Operation and maintenance workflow generation method and device

By fine-tuning the open source large language model and multimodal operation and maintenance knowledge base, we automatically extract operation steps from the operation and maintenance documents and generate task workflows, solving the complex operation and maintenance scenario sorting, irregular script writing and lack of agents in traditional operation and maintenance orchestration. The monitoring problem has achieved efficient and stable development and deployment of operation and maintenance tasks.

CN118939347APending Publication Date: 2024-11-12CCB FINTECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411101994.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The process-based orchestration of traditional operation and maintenance scripts has problems such as complex operation and maintenance scenario sorting, irregular script writing, and lack of monitoring and logging of agents, resulting in irregular operation and maintenance tasks development, difficult to maintain, and poor stability.

Method used

By using the fine-tuning open source large language model to automatically extract operation steps from the operation and maintenance document, and combine the multimodal operation and maintenance knowledge base to generate task workflows, and then perform audit tests and deploy to the production environment. , ensure the standardization and stability of operation and maintenance tasks.

Benefits of technology

It significantly improves the speed of operation and maintenance tasks development, reduces manual intervention, ensures the consistency and standardization of operation and maintenance tasks, and enhances the safety and stability of operation and maintenance tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939347A_ABST
    Figure CN118939347A_ABST
Patent Text Reader

Abstract

The invention discloses an operation and maintenance workflow generation method and device, and relates to the technical field of intelligent operation and maintenance. A specific embodiment of the method comprises the following steps: receiving an operation and maintenance document, and extracting operation steps from the operation and maintenance document by using a planner to obtain an operation step set; identifying a task scene of each operation step by using a scheduler, and calling a multi-modal operation and maintenance knowledge base according to a task workflow generation mode configured for each task scene in advance to obtain a task workflow of each operation step; and determining the types of the task workflows by using an executor, processing each task workflow according to an auditing test mode configured for each type in advance, and generating an operation and maintenance scene workflow set based on the task workflow set under the condition that an auditing test result is passed so as to be deployed into a production environment. According to the implementation mode, the operation and maintenance efficiency, normalization and safety are remarkably improved through conversion from the automatic operation and maintenance document to the operation and maintenance scene workflow, and meanwhile the operation and maintenance cost is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation and maintenance technology, and in particular to an operation and maintenance workflow generation method and device. Background Art

[0002] For the process-based orchestration of operation and maintenance scripts, traditional technical solutions require users to define operation and maintenance scenarios according to preset templates or manual sorting, build orchestration tasks step by step, and apply them through manual offline testing to deploy them in the production environment. This leads to irregular development and difficult maintenance of operation and maintenance tasks, lack of monitoring and logging, and error-prone execution of operation and maintenance tasks and poor stability. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a method and device for generating an operation and maintenance workflow, which can at least solve the problems in the prior art that operation and maintenance tasks are not standardized in development, difficult to maintain, and lack monitoring and log recording.

[0004] To achieve the above-mentioned purpose, according to one aspect of an embodiment of the present invention, there is provided a method for generating an operation and maintenance workflow, comprising: receiving an operation and maintenance document, extracting operation steps from the operation and maintenance document using a planner, and obtaining an operation step set; wherein, by fine-tuning an open source large language model, a planner, a scheduler, and an executor are obtained; using a scheduler to identify the task scenario of each operation step, and according to a task workflow generation method pre-configured for each task scenario, a multimodal operation and maintenance knowledge base is called to obtain a task workflow for each operation step; using an executor to determine the type of the task workflow, processing each task workflow according to an audit test method pre-configured for each type, and generating an operation and maintenance scenario workflow set based on the task workflow set for deployment to a production environment when the audit test result is passed.

[0005] Optionally, the use of a planner to extract operation steps from the operation and maintenance document to obtain an operation step set includes: splitting the operation and maintenance document into a sentence set, calculating the matching results and comprehensive similarity scores of each sentence in the sentence set and a preset step extraction prompt template; screening out matching sentences with a comprehensive similarity score greater than or equal to a preset similarity threshold from the sentence set to obtain a matching sentence set; obtaining a context of a preset length for each matching sentence from the matching sentence set to obtain a step context set; determining a corresponding matching sentence for each step context, and constructing a prompt statement based on the matching results of the matching sentence and the preset step extraction prompt template; extracting operation steps from each step context according to the prompt statement to obtain an operation step set for each step context.

[0006] Optionally, the process of calculating the comprehensive similarity score is: using a preset string similarity calculation method to calculate the string similarity score of each sentence and the preset step extraction prompt template; using a vector model to vectorize each sentence and the preset step extraction prompt template respectively, and calculating the vector similarity score of the two vectors; taking the sum of the vector similarity score and the string similarity score as the comprehensive similarity score of each sentence and the step extraction prompt template.

[0007] Optionally, the method further includes: extracting summary information from each step context, and using the summary information as scene description information of each operation step in the operation step set of each step context.

[0008] Optionally, the method further comprises: acquiring operation and maintenance knowledge of a historical time period, and determining a knowledge type of each operation and maintenance knowledge; wherein the knowledge type is an operation and maintenance script or a tool usage document;

[0009] Using the knowledge processing method pre-configured for each knowledge type, processing each operation and maintenance knowledge, and storing the processed operation and maintenance knowledge in the multimodal operation and maintenance knowledge base; using the scheduler to identify the task scenario of each operation step, calling the multimodal operation and maintenance knowledge base according to the task workflow generation method pre-configured for each task scenario, and obtaining the task workflow of each operation step, including: using the instruction learning method and the preset scenario intention recognition template to analyze the information of each operation step to determine whether each operation step meets the script coding conditions; if the judgment result is in compliance, calculating the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is the operation and maintenance script in the multimodal operation and maintenance knowledge base, according to the size relationship between the matching degree and the first preset matching degree threshold, using the corresponding script matching mode or script generation mode to obtain the script, and associating it to each operation step; if the judgment result is not in compliance, calculating the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is the tool usage document in the multimodal operation and maintenance knowledge base, according to the size relationship between the matching degree and the second preset matching degree threshold, using the corresponding tool task mode or preset mode to obtain the tool usage document, and associating it to each operation step.

[0010] Optionally, in the case where the knowledge type is an operation and maintenance script, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the use of the knowledge processing method pre-configured for each knowledge type to process the operation and maintenance knowledge and store the processed operation and maintenance knowledge in the multimodal operation and maintenance knowledge base includes: using a script understanding model to translate the operation and maintenance script to obtain script translation information, and storing the script translation information in the full-text retrieval database; and using a vector model to vectorize the script translation information to obtain a script translation vector, and storing the script translation vector in the vector database.

