training to optimize model performance. Hyperparameters include learning rate, batch size, number of training rounds, etc. Here are some hyperparameter tuning methods: Use grid search or random search to search in the hyperparameter space; dynamically adjust the value of hyperparameters based on the training effect and verification results of the model; use automated machine learning (AutoML) tools to perform hyperparameter tuning to improve tuning efficiency and accuracy.

[0026] The model evaluation and optimization module is used to evaluate the performance of the trained model and optimize the model based on the evaluation results. The following are the evaluation and optimization methods: Evaluation method: The model is evaluated using the test dataset to measure the accuracy and generalization ability of the model.

[0027] A variety of evaluation metrics, such as accuracy, recall, F1 score, etc., are used to comprehensively evaluate the performance of the model.

[0028] Optimization method: Based on the evaluation results, fine-tune the model, such as adjusting the model architecture, increasing training data, etc.

[0029] The prediction results of multiple models are combined by using ensemble learning methods to improve the accuracy and stability of the prediction, and the model is customized and optimized for specific tasks in the field of power information and communication.

[0030] It should be noted that in order to ensure data quality and consistency, the collected text data should be of high quality and consistency, and the noise and redundant information in the data should not have a negative impact on model training; and according to the characteristics and needs of the power information and communication field, the appropriate model architecture and training strategy should be selected. During the training process, it is necessary to pay close attention to the training effect and verification results of the model, and adjust the training parameters and strategies in time. Among them, hyperparameter tuning is a complex and time-consuming process, and appropriate methods and tools need to be used for search and optimization. At the same time, it is necessary to pay attention to avoid problems such as overfitting and underfitting; finally, when evaluating the performance of the model, a variety of evaluation indicators and methods need to be used to comprehensively measure the accuracy and generalization ability of the model. When optimizing the model, customized optimization and improvement are required for specific tasks in the field of power information and communication; at the same time, the fine-tuning system should be updated and optimized continuously with the development and changes in the field of power information and communication to adapt to new data and task requirements. At the same time, the system needs to be maintained and updated regularly to ensure the stability and reliability of the system.

[0031] See also Figure 1 and Figure 2 In an embodiment of the present application, the data preprocessing module also includes a domain-specific information extraction step for extracting specific information in the field of power information communication to provide rich features for subsequent model training.

[0032] Furthermore, the model architecture selection module selects the Transformer model architecture or its variant as the basic architecture of the large language model according to the characteristics and requirements of the power information communication field.

[0033] Furthermore, the hyperparameter tuning module adjusts the hyperparameters of learning rate, batch size, and number of iterations through grid search, random search, or Bayesian optimization to optimize model performance.

[0034] Furthermore, the model evaluation and optimization module uses evaluation indicators such as accuracy, recall, F1 score, etc. to evaluate the model, and fine-tune or retrain the model based on the evaluation results to improve the accuracy and generalization ability of the model.

[0035] Furthermore, the domain-specific vocabulary enhancement submodule in the generalization processing submodule performs vocabulary enhancement processing on text data using a domain dictionary or knowledge graph, including adding synonyms and hypernyms to enhance the model's understanding of vocabulary specific to the power information communication field; The context understanding optimization submodule in the generalization processing submodule analyzes the text structure and sentence characteristics in the field of power information communication, performs semantic analysis and context reasoning on the text, and optimizes the model's ability to understand complex sentences and paragraphs; The unseen word processing submodule in the generalization processing submodule processes or replaces unseen words appearing in the text by using spelling correction, context-based guessing or external resource-based methods.

[0036] The specific information includes the name of the power equipment, the type of power fault, the power market transaction information, the power network structure information, the power dispatching instructions, the power load data, the power safety information, the power communication network information and the power equipment operation status information.

[0037] Specifically, in this embodiment, the word segmentation submodule should use a word segmentation tool based on the field of electric power information communication to perform word segmentation on the text, and establish a custom dictionary for professional terms and phrases in the electric power field to improve the accuracy of word segmentation.

[0038] The denoising submodule identifies and removes irrelevant information in the text, such as advertisements, copyright notices, etc. Regular expressions or natural language processing techniques can be used to remove redundant characters and format errors in the text.

