Structured operation and maintenance document generation method and system and medium

Abstract

The invention relates to a structured operation and maintenance document generation method and system and a medium, and relates to the technical field of data structuralization, and the method comprises the following steps: obtaining an unstructured operation and maintenance document, and preprocessing the unstructured operation and maintenance document to obtain a preprocessed unstructured operation and maintenance document; analyzing the preprocessed unstructured operation and maintenance document to obtain a multi-modal fusion information packet; generating a cue word template, generating cue words according to the cue word template and the multi-modal fusion information packet, and adjusting the cue words to obtain optimized cue words; and generating a structured operation and maintenance document based on the optimized cue word and the multi-modal fusion information packet. According to the technical scheme, the unstructured operation and maintenance document of the equipment can be converted into the structured operation and maintenance document, and data structured processing of the unstructured operation and maintenance document is achieved.

Description

Technical Field

[0001] This application relates to the technical field of data structuring, and particularly to a method, system, and medium for generating structured operation and maintenance documents. Background Art

[0002] At the current stage, there are many shortcomings in the processing of unstructured operation and maintenance documents based on large models. At the actual operation process level, it overly relies on manual intervention and seriously lacks end-to-end information processing capabilities. For example, in key links such as determining key operation and maintenance information and screening and adapting solutions, designers have to intervene deeply, consuming a lot of energy, which becomes a huge obstacle to efficiency improvement; instruction design is deeply trapped in the dilemma of subjective experience, lacking unified and objective standards, resulting in poor generality and being difficult to be widely promoted in multiple scenarios; the evaluation system also has many problems. It is not only limited by preset goals but also faces the problem of time-consuming and laborious construction of complex test benches. Coupled with the token (a string or data unit used to verify identity, authorize access, or represent certain information in the system) limit of the API (Application Programming Interface), it is impossible to comprehensively and accurately measure the model performance. In addition, in the face of unstructured operation and maintenance text data, traditional multi-modal models expose serious deficiencies in generalization ability. Large language models can only handle single-modal data, have great difficulties in adapting to prompts, are extremely prone to generating hallucination information contrary to facts, have weak error handling capabilities, and it is difficult to balance efficiency and quality in the preprocessing strategy, and the text segmentation effect is also unsatisfactory.

[0003] Therefore, there is an urgent need to find a method to convert unstructured operation and maintenance documents into structured operation and maintenance documents to facilitate better processing of operation and maintenance text data. Summary of the Invention

[0004] The technical problem to be solved by this application is to provide a method, system, and medium for generating structured operation and maintenance documents.

[0005] The technical solution adopted by this application to solve its technical problems is: providing a method for generating structured operation and maintenance documents, including:

[0006] Obtaining an unstructured operation and maintenance document, preprocessing the unstructured operation and maintenance document to obtain a preprocessed unstructured operation and maintenance document;

[0007] Parsing the preprocessed unstructured operation and maintenance document to obtain a multi-modal fusion information package;

[0008] Generating a prompt template, generating a prompt according to the prompt template and the multi-modal fusion information package, and adjusting the prompt to obtain an optimized prompt;

[0009] Generate a structured operation and maintenance document based on the optimized prompt and the multimodal fusion information package.

[0010] Optionally, the step of preprocessing the unstructured operation and maintenance document includes:

[0011] Add metadata tags to the unstructured operation and maintenance document, where the metadata tags include the device unique identification number, the document generation time, the operation and maintenance project name, and the device operation and maintenance location.

[0012] Optionally, the steps of generating a prompt template and generating a prompt based on the prompt template and the multimodal fusion information package include:

[0013] Adjust the prompt large model based on the prompt template data;

[0014] Input the metadata of the multimodal fusion information package into the prompt large model to generate the prompt template;

[0015] Fill the key information of the multimodal fusion information package into the prompt template to generate the prompt.

[0016] Optionally, after the step of generating a structured operation and maintenance document based on the optimized prompt and the fusion information package, the following steps may further be included:

[0017] Screen out sample data from the structured operation and maintenance document;

[0018] Determine the sample metadata tags corresponding to the sample data in the metadata tags;

[0019] Screen out the actual data corresponding to the sample metadata tags from the unstructured operation and maintenance document;

[0020] Compare the sample data with the actual data to judge the consistency between the sample data and the actual data;

[0021] Determine feedback information based on the judgment result, and input the feedback suggestion into the prompt large model.