[0011] Optionally, the calculating of the matching degree between each operation step and each operation and maintenance knowledge whose knowledge type is an operation and maintenance script in the multimodal operation and maintenance knowledge base includes: calculating a first matching degree between each operation step and script translation information in a full-text retrieval database, and a second matching degree between each operation step and a script translation vector in a vector database; and integrating the first matching degree and the second matching degree according to the correspondence between the script translation information and the script translation vector to obtain a third matching degree between each operation step and each operation and maintenance knowledge.

[0012] Optionally, according to the size relationship between the matching degree and the first preset matching degree threshold, a corresponding script matching mode or script generation mode is used to obtain the script, and the script is associated with each operation step, including: determining the maximum matching degree, comparing the size relationship between the maximum matching degree and the first preset matching degree threshold; if the comparison result is greater than or equal to, then through the script matching mode, the target operation and maintenance script corresponding to the target operation and maintenance knowledge with the maximum matching degree is associated with each operation step; if the comparison result is less than, then through the script generation mode, the fine-tuning training template of the scheduler is used to construct a script writing prompt instruction, the script writing prompt instruction is input into the scheduler to write the operation and maintenance script code, and the written operation and maintenance script is associated with each operation step.

[0013] Optionally, the executor is used to determine the type of task workflow, and each task workflow is processed according to the audit test method pre-configured for each type, including: determining the type of mounted object in the task workflow, and when the mounted object is an operation and maintenance script, building a minimum executable image of the operation and maintenance script through container technology, sending the operation and maintenance script to a virtual environment for testing, obtaining the task execution log and sending it to a manual audit node, so that the manual audit node analyzes the test results according to the task execution log, and modifies the operation and maintenance script according to the test results until the test passes.

[0014] Optionally, in the case where the knowledge type is a tool usage document, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the use of the knowledge processing method pre-configured for each knowledge type to process each operation and maintenance knowledge, and storing the processed operation and maintenance knowledge in the multimodal operation and maintenance knowledge base includes: extracting text content from the tool usage document, using a recursive segmentation method to divide the text content into blocks, obtaining block content and storing it in the full-text retrieval database; and vectorizing each block content, and storing the block content vector in the vector database.

[0015] Optionally, the calculation of the matching degree between each operation step and each operation and maintenance knowledge whose knowledge type is tool usage document in the multimodal operation and maintenance knowledge base includes: calculating the fourth matching degree between each operation step and the block content in the full-text retrieval database, and the fifth matching degree between each operation step and the block content vector in the vector database; integrating the fourth matching degree and the fifth matching degree according to the correspondence between the block content and the block content vector to obtain the sixth matching degree between each operation step and each block content.

[0016] Optionally, according to the size relationship between the matching degree and the second preset matching degree threshold, the corresponding tool task mode or preset mode is used to obtain the tool usage document, and associate it with each operation step, including: determining the maximum matching degree, comparing the size relationship between the maximum matching degree and the second preset matching degree threshold; if the comparison result is greater than or equal to, then through the tool task mode, the target block content corresponding to the maximum matching degree is input into the language model, so that the language model processes the target block content, obtains tool usage information, and associates the tool usage information with each operation step; if the comparison result is less than, no tool usage information is generated.

[0017] Optionally, the multimodal operation and maintenance knowledge base also includes file storage, and the method also includes: when the tool usage document includes pictures, extracting the pictures from the tool usage document, determining the block content where the pictures are located, calculating the hash value of the text content in the block content, and storing the hash value as the identifier of the picture in the file storage; associating the tool usage information with each of the operation steps, also includes: calculating the target hash value of the text content in the target block content, and when the target hash value exists in the file storage, extracting the target picture corresponding to the target hash value from the file storage; generating a tool usage document based on the tool usage information and the target picture, and associating the generated tool usage document with each of the operation steps.

[0018] Optionally, the usage executor determines the type of the task workflow, and processes each task workflow according to the audit test method pre-configured for each type, including: determining the type of the mounted object in the task workflow, and when the mounted object is a tool usage document, sending the tool usage document to a manual review node for review and modification; the method also includes: determining a label for the target block content, and using the label as the label for the generated tool usage document.

[0019] To achieve the above-mentioned purpose, according to another aspect of an embodiment of the present invention, there is provided an operation and maintenance workflow generation device, comprising: a step extraction module, used to receive an operation and maintenance document, and use a planner to extract operation steps from the operation and maintenance document to obtain an operation step set; wherein, by fine-tuning the open source large language model, a planner, a scheduler and an executor are obtained; a task workflow generation module, used to use a scheduler to identify the task scenario of each operation step, and according to the task workflow generation method pre-configured for each task scenario, call a multimodal operation and maintenance knowledge base to obtain the task workflow of each operation step; a scenario workflow generation module, used to use an executor to determine the type of task workflow, and process each task workflow according to the audit test method pre-configured for each type. When the audit test result is passed, an operation and maintenance scenario workflow set is generated based on the task workflow set to be deployed to the production environment.

[0020] Optionally, the step extraction module is used to: split the operation and maintenance document into a sentence set, calculate the matching results and comprehensive similarity scores of each sentence in the sentence set and a preset step extraction prompt template; screen out matching sentences whose comprehensive similarity scores are greater than or equal to a preset similarity threshold from the sentence set to obtain a matching sentence set; obtain the context of a preset length of each matching sentence from the matching sentence set to obtain a step context set; determine the corresponding matching sentence for each step context, and construct a prompt statement based on the matching results of the matching sentence and the preset step extraction prompt template; extract operation steps from each step context according to the prompt statement to obtain an operation step set for each step context.

[0021] Optionally, the process of calculating the comprehensive similarity score is: using a preset string similarity calculation method to calculate the string similarity score of each sentence and the preset step extraction prompt template; using a vector model to vectorize each sentence and the preset step extraction prompt template respectively, and calculating the vector similarity score of the two vectors; taking the sum of the vector similarity score and the string similarity score as the comprehensive similarity score of each sentence and the step extraction prompt template.

[0022] Optionally, the step extraction module is further used to: extract summary information from each step context, and use the summary information as scene description information of each operation step in the operation step set of each step context.

