Substation text auditing method, system and equipment based on large model and medium

Abstract

The invention relates to the technical field of large models, in particular to a substation text auditing method, system and device based on a large model and a medium. According to the method, specialized preprocessing is carried out on power transformation text data to construct a training data set, then a low-rank adaptation and thinking chain combination technology is adopted to train a model, and finally an automatic auditing process is executed to generate a result. The method improves the accuracy and efficiency of text auditing, has high expansibility, can flexibly cope with continuously changing business requirements, and effectively saves resources of manual rechecking.

Description

Technical Field

[0001] The present application relates to the technical field of large models, and in particular to a large model-based substation text review method, system, device and medium. Background Art

[0002] With the continuous advancement of smart grid construction, the types and quantity of substation equipment are growing rapidly, and the operation and maintenance work is becoming more and more complicated. Substation operation, inspection, maintenance and other work involve a large amount of professional text materials, and the accuracy and standardization of these materials are directly related to the safe and stable operation of the power system.

[0003] At present, substation text processing mainly relies on the assistance of automated software systems, including basic functions such as text format checking, key information extraction, and data consistency verification. These systems are mainly based on traditional natural language processing technology, which performs preliminary structural analysis and processing of texts. They lack the ability to understand substation professional terms and complex semantics, have low processing accuracy, and require a lot of manual review. This situation needs to be further improved. Summary of the invention

[0004] In order to solve the problem that the existing substation text processing accuracy is low and a large amount of manual review is required, the present application provides a substation text review method, system, device and medium based on a large model, which adopts the following technical solutions: In a first aspect, the present application provides a large model-based substation text review method, comprising the following steps: Preprocess the substation text data to obtain the training data set; According to the training data set, a model is trained by using a combination of low-rank adaptation and thought chain technology to obtain a text review model; Based on the text audit model, the substation automation audit process is executed to obtain the audit result.

[0005] By adopting the above technical solution, due to the rapid growth in the types and number of substation equipment, the amount of professional text materials that substation operation and maintenance personnel need to handle has increased sharply. These materials include operating procedures, operation tickets, maintenance plans, etc. Each document must strictly follow professional standards and safety requirements; this application first performs professional preprocessing on the substation text materials to construct a training data set, and then uses low-rank adaptation and thinking chain combination technology to train the model, and finally executes an automated review process to generate results; it improves the accuracy and efficiency of text review, and has strong scalability, can flexibly respond to changing business needs, and effectively saves resources for manual review.

[0006] Optionally, the substation text data includes substation regulations and operation manuals; preprocessing the substation text data to obtain a training data set specifically includes the following steps: Read substation regulations and operation manual files, perform standardization processing, and obtain structured original texts; Based on the structured original text, term features are extracted and a knowledge graph is constructed, a term mapping relationship table is established, and term index data is obtained; Using the term index data, perform term standardization recognition and replacement processing on the text to generate a standardized text; A word segmentation algorithm and data cleaning rules are applied to the normalized text to generate a training data set in a standard format.

[0007] By adopting the above technical solution, this application first systematically collects substation regulations and operating manuals as original texts, constructs a professional dictionary system that includes standard definitions of terms, usage specifications and scenario associations, and then identifies and normalizes terms based on the dictionary. Finally, a training data set is generated through word segmentation and cleaning steps; it achieves a deep understanding and standardized processing of substation professional texts, which not only improves data quality, but also provides a reliable professional knowledge foundation for model training.

[0008] Optionally, according to the training data set, a model is trained by using a combination of low-rank adaptation and thought chain technology to obtain a text review model, which specifically includes the following steps: According to the training data set, low-rank adaptation training is performed using a low-rank adaptation technique to obtain a preliminary model; Construct a thinking chain template according to the substation audit requirements; Based on the preliminary model and the thought chain template, combined training is performed to obtain a text review model.