[0022] Optionally, the step of parsing the preprocessed unstructured operation and maintenance document to obtain a multimodal fusion information package includes:

[0023] Parse the text data, image data, and audio data of the preprocessed unstructured document based on a multimodal large model;

[0024] Associate and integrate the text data, the image data, and the audio data into the multimodal fusion information package.

[0025] Optionally, the step of generating a structured operation and maintenance document based on the optimized prompt and the fusion information packet includes:

[0026] Based on a large language model, generate a structured operation and maintenance document from the unstructured data in the multimodal fusion information packet according to the rules of the optimized prompt.

[0027] Optionally, after the step of generating a structured operation and maintenance document based on the optimized prompt and the multimodal fusion information packet, the following steps are further included:

[0028] Store the structured data of the structured operation and maintenance document;

[0029] Based on an entity extraction large model, extract the key entities of the stored structured data and the association relationships between the key entities;

[0030] Construct a knowledge spectrum graph based on the key entities and the association relationships.

[0031] Optionally, the step of storing the structured data of the structured operation and maintenance document includes:

[0032] Determine the classification level of the structured data;

[0033] Store the structured data based on the classification level.

[0034] In addition, to achieve the above object, the present application also provides a structured operation and maintenance document generation system, and the system includes the following modules:

[0035] A data processing module, configured to obtain an unstructured operation and maintenance document, preprocess the unstructured operation and maintenance document, and obtain a preprocessed unstructured operation and maintenance document;

[0036] A document parsing module, configured to parse the preprocessed unstructured operation and maintenance document to obtain a multimodal fusion information packet;

[0037] A prompt generation module, configured to generate a prompt template, generate a prompt according to the prompt template and the multimodal fusion information packet, and adjust the prompt to obtain an optimized prompt;

[0038] A document generation module, configured to generate a structured operation and maintenance document based on the optimized prompt and the multimodal fusion information packet.

[0039] In addition, to achieve the above object, the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the structured operation and maintenance document generation method as described above is implemented.

[0040] The beneficial effects of implementing the technical solution of this application are as follows: By obtaining unstructured operation and maintenance documents, and then preprocessing the unstructured operation and maintenance documents to obtain the preprocessed unstructured operation and maintenance documents; then parsing the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package; generating a prompt word template, generating prompt words according to the prompt word template and the multi-modal fusion information package, and obtaining optimized prompt words by adjusting the prompt words; according to the instructions of the optimized prompt words, generating a structured operation and maintenance document by combining the data in the multi-modal fusion information package, and finally realizing the conversion of the unstructured operation and maintenance document into a structured operation and maintenance document. Brief Description of the Drawings

[0041] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0042] Figure 1 is a schematic flowchart provided by the first embodiment of the structured operation and maintenance document generation method of this application;

[0043] Figure 2 is the first overall flowchart provided by the structured operation and maintenance document generation method of this application;

[0044] Figure 3 is the second overall flowchart provided by the structured operation and maintenance document generation method of this application. Detailed Description of the Embodiments

[0045] The following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, rather than all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0046] The embodiments of this application provide a structured operation and maintenance document generation method, referring to Figure 1 , Figure 1 is a schematic flowchart provided by the first embodiment of the structured operation and maintenance document generation method of this application. In this embodiment, the method includes steps S10 to S40:

[0047] Step S10, obtain an unstructured operation and maintenance document, preprocess the unstructured operation and maintenance document, and obtain a preprocessed unstructured operation and maintenance document;

[0048] It should be noted that unstructured operation and maintenance documents are data documents without predefined formats, standardized organizational structures, or fixed fields, and cannot be directly processed in the form of tables or databases. The information in unstructured operation and maintenance documents can come from devices, fault reports, maintenance records, operation manuals, etc., and this information has no clear tags or classifications.

[0049] Specifically, collect multi-modal unstructured operation and maintenance documents related to enterprise equipment, strive not to miss any document that may contain important operation and maintenance clues, ensure that the data sources are extensive and diverse, and comprehensively cover all aspects of daily inspection, sudden faults, maintenance and upgrades of equipment operation and maintenance. Then preprocess the unstructured operation and maintenance documents to obtain the preprocessed unstructured operation and maintenance documents.