[0023] Optionally, the device also includes a knowledge base construction module, which is used to: obtain operation and maintenance knowledge of historical time periods, determine the knowledge type of each operation and maintenance knowledge; wherein the knowledge type is an operation and maintenance script or a tool usage document; use the knowledge processing method pre-configured for each knowledge type to process each operation and maintenance knowledge, and store the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base; the task workflow generation module is used to: use the instruction learning method and the preset scene intention recognition template to analyze the information of each operation step to determine whether each operation step meets the script coding conditions; if the judgment result is in compliance, calculate the The matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is an operation and maintenance script in the multimodal operation and maintenance knowledge base is calculated, and according to the size relationship between the matching degree and the first preset matching degree threshold, the corresponding script matching mode or script generation mode is used to obtain the script, and the script is associated with each operation step; when the judgment result is that it does not meet the requirements, the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is a tool usage document in the multimodal operation and maintenance knowledge base is calculated, and according to the size relationship between the matching degree and the second preset matching degree threshold, the tool usage document is obtained by adopting the corresponding tool task mode or preset mode, and the script is associated with each operation step.

[0024] Optionally, when the knowledge type is an operation and maintenance script, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the task workflow generation module is used to: use a script understanding model to translate the operation and maintenance script to obtain script translation information, and store the script translation information in the full-text retrieval database; and use a vector model to vectorize the script translation information to obtain a script translation vector, and store the script translation vector in the vector database.

[0025] Optionally, the task workflow generation module is used to: calculate a first matching degree between each operation step and the script translation information in the full-text retrieval database, and a second matching degree between each operation step and the script translation vector in the vector database; integrate the first matching degree and the second matching degree according to the correspondence between the script translation information and the script translation vector to obtain a third matching degree between each operation step and each operation and maintenance knowledge.

[0026] Optionally, the task workflow generation module is used to: determine the maximum matching degree, compare the maximum matching degree with the first preset matching degree threshold; if the comparison result is greater than or equal to, then through the script matching mode, associate the target operation and maintenance script corresponding to the target operation and maintenance knowledge with the maximum matching degree with each operation step; if the comparison result is less than, then through the script generation mode, use the scheduler's fine-tuning training template to construct a script writing prompt instruction, input the script writing prompt instruction into the scheduler to write the operation and maintenance script code, and associate the written operation and maintenance script with each operation step.

[0027] Optionally, the scenario workflow generation module is used to: determine the type of mounted object in the task workflow, and when the mounted object is an operation and maintenance script, build a minimum executable image of the operation and maintenance script through container technology, send the operation and maintenance script to the virtual environment for testing, obtain the task execution log and send it to the manual review node, so that the manual review node analyzes the test results according to the task execution log, and modifies the operation and maintenance script according to the test results until the test passes.

[0028] Optionally, when the knowledge type is a tool usage document, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the task workflow generation module is used to: extract text content from the tool usage document, divide the text content into blocks using a recursive segmentation method, obtain the block content and store it in the full-text retrieval database; and vectorize each block content and store the block content vector in the vector database.

[0029] Optionally, the task workflow generation module is used to: calculate the fourth matching degree between each operation step and the block content in the full-text retrieval database, and the fifth matching degree between each operation step and the block content vector in the vector database; integrate the fourth matching degree and the fifth matching degree according to the correspondence between the block content and the block content vector to obtain the sixth matching degree between each operation step and each block content.

[0030] Optionally, the task workflow generation module is used to: determine the maximum matching degree, compare the maximum matching degree with a second preset matching degree threshold; if the comparison result is greater than or equal to, then through the tool task mode, input the target block content corresponding to the maximum matching degree into the language model, so that the language model processes the target block content, obtains tool usage information, and associates the tool usage information with each operation step; if the comparison result is less than, the tool usage information is not generated.

[0031] Optionally, the multimodal operation and maintenance knowledge base also includes file storage, and the knowledge base construction module is also used to: when the tool usage document includes pictures, extract the pictures from the tool usage document, determine the block content where the pictures are located, calculate the hash value of the text content in the block content, and store the hash value as the identifier of the picture in the file storage; the task workflow generation module is used to: calculate the target hash value of the text content in the target block content, and when the target hash value exists in the file storage, extract the target picture corresponding to the target hash value from the file storage; generate the tool usage document based on the tool usage information and the target picture, and associate the generated tool usage document with each of the operation steps.

[0032] Optionally, the scenario workflow generation module is used to: determine the type of mounted object in the task workflow, and when the mounted object is a tool usage document, send the tool usage document to a manual review node for review and modification; the device also includes: determining a label for the target block content, and using the label as a label for the generated tool usage document.

[0033] To achieve the above objective, according to another aspect of an embodiment of the present invention, an electronic device for generating an operation and maintenance workflow is provided.

[0034] The electronic device of an embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned operation and maintenance workflow generation methods.

[0035] To achieve the above objective, according to another aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored, and when the program is executed by a processor, any of the above-mentioned operation and maintenance workflow generation methods is implemented.

[0036] To achieve the above object, according to another aspect of an embodiment of the present invention, a computer program product is provided. A computer program product of an embodiment of the present invention includes a computer program, and when the program is executed by a processor, the operation and maintenance workflow generation method provided by the embodiment of the present invention is implemented.

[0037] According to the solution provided by the present invention, one embodiment of the above invention has the following advantages or beneficial effects: the planner, scheduler and executor obtained by fine-tuning the open source large language model can automatically extract operation steps from the operation and maintenance documents and generate task workflows. This process significantly improves the speed of operation and maintenance task development and reduces the need for manual intervention, thereby reducing time costs. At the same time, with the help of automation tools, the consistency and standardization of the operation and maintenance task workflows are ensured, avoiding inconsistencies and errors that may occur in the manual writing process. Combined with a multimodal operation and maintenance knowledge base, it is possible to more accurately match and generate task workflows suitable for different scenarios. In addition, only task workflows that have been reviewed and tested can be deployed to the production environment, further enhancing the security and stability of operation and maintenance tasks.