[0039] The special character and symbol processing submodule identifies and processes special characters and symbols in the text, such as mathematical symbols, punctuation marks, etc., and converts special characters and symbols into text representation or performs standardized processing.

[0040] The case conversion submodule is used to convert all letters in the text into uppercase or lowercase to improve the consistency of the text. According to the text characteristics in the field of power information and communication, professional terms and abbreviations can be kept in their original case.

[0041] The abbreviation processing submodule establishes an abbreviation expansion table in the field of power information communication, including common abbreviations and their complete forms, and uses the abbreviation expansion table to expand the abbreviations in the text into complete words or phrases.

[0042] The generalization processing submodule and the domain-specific vocabulary enhancement submodule will identify and enhance the specific vocabulary in the field of power information communication to improve the model's ability to recognize and understand domain vocabulary. The context understanding optimization submodule: optimizes the model's ability to understand the text by analyzing the context information in the text. For example, the context information is used to infer the exact meaning and usage of the vocabulary.

[0043] Unseen word processing submodule: For the unseen words that appear in the text (that is, words that the model has not encountered during the training process), spelling correction, word replacement or context-based inference methods are used to process them.

[0044] See also Figure 1 and Figure 2 In an embodiment of the present application, a method for training and fine-tuning a large language model for power information communication is provided, the method comprising the following steps: S1. Collect text data in the field of power information and communication using a data collection module; S2, preprocessing the collected text data through the data preprocessing module, including word segmentation, denoising, processing special characters and symbols, case conversion, processing abbreviations, and generalization; S3. According to the characteristics and needs of the power information communication field, a suitable large language model architecture is selected through the model architecture selection module; S4. Based on the preprocessed data, train the selected model architecture through the model training module; S5. During the model training process, the hyperparameters of learning rate, batch size, and number of iterations are adjusted through the hyperparameter tuning module to optimize the model performance; S6. Use the model evaluation and optimization module to evaluate the trained model, and fine-tune or retrain the model based on the evaluation results until the model reaches the expected accuracy and generalization ability.

[0045] Furthermore, in the step S2, a domain-specific information extraction sub-step is further included, and the domain-specific information extraction step in the data preprocessing module is used to extract specific information in the field of power information communication to provide rich features for subsequent model training.

[0046] Furthermore, in step S5, the model is comprehensively evaluated using evaluation indicators such as accuracy, recall, and F1 score, and based on the evaluation results, the model is fine-tuned or retrained in a targeted manner in combination with domain-specific vocabulary enhancement, context understanding optimization, and unseen word processing methods to further improve the accuracy and generalization ability of the model in the field of power information communication.

[0047] Specifically, the main task of the data collection module is to collect extensive text data in the field of power information and communication. This data can come from various documents, reports, technical manuals, specifications, news articles, forum discussions, etc. in the power industry. To ensure the diversity and comprehensiveness of the data, the following measures can be taken: Multi-channel collection: Collect data from multiple channels such as the official website of the power company, industry forums, technical communities, academic databases, etc.; Regular update: With the continuous development of power information and communication technology and the introduction of new policies, the data set needs to be updated regularly to ensure the timeliness and accuracy of the model; Data deduplication: During the collection process, use data deduplication technology to avoid the impact of duplicate data on model training.

[0048] The data preprocessing module performs a series of processing on the collected text data to improve the quality and applicability of the data. The processing steps include: word segmentation: use professional word segmentation tools to segment the text and split the text into smaller language units for subsequent processing; denoising: remove irrelevant information in the text, such as advertisements, irrelevant links, garbled characters, etc., to improve the quality of the data; process special characters and symbols: uniformly process or replace special characters and symbols in the text to avoid interference with model training; case conversion: uniformly process the uppercase and lowercase in the text, such as converting all to lowercase to reduce data inconsistency; process abbreviations: expand or replace common abbreviations in the field of power information and communication to improve the recognition ability of the model; generalization processing: generalize synonyms and near synonyms in the text to improve the generalization ability of the model.