[0009] By adopting the above technical solution, since the substation text review involves complex professional judgment and reasoning processes, such as the review of a maintenance plan needs to consider multiple dimensions such as equipment parameters, operation sequence, safety measures, etc., the traditional model training method often only focuses on the recognition of a single feature, and it is difficult to achieve deep logical reasoning; this application first uses low-rank adaptation technology to efficiently customize the pre-trained model so that it can acquire basic understanding capabilities in the substation field, and then constructs a thinking chain template containing professional judgment logic based on the substation review requirements, and finally organically combines the two technologies for model training; the professionalism of the model is guaranteed by low-rank adaptation, and the thinking chain technology is used to give the model clear reasoning capabilities, which not only improves the audit accuracy of the model, but also realizes the interpretability of the audit process.

[0010] Optionally, according to the training data set, low-rank adaptation training is performed using a low-rank adaptation technique to obtain a preliminary model, which specifically includes the following steps: According to the training data set, determining and freezing the pre-training model parameters to obtain a basic model; According to the basic model, a low-rank matrix is ​​added to the transformer to obtain an extended model; Based on the extended model, a bypass matrix is ​​trained to obtain fine-tuning parameters; The pre-trained model parameters are updated according to the fine-tuning parameters to obtain a preliminary model.

[0011] By adopting the above technical scheme, due to the professionalism and particularity of the substation field, it is difficult to accurately understand and process the substation text content by directly using a general pre-trained model, and the traditional full-parameter fine-tuning method has the problems of high training cost and low efficiency; the present application first freezes the pre-trained model parameters as the basic model, and then adds a low-rank matrix to the converter to build an extended model, and then obtains the fine-tuning parameters by training the bypass matrix, and finally uses these parameters to update the model; by introducing a low-rank matrix, the number of parameters that need to be trained is reduced, and at the same time, the design of the bypass matrix ensures the model's ability to quickly adapt to substation expertise; it not only greatly reduces the computing cost, but also improves the adaptability of the model in the substation field.

[0012] Optionally, according to the substation audit requirements, a thinking chain template is constructed, which specifically includes the following steps: Convert the substation audit requirements into a machine-recognizable instruction format to obtain an instruction set; Based on the instruction set, a standardized audit step sequence is generated to obtain a reasoning process; Formalizing the reasoning process, generating machine-executable verification rules, and obtaining a rule set; The rule set is converted into a structured thinking chain prompt template to obtain a thinking chain template.

[0013] By adopting the above technical solution, this application first extracts and converts the audit requirements in various regulations into a standardized instruction format, then constructs an audit step sequence with a clear sequence based on these instructions, and then converts the reasoning process into machine-executable verification rules through formal description, and finally generates a structured thinking chain prompt template; it not only ensures the integrity and traceability of the audit logic, but also provides a flexible template expansion mechanism, providing a reliable reasoning basis for the intelligent audit of substation text.

[0014] Optionally, based on the text audit model, a substation automation audit process is executed to obtain an audit result, which specifically includes the following steps: Receiving a text to be reviewed, inputting the text to be reviewed into the text review model, and obtaining a preliminary result; Based on the preliminary results, a reasoning process is generated to obtain an audit record; Based on the audit records, evaluation results and suggestions are output to obtain audit results.

[0015] By adopting the above technical solution, this application first inputs the text to be reviewed into the review model to obtain a preliminary judgment, then automatically generates a reasoning process containing key judgment nodes based on the output results of the model, and finally outputs instructive evaluation results and improvement suggestions by analyzing the review records; it not only achieves the transparency of the review process, but also provides targeted improvement directions, ensuring the reliability and practicality of the review results.

[0016] Optionally, outputting evaluation results and suggestions according to the audit records to obtain audit results specifically includes the following steps: Evaluate the use of substation terminology, calculate terminology accuracy, and obtain terminology scores; Evaluate the substation text format, determine the format compliance, and obtain the format score; Evaluate the substation operation steps, check the completeness of the steps, and obtain the completeness score; Based on the terminology score, format score and complete score, a substation safety assessment is performed, safety compliance is verified, and an audit result is output.