[0050] In a feasible implementation manner, step S10, the steps of preprocessing the unstructured operation and maintenance documents include:

[0051] Add metadata tags to the unstructured operation and maintenance documents, where the metadata tags include the unique device identification number, document generation time, operation and maintenance project name, and equipment operation and maintenance location.

[0052] Specifically, add detailed metadata tags to each collected unstructured document, including the unique device identification number, document generation time, specific associated operation and maintenance project name, key equipment operation and maintenance locations involved, etc. So as to quickly locate and conduct correlation analysis on the required documents in the follow-up.

[0053] In this implementation manner, by adding detailed metadata tags to the unstructured operation and maintenance documents, the efficiency of document management and search can be significantly improved.

[0054] Step S20, parse the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package;

[0055] It should be noted that the multi-modal fusion information package is a unified structure formed by fusing unstructured data of different modalities.

[0056] In a feasible implementation manner, the steps of step S20 include steps S21 - S22:

[0057] Step S21, parse the text data, image data, and audio data of the preprocessed unstructured document based on a multi-modal large model; Step S22, associate and integrate the text data, image data, and audio data into a multi-modal fusion information package.

[0058] It should be noted that a multimodal large model refers to a large deep learning model that can simultaneously process and understand multiple modalities (such as text, images, audio, etc.). Such models not only have the ability to process a single modality (for example, only process text or images), but also can integrate, analyze, and reason different types of data to provide a more comprehensive understanding and prediction.

[0059] Specifically, the unstructured data in the preprocessed unstructured operation and maintenance documents is input into the multimodal large model in batches. The multimodal large model uses a large amount of pre-trained knowledge to deeply analyze the semantic connotations of text data, accurately extract key text elements such as device names, technical parameters, operation instructions, keywords for fault descriptions, and maintenance suggestions, and at the same time parse the text logical structure to identify relationships such as causality, parallelism, and progression; for image data, it uses advanced computer vision algorithms to identify device models, appearance structures, and component states (signs of wear, corrosion, looseness), and combines image annotation technology to locate key parts; for audio data, it analyzes the running sounds of the device through audio processing and sound recognition technology, identifies the characteristics of abnormal noises, vibration sounds, or alarm sounds, and then determines whether the device has faults or abnormal states.

[0060] Exemplarily, taking the device fault scenario as an example, the position of the damaged component in the image of the faulty device is corresponded to the occurrence point of the fault phenomenon described in the text, and referring to the start time of the abnormal sound in the audio and the time stamp of the device operation state change, a multimodal fusion information packet is integrated and output in the form of a high-dimensional vector. Each dimension of the vector carries different modal key information and fusion weights, providing rich and accurate materials for subsequent structured conversion.

[0061] In this embodiment, by deeply fusing data of multiple modalities, a more comprehensive and accurate analysis of the device state can be achieved. Through the analysis of the multimodal large model, not only can key information such as fault descriptions and maintenance suggestions be extracted from the text, but also intuitive features such as the appearance state of the device and component damage can be identified by combining image data, and at the same time, audio data can be used to determine whether there are abnormal sounds or operation problems with the device. This cross-modal integration greatly enhances the data expression ability and information relevance. In addition, through the generated multimodal fusion information packet, output in the form of a high-dimensional vector, data of various modalities can be integrated into a unified structure, providing richer and more accurate materials for subsequent structured conversion, model training, and fault prediction.

[0062] Step S30, generate a prompt template, generate a prompt according to the prompt template and the multimodal fusion information packet, and adjust the prompt to obtain an optimized prompt;

[0063] It should be noted that the prompt template refers to a predefined structured format used to guide the generation of prompts for specific tasks or goals. The prompt template ensures the accuracy of the generated prompts through a fixed structure and keywords.

[0064] Specifically, by generating a prompt template, filling the data in the multimodal fusion package into the prompt template to generate a prompt, and then adjusting the prompt according to the semantics to obtain an optimized prompt.

[0065] Step S40: Generate a structured operation and maintenance document based on the optimized prompt and the multimodal fusion information package.

[0066] Optionally, step S40 further includes: based on the large language model, generating a structured operation and maintenance document for the unstructured data in the multimodal fusion information package according to the rules of the optimized prompt.