[0038] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0040] Figure 1 This is a main flow diagram of a method for generating an operation and maintenance workflow according to an embodiment of the present invention;

[0041] Figure 2 is a flowchart of an optional operation and maintenance workflow generation method according to an embodiment of the present invention;

[0042] Figure 3 is a flowchart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention;

[0043] Figure 4 is a schematic diagram of a process for constructing a multimodal operation and maintenance knowledge base according to an embodiment of the present invention;

[0044] Figure 5 is a flowchart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention;

[0045] Figure 6 is a flowchart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention;

[0046] Figure 7 is a flow chart of scenario workflow generation according to an embodiment of the present invention;

[0047] Figure 8 This is a schematic diagram of main modules of an operation and maintenance workflow generation device according to an embodiment of the present invention;

[0048] Fig. 9is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0049] Fig.10 It is a schematic diagram of the structure of a computer system of a mobile device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0051] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0052] In the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The acquisition, transmission, storage, use, and processing of data in the technical solution of the present invention are in compliance with the relevant provisions of national laws and regulations, and are used for legal and reasonable purposes, and are not shared, disclosed, or sold outside of these legal uses, and are subject to supervision and management by regulatory authorities.

[0053] For user information, necessary measures should be taken to prevent illegal access to such personal information data, ensure that persons who have access to personal information data comply with relevant laws and regulations, and ensure the security of user personal information. Once such user personal information data is no longer needed, risks should be minimized by limiting or even prohibiting data collection and / or deleting data. When applicable, including in certain relevant applications, protect user privacy by de-identifying data, such as by removing specific identifiers (e.g., date of birth, etc.), controlling the amount or specificity of stored data (e.g., collecting location data at the city level rather than at the specific address level), controlling how data is stored, and / or other methods of de-identification.

[0054] The current process of operation and maintenance script process orchestration based on Activiti technology includes the following steps: 1) The operation and maintenance personnel sort out the operation and maintenance scenarios and decompose the complex operation and maintenance scenarios with multiple steps into sub-steps that can be executed separately; 2) Write the corresponding operation and maintenance scripts according to the requirements of the sub-steps; 3) Freely arrange the sub-steps to form a complex operation and maintenance process according to the requirements of the operation and maintenance scenario, and solidify and save them to provide one-click execution capabilities; 4) The operation and maintenance personnel configure the execution resources and click to execute, and manage the flow of the process through Activiti; 5) According to the configured resources, randomly select the matching available Agent from the Agent connection pool, and then send the json string to the message queue; 6) After the corresponding self-developed Agent listens to the producer message, it obtains the json string and parses it, and then calls the underlying method to execute the command on the corresponding resource; 7) The execution result is passed to the operation platform through the message queue for storage and display.

[0055] The shortcomings of existing technologies mainly include the complexity of operation and maintenance scenario sorting, the consistency and maintainability of script writing, and the lack of monitoring and logging of agents. The details are as follows:

[0056] 1. Complexity of O&M scenario sorting: O&M personnel are required to manually sort out O&M scenarios, decompose process steps, and write O&M scripts. This results in high labor costs for task construction and low orchestration efficiency.

[0057] 2. Script writing consistency and maintainability: Manually writing operation and maintenance scripts can easily lead to non-standard script writing, poor environment compatibility, and high maintenance costs;

[0058] 3. Agent lacks monitoring and logging: Lack of detailed monitoring and logging makes it difficult to locate and solve problems.

[0059] 4. Robustness and fault tolerance of Agent: For the constructed script tasks, there is a lack of unified testing and auditing functions, which makes the execution of script tasks prone to errors and the production tasks unstable.

[0060] See also Figure 1 , which shows a main flow chart of a method for generating an operation and maintenance workflow provided by an embodiment of the present invention, including the following steps:

[0061] S101: receiving an operation and maintenance document, and extracting operation steps from the operation and maintenance document using a planner to obtain an operation step set; wherein a planner, a scheduler, and an executor are obtained by fine-tuning an open source large language model;

[0062] S102: using the scheduler to identify the task scenario of each operation step, and calling the multimodal operation and maintenance knowledge base according to the task workflow generation method pre-configured for each task scenario to obtain the task workflow of each operation step;

[0063] S103: Use the executor to determine the type of task workflow, process each task workflow according to the audit test method pre-configured for each type, and if the audit test result is passed, generate an operation and maintenance scenario workflow set based on the task workflow set to deploy to the production environment.

[0064] In the above implementation, the present invention uses artificial intelligence technology to automatically generate operation and maintenance scenario workflows and task test analysis by building a multimodal operation and maintenance knowledge base, converting operation and maintenance documents into operation and maintenance scenario workflows to form scenario solutions. For the process of building a multimodal operation and maintenance knowledge base, see the subsequent Figure 3 and Figure 4 The description shown is not repeated here.

[0065] For step S101, the operation and maintenance personnel upload the operation and maintenance documents (such as emergency plans, troubleshooting experience, other operation and maintenance operations, etc.) to the system. After receiving the operation and maintenance documents, the system removes the noise data such as duplicate directories and appendices, such as reading the document content and converting it into plain text format; identifying and removing the directory and appendix; dividing the document content into paragraphs or chapters, and removing duplicate content. This improves the quality and readability of the document.

[0066] This solution uses the open source large language model LLM (full name Large Language Model) as the base model, and uses SFT (Supervised Fine-Tuning) technology to train the model to obtain the planner, scheduler and executor. The planner extracts the operation steps or processes in the operation and maintenance documents to form an operation step set StepSetj=<stepj1,stepj2,…,stepjm> In particular, the planner generates operation schedules based on instruction learning methods. Figure 2 As shown in the figure, the planner solves the existing problem of "relying on operation and maintenance personnel to manually sort out the steps of the operation and maintenance scenario decomposition process".

[0067] Among them, LLM is a deep learning model based on the Transformer architecture, which is characterized by a very high number of parameters (usually more than one billion to hundreds of billions), trained with large-scale text data, and able to learn and simulate the complex structure and semantics of human language, showing excellent versatility and flexibility. This type of model can achieve excellent performance in a wide range of natural language processing tasks, especially in zero-sample learning and cross-task applications.

[0068] SFT is a model that uses supervised learning to fine-tune a pre-trained language model. The goal of SFT is to make pre-trained language models (such as GPT-3, BERT, etc.) better at completing specific tasks. It works by using a pre-trained language model as a basis and then using supervised learning to adjust the model's parameters. Supervised learning means using labeled data to train a model. In the SFT process, the label is usually the correct answer to the target task.