[0017] By adopting the above technical solution, this application first conducts a quantitative assessment of the standardization of terms in the text through a professional terminology library, then verifies the standardization of the text format based on the standard template, and then checks the completeness and logic of the operating steps based on the standard operating procedures. Finally, the scores of these three dimensions are weighted and integrated to perform a comprehensive safety assessment; the quantification of the quality of the substation text is achieved, and at the same time, a more reliable safety judgment is provided based on the comprehensive analysis of multi-dimensional scores, which not only improves the accuracy and credibility of the evaluation results, but also provides clear direction guidance for text quality improvement.

[0018] In a second aspect, the present application provides a large model-based substation text review system, comprising: The training data set acquisition module is used to pre-process the substation text data to obtain the training data set; A text audit model acquisition module is used to train a model using a low-rank adaptation and thought chain combination technology according to the training data set to obtain a text audit model; The audit result acquisition module is used to execute the substation automation audit process based on the text audit model to obtain the audit result.

[0019] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned large model-based substation text review method when executing the computer program.

[0020] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned large model-based substation text review method are implemented.

[0021] In summary, the present application includes at least one of the following beneficial technical effects: 1. This application first conducts professional preprocessing of substation text data to construct a training data set, then uses low-rank adaptation and thought chain combination technology to train the model, and finally executes the automated review process to generate results; it improves the accuracy and efficiency of text review, and has strong scalability, can flexibly respond to changing business needs, and effectively saves resources for manual review; 2. This application first systematically collects substation regulations and operating manuals as original texts, builds a professional dictionary system containing standard definitions of terms, usage specifications and scenario associations, then identifies and normalizes terms based on the dictionary, and finally generates a training data set through word segmentation and cleaning steps; it achieves a deep understanding and standardized processing of substation professional texts, which not only improves data quality, but also provides a reliable professional knowledge foundation for model training; 3. Since substation text review involves complex professional judgment and reasoning processes, such as the review of a maintenance plan needs to consider multiple dimensions such as equipment parameters, operation sequence, safety measures, etc., traditional model training methods often only focus on the recognition of a single feature, which makes it difficult to achieve deep logical reasoning; this application first uses low-rank adaptation technology to efficiently customize the pre-trained model so that it can acquire basic understanding capabilities in the substation field, and then constructs a thinking chain template containing professional judgment logic based on the substation review requirements, and finally organically combines the two technologies for model training; the professionalism of the model is guaranteed by low-rank adaptation, and the thinking chain technology is used to give the model clear reasoning capabilities, which not only improves the audit accuracy of the model, but also realizes the interpretability of the audit process. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of a large model-based substation text review method according to an embodiment of the present application; Figure 2 It is a flow chart of step S100 in a large model-based substation text review method according to an embodiment of the present application; Figure 3 It is a flowchart of step S200 in a large model-based substation text review method according to an embodiment of the present application; Figure 4 It is a flow chart of step S210 in a large model-based substation text review method according to an embodiment of the present application; Figure 5It is a flow chart of step S220 in a large model-based substation text review method according to an embodiment of the present application; Figure 6 It is a flowchart of step S300 in a large model-based substation text review method according to an embodiment of the present application; Figure 7 It is a flowchart of step S330 in a large model-based substation text review method according to an embodiment of the present application; Figure 8 It is a module schematic diagram of a substation text review system based on a large model according to an embodiment of the present application; Fig. 9 It is a diagram of the internal structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification and appended claims of the present application, the singular expressions "one", "a kind of", "said", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more listed items.

[0024] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0025] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0026] In the first aspect, the present application provides a method for substation text review based on a large model, referring to Figure 1 , including the following steps: S100, pre-processing the substation text data to obtain a training data set.

[0027] Among them, substation text materials include various types of professional documents such as substation operation procedures, equipment operation manuals, historical operation tickets, maintenance records, etc. These documents contain a large number of professional terms, standard operating procedures and safety regulations, and need to be systematically preprocessed before they can be used for model training. The purpose of preprocessing is to convert unstructured raw text into standardized and normalized training data sets.