[0049] In addition, the data preprocessing module also includes a domain-specific information extraction sub-step. This step uses natural language processing technology and domain knowledge base to extract specific information in the field of power information communication, such as equipment name, technical parameters, fault type, etc. This information can provide rich features for subsequent model training and improve the accuracy and generalization ability of the model.

[0050] When choosing a suitable large language model architecture, it is necessary to consider the characteristics and needs of the power information and communication field. The following are some recommended model architecture choices: Transformer-based models, such as BERT, GPT, etc., which have excellent performance in the field of natural language processing and are suitable for text understanding and generation tasks in the power information and communication field; Encoder-decoder models: such as BART, etc., are suitable for scenarios that need to handle text understanding and generation tasks at the same time; Decoder-only models: such as the GPT series, etc., are suitable for plain text generation tasks based on given prompts; When choosing a model architecture, it is also necessary to consider factors such as the size of the model, inference speed, and cost. According to budget constraints and actual needs, choose the most cost-effective model architecture.

[0051] In the model training module, the selected model architecture is trained using the preprocessed data. The training process includes: Data division: divide the preprocessed data into training set, validation set and test set to ensure the generalization ability and stability of the model; set training parameters: set appropriate training parameters such as learning rate, batch size, number of iterations, etc. according to the model architecture and data characteristics; train model: use the training set data to train the model, and verify and adjust the model through the validation set data.

[0052] During the model training process, the hyperparameter tuning module is responsible for adjusting hyperparameters such as learning rate, batch size, and number of iterations to optimize model performance. The specific steps are as follows: Preliminary tuning: Preliminary adjustment of hyperparameters based on experience or default settings, and observe changes in model performance; Fine tuning: Based on the preliminary tuning, use grid search, random search, or Bayesian optimization algorithms to fine-tune hyperparameters to find the optimal hyperparameter combination; Verification results: Use the validation set data to verify the tuned model to ensure that the improvement in model performance is stable and reliable.

[0053] During the hyperparameter tuning process, the model can also be comprehensively evaluated using evaluation indicators such as accuracy, recall, and F1 score. These evaluation indicators can reflect the performance of the model in different aspects, such as accuracy, completeness, and balance. Based on the evaluation results, the model can be fine-tuned or retrained to further improve the accuracy and generalization ability of the model in the field of power information communication.

[0054] In summary, the present invention integrates multiple modules including data collection, preprocessing, model architecture selection, training, hyperparameter tuning, evaluation and optimization, and realizes efficient processing and analysis of text data in the field of power information and communication. The system can accurately extract field-specific information, such as the name of power equipment and fault type, optimize the selection and adapt the Transformer advanced model architecture, improve the model's ability to understand complex texts through the generalization processing submodule, and use the hyperparameter tuning module to fine-tune key parameters such as learning rate and batch size to optimize model performance. At the same time, the system has also established a scientific evaluation system to comprehensively evaluate the trained model, and perform targeted fine-tuning or retraining based on the evaluation results, thereby significantly improving the accuracy and generalization ability of the model in processing texts in the field of power information and communication, providing strong support for the intelligent operation, maintenance and management of power systems, and promoting the intelligent development process in the field of power information and communication.

[0055] The above is a schematic description of the present invention and its implementation methods, which is not restrictive. The drawings show only one implementation method of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and creatively designs a structure and an embodiment similar to the technical solution without departing from the purpose of the invention, they should all fall within the protection scope of the present invention.

[0056] In addition, it should be understood that although this specification is described according to implementation modes, not every implementation mode contains only independent technical solutions. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should take the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A large language model training and fine-tuning system for power information communication, characterized in that: The system comprises: Data collection module, used to collect text data in the field of power information communication; A data preprocessing module is used to preprocess the text data, and the module further includes: The word segmentation submodule is used to divide text data into words; Denoising submodule, used to remove noise information from text data; Special character and symbol processing submodule, used to process special characters and symbols in text data; The case conversion submodule is used to convert the letters in the text data into uppercase or lowercase uniformly; an abbreviation processing submodule, for processing abbreviations, including establishing an abbreviation expansion table in the field of electric power information communication, and expanding abbreviations into complete words or phrases; The generalization processing submodule is used to improve the model's generalization ability for texts in the field of power information and communication, including a domain-specific vocabulary enhancer, a context understanding optimization submodule, and an unseen word processing submodule; Model architecture selection module, used to select a large language model architecture suitable for the power information communication field; Model training module, used to train the selected model architecture based on the preprocessed data; Hyperparameter tuning module, used to adjust hyperparameters during model training to optimize model performance; The model evaluation and optimization module is used to evaluate the performance of the trained model and optimize the model based on the evaluation results.