[0069] For step S102, after obtaining the set of operation steps through the planner, the scheduler identifies the task scenarios of the operation steps stepjm in each StepSetj, such as "script matching", "script generation", "tool operation", and "other manual operation". Different task scenarios generate task workflows in different ways. For details, see the following Figure 5 and Figure 6 Here is a brief overview:

[0070] 1. First, determine whether step stepjm is suitable for scripting. Specifically, use the instruction learning method to determine whether step stepjm can be scripted (shell, python), such as analyzing the step content to understand the specific meaning and purpose of step stepjm. Identify the encodable elements to determine which parts of the step can be converted into script code. Determine the feasibility of coding to determine whether it can be implemented using shell or Python coding based on the step content.

[0071] P(yes,no|stepjm)=llm(stepjm,intent_prompt) Where stepjm is step m in step set j, j,m are positive integers, and m<=j. P represents the probability of predicting the result as "yes" or "no" given stepjm. Intent_prompt refers to the preset scenario intention recognition prompt, which is used to identify each step task scenario, such as whether the step is suitable for scripting. Among them, Prompt: an interaction method based on natural language processing, which realizes communication between users and machines through machine parsing of natural language. The main implementation method of Prompt is to convert natural language into machine-recognizable instructions by establishing a corresponding corpus and semantic parsing model.

[0072] 2. If it matches, it is identified as using a script to complete the operation and maintenance task. Use text matching to determine the matching degree between the step stepjm and the corresponding knowledge type object in the multimodal operation and maintenance knowledge base. If it matches, it is recorded as "script matching" and the matching operation and maintenance script in the multimodal operation and maintenance knowledge base is associated with the step; otherwise, it is recorded as "script generation". According to the SFT training template of the scheduler, a script writing prompt is constructed, and the script writing prompt is input into the scheduler model so that the model writes the script code and the script code is associated with the step;

[0073] 3. If it does not match, it is identified as the use of tools to complete the operation and maintenance task. The text matching method is used to determine the matching degree of the step stepjm with the corresponding knowledge type object in the multimodal operation and maintenance knowledge base. If it matches, it is recorded as "tool task". The text content in the multimodal operation and maintenance knowledge base is input into the language model, and the tool usage information is obtained by combining the language model's understanding ability. Otherwise, it is recorded as "other manual operation" and no processing is performed.

[0074] For step S103, after the task workflow of each operation step stepjm is obtained through the above operation, the task workflow set<TaskDag1, TaskDag,…, TaskDagj> , you can review and test. For the task workflow TestDagj, different mounted objects have different review and test methods. For example, the operation and maintenance script needs to be reviewed and tested, while the tool usage document only needs to be reviewed and modified. Repeat the above steps to get the scene workflow set<SceneDag1,SceneDag2,…, SceneDagj> After that, the status of the completed scenario workflow set is updated to published, and the published scenario workflow is deployed to the production environment, providing a one-click execution function.

[0075] The method provided in the above embodiment, by fine-tuning the open source large language model to obtain the planner, scheduler and executor, can automatically extract operation steps from the operation and maintenance documents and automatically generate task workflows, significantly improving the speed of operation and maintenance task development, greatly reducing the need for manual intervention, and reducing time costs. In addition, the consistency and standardization of the operation and maintenance task workflows are ensured through automated tools, avoiding inconsistencies and errors in manual writing. Combined with the multimodal operation and maintenance knowledge base, it can more accurately understand and generate task workflows suitable for different scenarios. Only after passing the test can it be deployed to the production environment, enhancing the security and stability of the operation and maintenance tasks.

[0076] See also Figure 2 , shows a schematic flow chart of an optional operation and maintenance workflow generation method according to an embodiment of the present invention, comprising the following steps:

[0077] S201: splitting the operation and maintenance document into a sentence set, and calculating a matching result and a comprehensive similarity score between each sentence in the sentence set and a preset step extraction prompt template;

[0078] S202: Filter out matching sentences whose comprehensive similarity scores are greater than or equal to a preset similarity threshold from the sentence set to obtain a matching sentence set;

[0079] S203: Obtaining a context of a preset length of each matching sentence from the matching sentence set to obtain a step context set;

[0080] S204: determining a corresponding matching sentence for each step context, extracting a matching result of a prompt template according to the matching sentence and a preset step, and constructing a prompt sentence;

[0081] S205: Extracting operation steps from each step context according to the prompt statement to obtain an operation step set of each step context.

[0082] This implementation is used to describe how to use a planner to extract operation steps from an operation and maintenance document to obtain an operation step set.

[0083] For step S201, assuming that the operation and maintenance document is i, firstly, the operation and maintenance document i is split into a set of sentences using punctuation marks (such as periods) as dividing lines.<sent1, sent2,... ,sentn> , n is a positive integer. Calculate the comprehensive similarity score and matching result between each sentence in the sentence set and the step extraction prompt (i.e., the step extraction prompt template).

[0084] The step extraction prompt here can be regarded as a template composed of multiple template sentences. Each template sentence includes some keywords, such as operation steps, analysis flow, and analysis process, because there may be multiple scenes in a document, and each scene has extraction steps. By designing a reasonable prompt template, this solution can flexibly adapt to operation and maintenance documents of different types and formats, enhancing the adaptability and flexibility of the system. The matching results here can include the hits of keywords in different scenarios, and then obtain matching results. The formula for calculating the comprehensive similarity score is:

[0085] SimScore<StepPrompt, Senti>=EmbedScore<StepPrompt, Senti>+StrSimScore<StepPrompt, Senti>, where 1 < i < n and n is a positive integer. Here, SimScore<StepPrompt, Senti> represents the comprehensive similarity score between sentence i and the step extraction prompt; EmbedScore <StepPrompt, Senti> represents the vector similarity score between sentence i and the step extraction prompt, which is obtained by calculating the cosine of their vectors. Here, a vector model can be used to vectorize sentence i and the step extraction prompt respectively, and then calculate the similarity of the two vectors. Among them, the vector space model (VSM) is a classic theoretical framework in the fields of information retrieval and natural language processing. It transforms text information into vector representations in a high-dimensional space to quantify the similarity and relevance between texts. In this model, both documents and queries are regarded as vectors composed of a set of feature terms, and each feature term corresponds to a weight value, which is used to represent its importance in the document or query.

[0086] StrSimScore <StepPrompt, Senti> represents the string similarity score between sentence i and the step extraction prompt, which is obtained by calculating their BM25 (Best Matching 25). BM25 is a scoring function used for information retrieval and is mainly used to evaluate the relevance between a document and a query. For example, string 1: "This is an example string", string 2: "This is another example string". First, each string is split into terms, resulting in string 1: ["This", "is", "an", "example", "string"], string 2: ["This", "is", "another", "example", "string"]. Then, for each term, its TF-IDF value is calculated. For example, to calculate the TF-IDF value of the term "example", and then the similarity score between the two strings is calculated according to the BM25 formula. The closer the result value is to 1, the more similar the two strings are.