[0028] Specifically, the text standardization technology is used to first unify the format and convert the encoding of the document, then use professional field word segmentation tools to perform word segmentation, then use regular expressions to identify the chapter structure and key information of the document, and finally organize the processed text into a training data set according to a predefined format. For example, for a substation switching operation ticket, the preprocessing process will extract key information such as the name of the operating equipment, operating steps, and safety measures, and convert it into a structured data format.

[0029] S200. Based on the training data set, the model is trained using low-rank adaptation and thought chain combination technology to obtain a text review model.

[0030] The model training adopts the combination of low-rank adaptation and thought chain technology, which can quickly adapt to the professional knowledge in the substation field while maintaining the core capabilities of the pre-trained model. The low-rank adaptation technology achieves efficient field adaptation by adding a small number of trainable parameters to the model; the thought chain technology introduces a structured reasoning template to enhance the logical reasoning ability of the model.

[0031] Specifically, low-rank adaptation training is first performed based on a large-scale pre-trained model, and domain features are captured by adding a training parameter matrix; then, a thinking chain reasoning template is constructed in combination with the audit specifications in the substation field to guide the model to perform standardized reasoning and judgment. For example, when reviewing a substation operation ticket, the model can perform step-by-step reasoning according to the thinking chain of "equipment name verification -> operation step integrity check -> safety measure verification".

[0032] S300: Based on the text audit model, execute the substation automation audit process to obtain the audit result.

[0033] The automated audit process is a complete text analysis and evaluation process, including input text preprocessing, model reasoning, result evaluation, etc. The audit model can conduct a comprehensive analysis of the input substation text and ultimately output detailed audit results and improvement suggestions.

[0034] Specifically, after receiving the text to be reviewed, the system first preprocesses and formats the text, and then inputs the processed text into the review model for analysis. The model will infer based on the trained parameters and thought chain templates to generate evaluation results containing multiple dimensions. For example, the review results of a substation operation ticket will include terminology standardization scores, step completeness scores, safety measures adequacy scores, etc., and provide specific problem descriptions and improvement suggestions.

[0035] In one embodiment, referring to Figure 2 The substation text data includes substation regulations and operation manuals; in step S100, the substation text data is preprocessed to obtain a training data set, which specifically includes the following steps: S110, read the substation regulations and operation manual files, perform standardization processing, and obtain structured original text.

[0036] Among them, the substation regulations mainly include equipment operating specifications, safety operating requirements and accident handling procedures, while the operating manual contains the operating steps, technical parameters and precautions for specific equipment.

[0037] Specifically, first use a document parsing tool to convert files of different formats into plain text format, and then perform standardization processing such as unified encoding conversion, special character processing, and line break unification on the text.

[0038] S120. Based on the structured original text, extract term features and construct a knowledge graph, establish a term mapping relationship table, and obtain term index data.

[0039] Among them, there are a large number of professional terms and technical terms in the field of substation, and these terms have complex semantic associations. By constructing a knowledge graph, the hierarchical relationship and association relationship between terms can be clarified, which is helpful for the subsequent terminology standardization processing. The nodes of the knowledge graph represent the term concepts, and the edges represent the relationship between terms.

[0040] Specifically, professional terms are identified from structured text, including equipment names, operating actions, technical parameters, etc. Then, based on the co-occurrence relationship of terms and context information, the semantic associations between terms are analyzed to construct a knowledge graph.

[0041] S130. Using the term index data, perform term standardization recognition and replacement processing on the text to generate a normalized text.

[0042] Among them, different expressions of the same term often appear in substation texts, which need to be standardized based on term index data. This process includes steps such as term variant identification, standardized replacement, and consistency checking, with the aim of ensuring that the terminology used in the text meets the requirements of the specification.