[0067] Specifically, input the optimized prompt and the fusion information package into the large language model for structured data conversion, prompting the model to perform structured conversion on the multimodal information according to the guidance of the prompt, using its powerful semantic understanding and logical reasoning capabilities, and outputting structured data or structured texts such as fault diagnosis reports, equipment maintenance plans, and performance optimization solutions, providing strong support for subsequent operation and maintenance decision-making and operations.

[0068] In this embodiment, by obtaining the unstructured operation and maintenance document, preprocessing the unstructured operation and maintenance document to obtain the preprocessed unstructured operation and maintenance document; then parsing the preprocessed unstructured operation and maintenance document to obtain the multimodal fusion information package; generating a prompt template, generating a prompt according to the prompt template and the multimodal fusion information package, and obtaining an optimized prompt by adjusting the prompt; generating a structured operation and maintenance document according to the instructions of the optimized prompt, combined with the data in the multimodal fusion information package, finally realizing the conversion of the unstructured operation and maintenance document into a structured operation and maintenance document.

[0069] The present application also provides a second embodiment. The same or similar content in the second embodiment and the above embodiments will not be elaborated here. In this embodiment, the steps of generating a prompt template in step S30 and generating a prompt according to the prompt template and the multimodal fusion information package include steps S31 to S33:

[0070] Step S31: Adjust the prompt large model based on the prompt template data; Step S32: Input the metadata of the multimodal fusion information package into the prompt large model to generate a prompt template; Step S33: Fill the key information of the multimodal fusion information package into the prompt template to generate a prompt.

[0071] It should be noted that the prompt template data refers to a structured data set used to adjust the prompt large model, which includes descriptive information related to the task, objectives, input data types, required output formats, and task context, etc. These data provide the framework and direction for the model to generate prompts, enabling the model to accurately generate prompts according to the task requirements. The prompt template refers to a predefined structured format used to guide the generation of prompts (prompts) for specific tasks or objectives. The prompt template ensures the accuracy of the generated prompts through a fixed structure and keywords. The prompt refers to the specific text content generated through the prompt template, which is used to guide the large model to perform specific tasks or analysis processes. By filling in the key information in the template, the prompt provides clear instructions or information guidance to help the large model accurately analyze, judge, or predict when processing multimodal data.

[0072] In addition, the prompt large model is a model specifically used to generate prompts. The key information refers to the information in the metadata related to the operation and maintenance of the designed device, such as the device failure location, fault details described in the text, etc. The device metadata input into the prompt large model refers to data such as device model, timestamp, business scenario identifier, etc.

[0073] Specifically, input the prompt template data into the prompt large model, and use these data to train the prompt model to generate accurate prompt templates. Then input the metadata in the multimodal fusion information package (such as device model, timestamp, business scenario identifier, etc.) into the prompt large model to automatically generate an adapted prompt template. Then, fill the key information of the multimodal fusion information package into the parameter positions in the prompt template through the large language model, and combine the current state of the model (such as knowledge update situation, recent processing accuracy) with the actual operation and maintenance requirements to fine-tune and optimize the filled prompt to ensure its accurate adaptation to the current task.

[0074] In this embodiment, by adjusting the prompt large model, the large model can better understand the task requirements and generate prompt templates that meet the requirements. By inputting the metadata of the multimodal fusion information package into the prompt large model, the model can combine the key information of the device (such as device model, timestamp, business scenario identifier, etc.) to generate more accurate and practical operation and maintenance requirement-compliant prompt templates. By fine-tuning and optimizing the prompts, the model can dynamically adapt to the current device status and the latest operation and maintenance knowledge, further improving the processing accuracy.

[0075] This application also provides a third embodiment. The same or similar content as in the above embodiments will not be repeated here. In this embodiment, after step S40 of generating a structured operation and maintenance document based on the optimized prompt and the fusion information package, steps A41 to A45 can also be included:

[0076] Step A41: Screen out sample data from the structured operation and maintenance documents;

[0077] Specifically, according to the pre-set rules and standards, representative sample data is screened out from the massive structured data. These samples cover various types of data under different equipment types, operation and maintenance scenarios, and time spans.