[0087] For steps S202 and S203, when SimScore is higher than or equal to (or only consider the case of being higher than) the threshold, a set of matching sentences <Senti1, Senti2,..., Sentij> is obtained, where i and j are positive integers. For each matching sentence Sentij, from the set of matching sentences, the context of a specified length of this matching sentence is extracted to obtain a set of step contexts <StepText1, StepText2,..., StepTextj>.

[0088] Assume that the operation and maintenance document is divided into 10 sentences, and three matching sentences ABC are selected according to the score. For A, there is only one context BC. Assuming the specified length is 100 characters, 100 characters need to be extracted from BC in order as the context of A. For B, A is its previous context and C is its next context. In the case that AC are both long, these 100 characters can be extracted from AC. However, if A is shorter, such as only 40 characters, 60 characters need to be extracted from C, or if C is shorter, such as only 40 characters, 60 characters need to be extracted from A and 40 characters from C. Assuming that the sum of the lengths of AC does not meet 100, the method of filling in blanks can be adopted, or the actual length shall prevail. For C, there is only the previous context, so 100 characters need to be extracted from BA in reverse order as the context.

[0089] For steps S204 and S205, for each step context StepTextj, determine the matching result of the corresponding matching sentence and step extraction prompt, such as hitting the keyword "operation step", and design a prompt sentence prompt based on the matching result, such as designing a prompt sentence prompt including the keyword "operation step". Then, according to the prompt sentence prompt, extract the operation steps from the step context StepTextj to obtain the operation step set corresponding to the step context StepTextj, such as StepSetj=<stepj1,stepj2,…,stepjm> .

[0090] As mentioned above, the three matching sentences ABC get three step contexts, j=1,2,3, and get StepSet1=<stepj1,stepj2,…,stepjm1> ,StepSet2=<stepj1,stepj2,…,stepjm2> ,StepSet3=<stepj1,stepj2,…,stepjm3> These three sets of operation steps. Here, m1, m2, and m3 can be different, depending on the number of steps actually extracted. In addition, the summary information of the three matching sentences ABC can be extracted separately, such as the summary information of A is StepSet1=<stepj1,stepj2,…,stepjm1> Combining the above results, we can get<StepSet1,StepSet2,…,StepSetj> and its scene description information.

[0091] Assume there is a step context StepTextj: "Open the browser, enter the URL www.example.com, click the login button, and enter the user name and password". The matching results include the keyword operation steps. According to the keyword operation steps, a prompt sentence prompt is constructed: "Please list the operation steps". Then the prompt sentence prompt is used to extract the operation steps from the step context StepTextj, such as {"Open the browser", "Enter the URL www.example.com", "Click the login button", "Enter the user name and password"}. Through this process, the operation steps can be accurately extracted from the step context to form an ordered set of operation steps.

[0092] In the method provided in the above embodiment, the operation and maintenance document includes multiple sentences, each sentence may correspond to a different scenario, so the context of each scenario in the document is first extracted, and then the operation steps are extracted from the context of each scenario, which solves the problem that the prior art relies on operation and maintenance personnel to manually sort out the operation and maintenance scenario decomposition process steps, significantly reduces the degree of manual participation, and at the same time improves the overall efficiency of operation and maintenance scenario sorting, ensuring the accuracy and completeness of the extraction results. By screening matching sentences, the amount of data for subsequent processing is reduced and processing efficiency is improved; obtaining the step context helps to understand the specific meaning of the operation steps and their contextual relationship in the document; constructing prompt statements based on the matching results for step extraction ensures the coherence and consistency of the extraction process.

[0093] See also Figure 3 , shows a schematic flow chart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention, comprising the following steps:

[0094] S301: Acquire operation and maintenance knowledge in a historical period and determine the knowledge type of each operation and maintenance knowledge; wherein the knowledge type is an operation and maintenance script or a tool usage document;

[0095] S302: Using the knowledge processing method pre-configured for each knowledge type, process each operation and maintenance knowledge, and store the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base.

[0096] This implementation method is used to build a multimodal operation and maintenance knowledge base. First, collect historical stock operation and maintenance knowledge, such as all operation and maintenance knowledge from the creation of the system to the present, such as obtaining known operation and maintenance knowledge in the past three months. The knowledge types of operation and maintenance knowledge include operation and maintenance scripts and tool usage documents. This solution pre-sets different knowledge processing methods according to different knowledge types.

[0097] In the first embodiment, the knowledge type is an operation and maintenance script.

[0098] The operation and maintenance script is translated by the script translation service to obtain the Chinese interpretation of the operation and maintenance script (i.e., script translation information). In particular, the script translation service is an HTTP (HyperText TransferProtocol) service based on the script understanding model. The script understanding model uses any open source large language model as the basic model and is obtained through SFT fine-tuning, such as fine-tuning the parameters related to "script content understanding". The script translation information is then stored in a full-text retrieval database (such as elasticsearch), and the script translation information is vectorized using the vector encoding service of the vector model to obtain the script translation vector, which is then stored in a vector database (such as milvus). In particular, the vector service relies on any open source vector model.

[0099] In the second embodiment, the knowledge type is tool usage document.

[0100] When the knowledge type is a tool usage document, extract the text content part of the document, and use recursive segmentation to divide the text content into blocks to obtain the block content. When processing large amounts of text data, recursive segmentation is an effective way to divide the text into smaller, more manageable parts. For example, define the basic situation: if the text length is less than or equal to a certain threshold, no further segmentation is required. Recursive segmentation: split the text from the middle, and recursively perform the same operation on the left and right parts until all parts are less than or equal to the threshold. The block content is then stored in the full-text retrieval database (elasticsearch), and the block content is vectorized, and the vectorization results are stored in the vector database (milvus).

[0101] If the tool usage document includes pictures, it is also necessary to extract the pictures in the tool document, calculate the MD5 (Message-Digest Algorithm 5) value for the text content in the block where the picture is located, and use the MD5 value as the picture ID and store it in the file storage. Among them, MD5 is a commonly used hash function that can accept data input of any length and generate a fixed length (usually 128 bits or 16 bytes) hash value, usually expressed in the form of 32 hexadecimal digits.