[0043] Specifically, the system first uses the term mapping table to match the term variants in the text, and then replaces the identified non-standard expressions with the corresponding standard terms. For example, different expressions such as "switch", "circuit breaker", and "switchgear" in the text are standardized into the standard term "circuit breaker" to ensure the consistency of terminology use.

[0044] S140. Apply word segmentation algorithm and data cleaning rules to the normalized text to generate a training data set in a standard format.

[0045] Among them, the normalized text needs to be further processed before it can be used for model training, which includes steps such as text segmentation, noise removal, and format normalization.

[0046] Specifically, a word segmentation dictionary optimized for the substation field is used to segment the text, removing irrelevant information such as stop words and special symbols in the text. Then the processed text is organized according to a predefined data format, including the input text and the standard answer.

[0047] In one embodiment, referring to Figure 3 In step S200, according to the training data set, a low-rank adaptation and thought chain combination technology is used to train the model to obtain a text review model, which specifically includes the following steps: S210. According to the training data set, low-rank adaptation training is performed using low-rank adaptation technology to obtain a preliminary model.

[0048] Among them, low-rank adaptation technology is an efficient domain adaptation method that achieves targeted adjustment of model capabilities by adding a small number of trainable parameters to the pre-trained model. This method does not require fine-tuning the entire model, but captures domain knowledge by training specific low-rank matrices, which not only maintains the general capabilities of the model, but also achieves rapid adaptation to the substation field.

[0049] Specifically, we first select a suitable pre-trained language model as the basis, and then insert low-rank adaptation matrices between the key layers of the model. The rank of these matrices is much smaller than the dimension of the original parameter matrix, and is usually selected to be 5%-10% of the original dimension. For example, for a transformation matrix with a dimension of 1024, a low-rank adaptation matrix with a rank of 64 can be inserted. During the training process, only the parameters of these low-rank matrices are updated, while the other parameters of the pre-trained model are kept fixed, so as to efficiently learn the feature representation of the substation field.

[0050] S220. Construct a thinking chain template based on the substation audit requirements.

[0051] Among them, the thinking chain template is a structured reasoning framework that guides the model to perform logical reasoning and judgment according to predefined steps. In the substation text review scenario, according to the review requirements of different types of documents, it is necessary to build a corresponding thinking chain template, including key checkpoints, judgment criteria, and reasoning paths.

[0052] S230. Based on the preliminary model and the thinking chain template, combined training is performed to obtain a text review model.

[0053] Among them, combined training is the process of fusing the preliminary model after low-rank adaptation with the thinking chain template, with the aim of enabling the model to perform standardized reasoning and judgment according to the predefined thinking chain template.

[0054] Specifically, the training data is first organized according to the structure of the thinking chain template, and corresponding training samples are prepared for each reasoning step. Then the loss function is designed, taking into account the matching degree between the model output and the standard answer, as well as the consistency between the reasoning process and the thinking chain template. For example, for the review of the substation operation ticket, the model needs to learn to first check the standardization of the equipment name, and then evaluate the completeness of the operation steps, the rationality of the operation sequence, etc., and finally give a comprehensive review result.

[0055] In one embodiment, referring to Figure 4 In step S210, low-rank adaptation training is performed using low-rank adaptation technology according to the training data set to obtain a preliminary model, which specifically includes the following steps: S211. According to the training data set, determine the pre-training model parameters and freeze them to obtain a basic model.

[0056] Among them, the pre-trained model is a language model trained on a large-scale general corpus, which has strong language understanding and generation capabilities. When applied to the substation field, it is necessary to first determine the appropriate pre-trained model and its parameter configuration, and temporarily freeze these parameters to prevent the destruction of the original capabilities during subsequent training.

[0057] S212. According to the basic model, a low-rank matrix is ​​added to the transformer to obtain an extended model.

[0058] Among them, the low-rank matrix is ​​a parameter matrix with small dimensions but can effectively capture the characteristics of the domain. These matrices are inserted into key positions of the transformer, including the self-attention layer and the feedforward network layer.