[0078] Step A42: Determine the sample metadata tags corresponding to the sample data in the metadata tags; Step A43: Screen out the actual data corresponding to the sample metadata tags from the unstructured operation and maintenance documents;

[0079] Specifically, among the types of metadata tags, the metadata tags corresponding to the sample data are determined, that is, the sample metadata tags. Then, the multi-modal large model is used to extract the actual data under the sample metadata tags from the unstructured operation and maintenance documents matching the sample metadata tags. The actual data here is the same as the key information mentioned in the above embodiments.

[0080] Step A44: Compare the sample data with the actual data to judge the consistency between the sample data and the actual data;

[0081] Specifically, the large language model is used to compare the structured sample data with the actual data to judge whether they are consistent.

[0082] Optionally, after the comparison by the large language model, manual comparison can also be carried out for secondary verification.

[0083] Specifically, professional operation and maintenance personnel are organized to conduct secondary verification on the extracted and preliminarily compared samples. Relying on rich practical experience and on-site intuitive feelings, the operation and maintenance personnel carefully examine the data from the perspectives of the feasibility of actual operation and maintenance and the rationality of details.

[0084] Step A45: Determine the feedback information based on the judgment result and input the feedback suggestions into the prompt large model.

[0085] Specifically, based on the judgment result, the feedback information is determined. The large language model is used to integrate the feedback information, and the inconsistent data points, potential error points, and optimization suggestions are detailedly marked, and then regularly fed back to the prompt large model. After receiving the feedback, the prompt large model focuses on optimizing the prompt framework for structured data generation. Through in-depth learning of the feedback information, the semantic expression, logical structure, and parameter settings of the prompt are readjusted to enable it to more accurately guide the subsequent generation of structured data.

[0086] Optionally, the feedback information can be equivalent to the prompt template data for fine-tuning the prompt large model.

[0087] Exemplarily, if it is found that the structured data for a certain type of equipment fault diagnosis frequently shows deviations, the prompt word large model will specifically refine the prompt words related to the equipment fault, highlight the key diagnosis elements, optimize the association guidance between the fault phenomenon and the cause, ensure that the structured data generated in the next round can be significantly improved in terms of accuracy, integrity, and practicality, and then continuously improve the entire intelligent processing process of the operation and maintenance documents, providing more reliable data support for the enterprise equipment operation and maintenance.

[0088] In this embodiment, by verifying the structured data and transmitting the feedback information to the prompt word large model to optimize the prompt word large model, the subsequent generated prompt word template is made more accurate.

[0089] Optionally, after step S40 of generating the structured operation and maintenance document based on the optimized prompt words and the fusion information package, steps B41 to S43 are further included:

[0090] Step B41, storing the structured data of the structured operation and maintenance document;

[0091] Specifically, the structured data generated by the large language model or structured texts such as fault diagnosis reports, equipment maintenance plans, and performance optimization plans are stored in the enterprise operation and maintenance database strictly in accordance with the unified data format specifications.

[0092] In a feasible embodiment, the step of step B41 of storing the structured data of the structured operation and maintenance document further includes: determining the classification level of the structured data; storing the structured data based on the classification level.

[0093] Specifically, for various structured data, a classification storage architecture is carefully built. For example, the equipment name is used as the first-level classification directory, the specific category of the operation and maintenance project is used as the second-level directory, and the third-level directory is set in chronological order.

[0094] In this embodiment, by classifying and storing the structured data, it is ensured that the data is stored in an orderly manner, effectively guaranteeing the integrity and coherence of the operation and maintenance data, and building a complete data resource treasure house for long-term operation and maintenance analysis work.

[0095] Step B42, extracting the key entities and the association relationships between the key entities of the stored structured data based on the entity extraction large model;

[0096] Specifically, the entity extraction large model is fine-tuned, and the entity extraction large model is used to accurately extract the key entities, covering core elements such as equipment, components, fault categories, and maintenance means. At the same time, the mutual relationships between the entities are carefully identified, such as the composition relationship between the equipment and its affiliated components, and the matching relationship between a specific fault type and the corresponding maintenance method.

[0097] Step B43: Construct a knowledge spectrum graph based on key entities and their association relationships.

[0098] Specifically, different entities (such as devices, components, fault categories, maintenance methods, etc.) are represented in the form of nodes, and directed or undirected edges are established through the association relationships between entities. These edges reflect various relationships between entities, such as the composition relationship between a device and its components, the causal relationship between a fault and a maintenance method, the similarity between different devices, etc., and finally form a knowledge graph. In addition, with the continuous update of new structured data, the knowledge graph is continuously iteratively optimized to dynamically reflect the latest knowledge achievements and practical experiences in the enterprise operation and maintenance field at all times.