[0102] See the overall process Figure 4 As shown in the figure, the operation and maintenance knowledge processing logic of different knowledge types is different, so we can consider only one of the knowledge types, or we can consider both knowledge types at the same time:

[0103] 1. When the knowledge type is an operation and maintenance script, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database. The operation and maintenance script is translated using a script understanding model to obtain script translation information, and the script translation information is stored in the full-text retrieval database; and the script translation information is vectorized using a vector model to obtain a script translation vector, and the script translation vector is stored in the vector database.

[0104] 2. When the knowledge type is tool usage documents, the multimodal operation and maintenance knowledge base includes a full-text retrieval database, a vector database, and file storage. Extract text content from the tool usage document, use recursive segmentation to divide the text content into blocks, obtain the block content and store it in the full-text retrieval database; vectorize each block content, and store the block content vector in the vector database; when the tool usage document includes pictures, extract the picture from the tool usage document, determine the block content where the picture is located, calculate the hash value of the text content in the block content, and use the hash value as the image identifier to store it in the file storage.

[0105] The method provided in the above embodiment can efficiently process and store operation and maintenance knowledge by clearly distinguishing the knowledge types of operation and maintenance knowledge and configuring the knowledge processing method for each knowledge type, ensuring the structuring and standardization of knowledge, and facilitating subsequent query and use. Storing the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base can improve the reusability of knowledge, reduce duplication of work, and improve work efficiency. The clear division of knowledge types and structured storage methods help to improve the efficiency of knowledge retrieval, allowing operation and maintenance personnel to quickly find the required operation and maintenance scripts or tool usage documents, and also facilitate personnel with lower professional capabilities to quickly familiarize themselves with the process, reduce learning costs, reduce risks caused by human errors, and improve the reliability and safety of operation and maintenance work.

[0106] See also Figure 5 , shows a schematic flow chart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention, comprising the following steps:

[0107] S501: using the instruction learning method and the preset scene intention recognition template, analyzing the information of each operation step to determine whether each operation step meets the script coding conditions;

[0108] S502: When the judgment result is consistent, calculating a first matching degree between each operation step and the script translation information in the full-text search database, and a second matching degree between each operation step and the script translation vector in the vector database;

[0109] S503: According to the correspondence between the script translation information and the script translation vector, the first matching degree and the second matching degree are integrated to obtain a third matching degree between each operation step and each operation and maintenance knowledge;

[0110] S504: Determine a maximum third matching degree, and compare the maximum third matching degree with a first preset matching degree threshold;

[0111] S505: If the comparison result is greater than or equal to, then through the script matching mode, the target operation and maintenance script corresponding to the target operation and maintenance knowledge with the maximum third matching degree is associated with each of the operation steps;

[0112] S506: If the comparison result is less than, then through the script generation mode, use the fine-tuning training template of the scheduler to construct a script writing prompt instruction, input the script writing prompt instruction into the scheduler to write the operation and maintenance script code, and associate the written operation and maintenance script with each of the operation steps;

[0113] S507: Determine the type of the mounted object in the task workflow. If the mounted object is an operation and maintenance script, build a minimum executable image of the operation and maintenance script through container technology.

[0114] S508: Send the operation and maintenance script to the virtual environment for testing, obtain the task execution log and send it to the manual review node, so that the manual review node analyzes the test results according to the task execution log, and modifies the operation and maintenance script according to the test results until the test passes.

[0115] In the above implementation, steps S501 to S506 are used to describe the script matching process in the case of script coding. A text matching method (vector matching + full-text search) is used to determine whether step stepjm matches the operation and maintenance knowledge in the multimodal operation and maintenance knowledge base. If so, it is recorded as "script matching" and the matching script in the multimodal operation and maintenance knowledge base is associated with the step.

[0116] The calculation matching process here includes: calculating the first matching degree between the operation step stepjm and the script translation information, vectorizing the operation step stepjm using a vector model, and calculating the second matching degree between the vector and the script translation vector. In addition, the operation step stepjm and the scene description information can be integrated into a document, the first matching degree between the document and the script translation information is calculated, the document is vectorized using a vector model, and the second matching degree between the vector and the script translation vector is calculated. This solution preferably only considers the operation step stepjm.

[0117] According to the correspondence between the script translation information and the script translation vector, the first matching degree and the second matching degree are integrated, such as summing, weighted summing, or averaging, to obtain the third matching degree between the operation step stepjm and each operation and maintenance knowledge. Since the number of operation and maintenance knowledge, that is, the number of stock operation and maintenance scripts, may be multiple, it is necessary to determine the maximum third matching degree and compare it with the first preset matching degree threshold. If the maximum third matching degree is greater than or equal to the first preset matching degree threshold, it is represented as a script match, and its corresponding target operation and maintenance script is associated with the operation step stepjm.

[0118] If it is less than, it means that all matching relationships do not meet the requirements and a new script needs to be generated, which is recorded as "script generation". According to the SFT training template of the scheduler, a script writing prompt is constructed, and the script writing prompt is input into the scheduler model so that the model writes the script code and associates the script code with the operation step stepjm. This method can generate standardized operation and maintenance scripts, unify the script development specifications, and improve the maintainability and consistency of the scripts.

[0119] For steps S507 and S508, for the generated task workflow TestDagj, when step stepjm mounts the operation and maintenance script, the operation and maintenance personnel will review and modify the mounted operation and maintenance script. Specifically, through K8S container technology or other containerization technologies, a minimum executable image of the operation and maintenance script is built, and it is sent to the virtual environment for testing through microservice technology, thereby improving the efficiency and accuracy of script testing. The task execution log is sent back to the operation and maintenance personnel, who will analyze the test results. The purpose here is to test the execution of the operation and maintenance script to improve the executableness of the script so that it can be put into production. The method used is virtualization technology to establish a container environment to test the execution of the script. At the same time, it solves the existing problem of "lack of detailed monitoring and logging, making problem location and resolution difficult."

[0120] Collecting task execution logs online can comprehensively monitor script execution, identify problems in a timely manner and make adjustments, thus improving the completeness and effectiveness of task monitoring. By reviewing and modifying scripts, the quality of scripts is ensured, and the effectiveness and stability of task execution are improved. Operation and maintenance personnel modify task scripts according to test results until the test passes. After completion, the task workflow is stored in the scenario orchestration knowledge base to solve the existing problem of "lack of unified testing and review functions for constructed script tasks, resulting in error-prone execution of script tasks and unstable production tasks". After the operation and maintenance personnel review the modified operation and maintenance scripts, they can be transferred to the multimodal operation and maintenance knowledge base for analysis and storage to expand the operation and maintenance scripts in the multimodal operation and maintenance knowledge base.