[0059] Specifically, two low-rank matrices are inserted into the attention calculation module of each transformer layer, which are used to adjust the calculation of the query vector and the key-value vector respectively. The rank of the matrix is ​​set to about 8% of the original dimension. For example, for a hidden layer with 768 dimensions, a low-rank matrix of rank 60 can be inserted. These matrices are implemented by decomposing them into two smaller matrices, namely the product of a 768×60 matrix and a 60×768 matrix.

[0060] S213. Based on the extended model, train the bypass matrix to obtain fine-tuning parameters.

[0061] Among them, the bypass matrix refers to the trainable parameter matrix added outside the original model structure. They are parallel to the main calculation path of the model and are used to adjust the output of the model.

[0062] Specifically, a set of bypass matrices are designed for each transformer layer, including an input mapping matrix and an output mapping matrix. The dimensions of these matrices match the hidden state dimensions of the corresponding layer, but low-rank decomposition is used to reduce the number of parameters.

[0063] S214. Update the pre-trained model parameters according to the fine-tuning parameters to obtain a preliminary model.

[0064] The fine-tuning parameters include the training results of the low-rank matrix and the bypass matrix, which need to be integrated with the parameters of the original pre-trained model to form a unified model. The integration process needs to ensure that the new parameters can effectively affect the behavior of the model without destroying the model's original language understanding ability.

[0065] Specifically, the trained low-rank matrix and bypass matrix are fused with the original model parameters by parameter combination. For each transformer layer, the effect of the low-rank matrix is ​​added to the original calculation path by residual connection. In this embodiment, the output of the original layer is h, and the output of the low-rank matrix is ​​h', then the final output is h+αh', where α is a learnable scaling factor.

[0066] In one embodiment, referring to Figure 5 In step S220, according to the substation audit requirements, a thinking chain template is constructed, which specifically includes the following steps: S221. Convert the substation audit requirements into a machine-recognizable instruction format to obtain an instruction set.

[0067] S222. Based on the instruction set, a standardized audit step sequence is generated to obtain the reasoning process.

[0068] Among them, the audit step sequence refers to organizing scattered audit instructions into a logically related execution sequence to form a complete reasoning process.

[0069] Specifically, the audit steps are organized in a directed acyclic graph, and the steps with dependencies are connected through edges. For example, when auditing a substation operation ticket, a step sequence such as "equipment name verification -> operation environment inspection -> operation step verification -> safety measures confirmation" can be constructed.

[0070] S223. Formalize the reasoning process, generate machine-executable verification rules, and obtain a rule set.

[0071] Among them, formal description is the process of converting the reasoning process into strict logical expressions and judgment rules.

[0072] Specifically, use conditional statements and logical operators to describe validation rules.

[0073] S224. Convert the rule set into a structured thinking chain prompt template to obtain a thinking chain template.

[0074] Among them, the thinking chain prompt template is a structured text that guides the model to reason, which converts the verification rules into a series of coherent reasoning steps.

[0075] Specifically, the template structure of "task description-reasoning steps-judgment criteria-output format" is adopted. For example, for the operation ticket review, the template can include the following content: "Please review the standardization of the following operation ticket. Step 1: Check whether the equipment name is standardized, refer to the equipment naming specification... Step 2: Verify the integrity of the operation steps and confirm whether the necessary safety measures are included..." Each step contains clear inspection content and judgment criteria, guiding the model to conduct systematic reasoning analysis.

[0076] In one embodiment, referring to Figure 6 In step S300, based on the text audit model, the substation automation audit process is executed to obtain the audit result, which specifically includes the following steps: S310, receiving the text to be reviewed, inputting the text to be reviewed into a text review model, and obtaining a preliminary result.

[0077] S320. Based on the preliminary results, generate a reasoning process and obtain an audit record.

[0078] S330. Output the evaluation results and suggestions based on the audit records to obtain the audit results.