[0099] When constructing the knowledge spectrum graph, the entity extraction large model will also assign weights to different nodes and edges according to factors such as the strength and importance of the entities and their relationships, further optimizing the structure of the graph. The structured knowledge graph can not only help visually display the knowledge network of equipment operation and maintenance, but also perform in-depth analysis through graph algorithms to discover potential laws and hidden information, such as early warnings of equipment failures and key factors of component damage, thus providing a strong foundation for the decision support system.

[0100] Exemplarily, for the sake of understanding, in combination with Figure 2 、 3 and the above embodiments, the overall technical solution of this application is described as follows:

[0101] Referring to Figure 2 , the overall process of this application combining the methods of the above multiple embodiments is as follows: First, unstructured documents are collected, such as equipment fault records, maintenance manuals, operation guides, etc. Then, the collected unstructured documents are processed by the large model. The document processing includes a document parsing layer and a structured processing layer. After parsing and structured processing, the large model generates a structured document containing structured data. Then, for these structured data, the large model stores the structured data in the form of a knowledge graph and classification storage. Finally, the large model and manual verification feedback are used to verify whether the structured data and the unstructured data under the same metadata label are consistent, and feedback information is obtained according to the judgment result, and the large model is fine-tuned based on the feedback information to make the structured data in the generated structured document more accurate.

[0102] Referring to Figure 3, taking the operation and maintenance of a wind farm as an example, in the operation and maintenance scenario of a wind power plant, the collection and annotation of multi-modal unstructured operation and maintenance documents is the first step. When performing daily inspections, maintenance personnel use a tablet computer to record the conditions of various components of the wind turbine and attach photos of key parts; there are also online meeting records of exchanges with the wind turbine manufacturer. After that, intelligent tools are used to add metadata tags such as equipment numbers and generation times to each document to make the materials organized.

[0103] Then it enters the non-structured document parsing link based on the multi-modal large model. Input various types of preprocessed modal data into the multi-modal large model, which can analyze the semantics of the wind turbine inspection text, extract key information about equipment and faults, and analyze the text logic; identify the equipment model and damaged parts for the fault images. Finally, integrate the multi-modal information and output a fusion information package to provide accurate materials for the subsequent process.

[0104] Then start the generation of structured data based on the large language model and prompt engineering. Use the prompt large model to generate a suitable prompt framework according to the content of the wind turbine fusion information package, then fill in the key information of the fusion information package, optimize the prompt in combination with the state of the prompt large model and operation and maintenance requirements to obtain the final optimized prompt, and then input the optimized prompt and the wind turbine fusion information package into the large language model to prompt it to output a structured document containing structured data.

[0105] The structured data in the structured document needs to be stored for subsequent use. On the one hand, it is strictly stored in the operation and maintenance database according to the format specifications, classified and stored by equipment name, operation and maintenance items, time, etc., and new and old data will be integrated, redundant information will be cleaned, and the cross-equipment fault relationship will be analyzed by association; on the other hand, a knowledge graph will be constructed, entities and relationships such as wind turbines, components, faults, and maintenance methods will be extracted, and visualized for display, so that maintenance personnel can quickly retrieve and assist in decision-making when encountering faults.

[0106] Finally, in the structured data verification and feedback stage based on the large model and manual comparison. Randomly select structured data samples, automatically match them with the actual operation and maintenance documents according to the metadata tags, use the multi-modal large model to extract the key information of the documents, and then use the large language model to match the core content of the extracted data with the key information to verify the accuracy and integrity of the key information. Then arrange for maintenance personnel to conduct manual verification, supplement practical feasibility and detail rationality issues that the model may ignore, integrate the comparison results and feedback them to the prompt large language model to prompt it to optimize the content of the prompt engineering, improve the quality of subsequent structured data, and ensure the efficient development of wind farm operation and maintenance work.