2. The power information communication large language model training and fine-tuning system according to claim 1 is characterized in that: The data preprocessing module also includes a domain-specific information extraction step for extracting specific information in the field of power information communication to provide rich features for subsequent model training.

3. The power information communication large language model training and fine-tuning system according to claim 1 is characterized in that: The model architecture selection module selects the Transformer model architecture or its variant as the basic architecture of the large language model according to the characteristics and requirements of the power information communication field.

4. The power information communication large language model training and fine-tuning system according to claim 1 is characterized in that: The hyperparameter tuning module adjusts the hyperparameters of learning rate, batch size, and number of iterations through grid search, random search, or Bayesian optimization to optimize model performance.

5. The power information communication large language model training and fine-tuning system according to claim 1 is characterized in that: The model evaluation and optimization module uses evaluation indicators such as accuracy, recall, and F1 score to evaluate the model, and fine-tunes or retrains the model based on the evaluation results to improve the accuracy and generalization ability of the model.

6. The power information communication large language model training and fine-tuning system according to claim 1 is characterized in that: The domain-specific vocabulary enhancement submodule in the generalization processing submodule performs vocabulary enhancement processing on text data using a domain dictionary or knowledge graph, including adding synonyms and hypernyms to enhance the model's understanding of vocabulary specific to the power information communication field; The context understanding optimization submodule in the generalization processing submodule analyzes the text structure and sentence characteristics in the field of power information communication, performs semantic analysis and context reasoning on the text, and optimizes the model's ability to understand complex sentences and paragraphs; The unseen word processing submodule in the generalization processing submodule processes or replaces unseen words appearing in the text by using spelling correction, context-based guessing or external resource-based methods.

7. The power information communication large language model training and fine-tuning system according to claim 2 is characterized in that: The specific information includes the name of the power equipment, the type of power fault, the power market transaction information, the power network structure information, the power dispatching instructions, the power load data, the power safety information, the power communication network information and the power equipment operation status information.

8. A method for training and fine-tuning a large language model for power information communication, characterized in that: Applied to the power information communication large language model training and fine-tuning system according to any one of claims 1 to 7, the method comprising the following steps: S1. Collect text data in the field of power information and communication using a data collection module; S2, preprocessing the collected text data through the data preprocessing module, including word segmentation, denoising, processing special characters and symbols, case conversion, processing abbreviations, and generalization; S3. According to the characteristics and needs of the power information communication field, a suitable large language model architecture is selected through the model architecture selection module; S4. Based on the preprocessed data, train the selected model architecture through the model training module; S5. During the model training process, the hyperparameters of learning rate, batch size, and number of iterations are adjusted through the hyperparameter tuning module to optimize the model performance; S6. Use the model evaluation and optimization module to evaluate the trained model, and fine-tune or retrain the model based on the evaluation results until the model reaches the expected accuracy and generalization ability.

9. The method for training and fine-tuning a large language model for electric power information communication according to claim 8, characterized in that: In the step S2, a domain-specific information extraction sub-step is further included, and the domain-specific information extraction step in the data preprocessing module is used to extract specific information in the field of power information communication to provide rich features for subsequent model training.

10. The method for training and fine-tuning a large language model for electric power information communication according to claim 8, characterized in that: In step S5, the model is comprehensively evaluated using evaluation indicators such as accuracy, recall, and F1 score, and based on the evaluation results, the model is fine-tuned or retrained in a targeted manner in combination with domain-specific vocabulary enhancement, context understanding optimization, and unseen word processing methods to further improve the accuracy and generalization ability of the model in the field of power information communication.

Images