[0079] In one embodiment, referring to Figure 7 In step S330, the evaluation results and suggestions are output according to the audit records to obtain the audit results, which specifically includes the following steps: S331. Evaluate the use of substation terminology, calculate the accuracy of terminology, and obtain terminology scores.

[0080] S332. Evaluate the substation text format, determine the format compliance, and obtain the format score.

[0081] S333. Evaluate the substation operation steps, check the completeness of the steps, and obtain a complete score.

[0082] S334. Based on the terminology score, format score and complete score, perform substation safety assessment, verify safety compliance, and output the audit results.

[0083] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0084] On the second aspect, the present application provides a substation text review system based on a large model. The substation text review system based on a large model of the present application is described below in combination with the above-mentioned substation text review method based on a large model.

[0085] Reference Figure 8, a substation text review system based on a large model, comprising: The training data set acquisition module is used to pre-process the substation text data to obtain the training data set; A text review model acquisition module is used to train a model based on a training data set by using a combination of low-rank adaptation and thought chain technology to obtain a text review model; The audit result acquisition module is used to execute the substation automation audit process based on the text audit model and obtain the audit results.

[0086] In one embodiment, the substation text data includes substation regulations and operation manuals; the training data set acquisition module includes: The text reading unit is used to read the substation regulations and operation manual files, perform standardization processing, and obtain the structured original text; The term extraction unit is used to extract term features and construct a knowledge graph based on the structured original text, establish a term mapping relationship table, and obtain term index data; A term processing unit, used to perform term standardization recognition and replacement processing on the text using the term index data to generate a standardized text; The data cleaning unit is used to apply word segmentation algorithms and data cleaning rules to the normalized text to generate a training data set in a standard format.

[0087] In one embodiment, the text audit model acquisition module includes: A low-rank adaptation training unit is used to perform low-rank adaptation training using a low-rank adaptation technique according to a training data set to obtain a preliminary model; A thinking chain template building unit is used to build a thinking chain template according to the substation audit requirements; The combined training unit is used to conduct combined training based on the preliminary model and the thinking chain template to obtain a text review model.

[0088] In one embodiment, the low-rank adaptation training unit includes: The basic model acquisition subunit is used to determine and freeze the pre-trained model parameters according to the training data set to obtain the basic model; An extended model acquisition subunit is used to add a low-rank matrix to the transformer according to the basic model to obtain an extended model; A parameter training subunit, used for training a bypass matrix based on the extended model to obtain fine-tuning parameters; The model updating subunit is used to update the pre-trained model parameters according to the fine-tuning parameters to obtain a preliminary model.

[0089] In one embodiment, the thought chain template construction unit includes: An instruction conversion subunit, used for converting the substation audit requirements into an instruction format recognizable by a machine to obtain an instruction set; The reasoning process generation subunit is used to generate a standardized audit step sequence based on the instruction set to obtain the reasoning process; The rule generation subunit is used to formally describe the reasoning process, generate machine-executable verification rules, and obtain a rule set; The template conversion subunit is used to convert the rule set into a structured thinking chain prompt template to obtain the thinking chain template.

[0090] In one embodiment, the audit result acquisition module includes: A preliminary result acquisition unit is used to receive the text to be reviewed, input the text to be reviewed into the text review model, and obtain a preliminary result; A reasoning process generating unit, used for generating a reasoning process based on the preliminary result to obtain an audit record; The result output unit is used to output the evaluation results and suggestions according to the audit records to obtain the audit results.

[0091] In one embodiment, the result output unit includes: The term evaluation subunit is used to evaluate the use of substation terminology, calculate the accuracy of terminology, and obtain the terminology score; The format evaluation subunit is used to evaluate the format of the substation text, determine the format compliance, and obtain the format score; The step evaluation subunit is used to evaluate the substation operation steps, check the completeness of the steps, and obtain the completeness score; The safety assessment subunit is used to perform substation safety assessment based on terminology scores, format scores and complete scores, verify safety compliance, and output audit results.