[0107] In addition, the present application also provides a structured operation and maintenance document generation system, and the system includes the following modules:

[0108] A data processing module, configured to obtain unstructured operation and maintenance documents, preprocess the unstructured operation and maintenance documents to obtain preprocessed unstructured operation and maintenance documents;

[0109] A document parsing module, configured to parse the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package;

[0110] A prompt word generation module, configured to generate a prompt word template, generate a prompt word according to the prompt word template and the multi-modal fusion information package, and adjust the prompt word to obtain an optimized prompt word;

[0111] A document generation module, configured to generate a structured operation and maintenance document based on the optimized prompt word and the multi-modal fusion information package.

[0112] In addition, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the structured operation and maintenance document generation method as described above is implemented.

[0113] In addition, those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by functional modules in the device, which will not be elaborated here.

[0114] The above-disclosed are only the preferred embodiments of the present application. Of course, the scope of the rights of the present application cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of the above embodiments, and the equivalent changes made according to the claims of the present application still fall within the scope covered by the invention.

Claims

1. A method for generating a structured operation and maintenance document, characterized in that, The method includes: Obtain unstructured operation and maintenance documents, preprocess the unstructured operation and maintenance documents to obtain preprocessed unstructured operation and maintenance documents; Parse the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package; Generate a prompt template, generate a prompt according to the prompt template and the multi-modal fusion information package, and adjust the prompt to obtain an optimized prompt; Generate a structured operation and maintenance document based on the optimized prompt and the multi-modal fusion information package.

2. The structured operation and maintenance document generation method according to claim 1, wherein The step of preprocessing the unstructured operation and maintenance documents includes: Add metadata tags to the unstructured operation and maintenance documents, where the metadata tags include device unique identification numbers, document generation times, operation and maintenance project names, and device operation and maintenance parts.

3. The structured operation and maintenance document generation method according to claim 2, wherein The steps of generating a prompt template and generating a prompt according to the prompt template and the multi-modal fusion information package include: Adjust the prompt large model based on prompt template data; Input the metadata of the multi-modal fusion information package into the prompt large model to generate the prompt template; Fill the key information of the multi-modal fusion information package into the prompt template to generate the prompt.

4. The structured operation and maintenance document generation method according to claim 2, wherein After the step of generating a structured operation and maintenance document based on the optimized prompt and the fusion information package, the following steps may further be included: Screen out sample data from the structured operation and maintenance document; Determine the sample metadata tags corresponding to the sample data in the metadata tags; Screen out the actual data corresponding to the sample metadata tags from the unstructured operation and maintenance documents; Compare the sample data with the actual data to judge the consistency between the sample data and the actual data; Determine feedback information based on the judgment result and input the feedback suggestion into the prompt large model.

5. The structured operation and maintenance document generation method according to claim 1, wherein The step of parsing the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package includes: Parse the text data, image data, and audio data of the preprocessed unstructured document based on a multi-modal large model; Associate and integrate the text data, the image data, and the audio data into the multi-modal fusion information package.

6. The structured operation and maintenance document generation method according to claim 1, wherein The step of generating a structured operation and maintenance document based on the optimized prompt and the fusion information package includes: Based on a large language model, generate a structured operation and maintenance document for the unstructured data in the multi-modal fusion information package according to the rules of the optimized prompt.

7. The structured operation and maintenance document generation method according to claim 1, wherein, After the step of generating a structured operation and maintenance document based on the optimized prompt and the multi-modal fusion information package, the following steps are further included: Store the structured data of the structured operation and maintenance document; Extract the key entities and the association relationships between the key entities of the stored structured data based on an entity extraction large model; Construct a knowledge spectrum diagram based on the key entities and the association relationships.

8. The structured operation and maintenance document generation method according to claim 7, wherein, The step of storing the structured data of the structured operation and maintenance document includes: Determine the classification level of the structured data; Store the structured data based on the classification level.

9. A structured operation and maintenance document generation system, characterized in that, The system includes the following modules: A data processing module, configured to obtain unstructured operation and maintenance documents, preprocess the unstructured operation and maintenance documents to obtain preprocessed unstructured operation and maintenance documents; A document parsing module, configured to parse the preprocessed unstructured operation and maintenance documents to obtain a multi-modal fusion information package; A prompt word generation module, configured to generate a prompt word template, generate prompt words according to the prompt word template and the multi-modal fusion information package, and adjust the prompt words to obtain optimized prompt words; A document generation module, configured to generate structured operation and maintenance documents based on the optimized prompt words and the multi-modal fusion information package.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the structured operation and maintenance document generation method according to any one of claims 1 to 8.

Images