[0092] In one embodiment, the present application provides an electronic device, which may be a server, and its internal structure diagram may be as follows: Fig. 9 As shown. The electronic device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store data. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a substation text review method based on a large model is implemented.

[0093] Those skilled in the art will understand that Fig. 9The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0094] In one embodiment, an electronic device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0095] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0096] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A substation text review method based on a large model, characterized in that: The steps include: Preprocess the substation text data to obtain the training data set; According to the training data set, a model is trained by using a combination of low-rank adaptation and thought chain technology to obtain a text review model; Based on the text audit model, the substation automation audit process is executed to obtain the audit result.

2. The large model-based substation text review method according to claim 1 is characterized in that: The substation text data includes substation regulations and operation manuals; preprocessing the substation text data to obtain a training data set specifically includes the following steps: Read substation regulations and operation manual files, perform standardization processing, and obtain structured original texts; Based on the structured original text, term features are extracted and a knowledge graph is constructed, a term mapping relationship table is established, and term index data is obtained; Using the term index data, perform term standardization recognition and replacement processing on the text to generate a standardized text; A word segmentation algorithm and data cleaning rules are applied to the normalized text to generate a training data set in a standard format.

3. The large model-based substation text review method according to claim 1 is characterized in that: According to the training data set, a model is trained by combining low-rank adaptation and thought chain technology to obtain a text review model, which specifically includes the following steps: According to the training data set, low-rank adaptation training is performed using a low-rank adaptation technique to obtain a preliminary model; Construct a thinking chain template according to the substation audit requirements; Based on the preliminary model and the thought chain template, combined training is performed to obtain a text review model.

4. The large model-based substation text review method according to claim 3 is characterized in that: According to the training data set, low-rank adaptation training is performed using low-rank adaptation technology to obtain a preliminary model, which specifically includes the following steps: According to the training data set, determining and freezing the pre-training model parameters to obtain a basic model; According to the basic model, a low-rank matrix is ​​added to the transformer to obtain an extended model; Based on the extended model, a bypass matrix is ​​trained to obtain fine-tuning parameters; The pre-trained model parameters are updated according to the fine-tuning parameters to obtain a preliminary model.

5. The large model-based substation text review method according to claim 3 is characterized in that: According to the substation audit requirements, a thinking chain template is constructed, which includes the following steps: Convert the substation audit requirements into a machine-recognizable instruction format to obtain an instruction set; Based on the instruction set, a standardized audit step sequence is generated to obtain a reasoning process; Formalizing the reasoning process, generating machine-executable verification rules, and obtaining a rule set; The rule set is converted into a structured thinking chain prompt template to obtain a thinking chain template.

6. The large model-based substation text review method according to claim 1 is characterized in that: Based on the text audit model, the substation automation audit process is executed to obtain the audit result, which specifically includes the following steps: Receiving a text to be reviewed, inputting the text to be reviewed into the text review model, and obtaining a preliminary result; Based on the preliminary results, a reasoning process is generated to obtain an audit record; Based on the audit records, evaluation results and suggestions are output to obtain audit results.

7. The large model-based substation text review method according to claim 6 is characterized in that: According to the audit records, output the evaluation results and suggestions to obtain the audit results, which specifically includes the following steps: Evaluate the use of substation terminology, calculate terminology accuracy, and obtain terminology scores; Evaluate the substation text format, determine the format compliance, and obtain the format score; Evaluate the substation operation steps, check the completeness of the steps, and obtain the completeness score; Based on the terminology score, format score and complete score, a substation safety assessment is performed, safety compliance is verified, and an audit result is output.

8. A substation text review system based on a large model, characterized in that: include: The training data set acquisition module is used to pre-process the substation text data to obtain the training data set; A text audit model acquisition module is used to train a model using a low-rank adaptation and thought chain combination technology according to the training data set to obtain a text audit model; The audit result acquisition module is used to execute the substation automation audit process based on the text audit model to obtain the audit result.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the large model-based substation text review method described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the large model-based transformer text review method described in any one of claims 1-7 are implemented.

Images