[0121] The method provided in the above embodiment, through an intelligent script matching and generation mechanism, combines virtualization technology to perform script testing, which not only improves the standardization level and generation efficiency of operation and maintenance scripts, but also enhances the testability of scripts and the effectiveness and stability of task execution, and promotes the accumulation and inheritance of operation and maintenance scripts in a multimodal operation and maintenance knowledge base.

[0122] See also Figure 6 , shows a schematic flow chart of another optional operation and maintenance workflow generation method according to an embodiment of the present invention, comprising the following steps:

[0123] S601: using the instruction learning method and the preset scene intention recognition template, analyzing the information of each operation step to determine whether each operation step meets the script coding conditions;

[0124] S602: When the result of the judgment is that the operation steps do not match, a fourth matching degree between each operation step and the block content in the full-text search database and a fifth matching degree between each operation step and the block content vector in the vector database are calculated;

[0125] S603: Integrate the fourth matching degree and the fifth matching degree according to the correspondence between the block content and the block content vector to obtain a sixth matching degree between each operation step and each block content;

[0126] S604: Determine the maximum sixth matching degree, and compare the maximum sixth matching degree with a second preset matching degree threshold;

[0127] S605: If the comparison result is greater than or equal to, then through the tool task mode, the target block content corresponding to the sixth maximum matching degree is input into the language model, so that the language model processes the target block content, obtains tool usage information, and associates the tool usage information with each of the operation steps;

[0128] S606: If the comparison result is less than, no tool usage information is generated;

[0129] S607: Determine the type of the mounted object in the task workflow, and if the mounted object is a tool usage document, send the tool usage document to a manual review node for review and modification.

[0130] In the above implementation, for steps S601 to S606, this implementation is used to describe the matching process of tool usage documents that do not conform to script coding. Use text matching (vector matching + full-text search) to determine whether step stepjm matches the operation and maintenance knowledge in the multimodal operation and maintenance knowledge base. If it matches, it is recorded as "tool task". If it does not match, it is recorded as "other manual operation" and is not processed.

[0131] The calculation matching process here includes: calculating the fourth matching degree between the operation step stepjm and the block content, vectorizing the operation step stepjm using a vector model, and calculating the fifth matching degree between the vector and the block content vector. In addition, the operation step stepjm and the scene description information can be integrated into a document, the fourth matching degree between the document and the block content is calculated, the document is vectorized using a vector model, and the fifth matching degree between the vector and the block content vector is calculated. This solution preferably only considers the operation step stepjm.

[0132] According to the correspondence between the block content and the block content vector, the fourth matching degree and the fifth matching degree are integrated, such as summing, weighted summing, or averaging, to obtain the sixth matching degree of the operation step stepjm and each operation and maintenance knowledge. Since the operation and maintenance knowledge, that is, the number of existing block contents, may be multiple, it is necessary to determine the maximum sixth matching degree and compare it with the second preset matching degree threshold. If the maximum sixth matching degree is greater than or equal to the second preset matching degree threshold, it is represented as a tool task. At this time, the target block content corresponding to the maximum sixth matching degree in the multimodal operation and maintenance knowledge base needs to be input into the language model, and the tool usage instructions are obtained in combination with the language model understanding ability.

[0133] If the target block content also includes pictures, the MD5 value of the text content in the target block content can be calculated, and a query can be performed in the file storage based on the MD5 value. If the query exists, it means that the target block content also includes pictures, otherwise it means that it does not include them. Just associate the above tool usage information with the operation step stepjm. If included, it is necessary to generate a tool usage document with the tool usage information and the target picture, and associate the generated tool usage document with the operation step stepjm, so as to ensure the standardization and consistency of the tool usage document.

[0134] For step S607, for the generated task workflow TestDagj, when the step stepjm mounts the tool usage document, the operation and maintenance personnel shall review and modify the mounted tool usage document. After the review is passed, the task workflow is stored in the scenario arrangement knowledge base. This operation ensures the quality of the document and improves the effectiveness and stability of the tool usage document. When the step is not mounted, the task workflow is directly stored in the scenario arrangement knowledge base. For the tool usage document that has been reviewed and modified, it is necessary to match it with the target block content associated with it. Assuming that the label of the target block content is XX-User Manual, this label needs to be used as the label of the tool usage document.

[0135] The method provided in the above embodiment, through an intelligent tool usage information generation mechanism combined with manual review, not only improves the efficiency and quality of tool usage document generation, but also enhances the standardization and consistency of tool usage documents, and promotes the accumulation and inheritance of operation and maintenance knowledge in the multimodal operation and maintenance knowledge base.

[0136] See also Figure 7 As shown, a schematic diagram of a specific process of generating a scene workflow is shown, including the following steps:

[0137] 1. Use the instruction learning method and the preset scene intention recognition template to analyze the information of each operation step stepjm to determine whether each operation step meets the script coding conditions. If so, jump to step b, otherwise jump to step c.

[0138] 2. Use the text matching method (vector matching + full-text search) to determine whether step stepjm matches the script translation result (including script translation information and script translation vector) in the multimodal operation and maintenance knowledge base. If so, record it as "script matching", otherwise record it as "script generation";

[0139] 3. Use text matching (vector matching + full-text search) to calculate whether the steps match the tool usage documents (including block content and block content vectors) in the multimodal operation and maintenance knowledge base. If they match, they are identified as "tool tasks", otherwise they are recorded as "other manual operations".

[0140] 4. Perform the above operations on each step in the above step set to obtain the task workflow set in the test state<TaskDag1, TaskDag,…, TaskDagj> Among them, workflow generation: According to the task scenario classification results, the following generation tasks are performed for stepjm in each operation step set StepSej.

[0141] 1) When the task scenario is "script matching", the operation and maintenance script matched in the multimodal operation and maintenance knowledge base is associated with the step stepjm.

[0142] 2) When the task scenario is "script generation", build a script writing prompt according to the SFT training format of the scheduler, use the scheduler model to write the script code, and associate the script code with step stepjm.

[0143] 3) When the task scenario is "tool task", the block content in the multimodal operation and maintenance knowledge base is input into the language model, and the tool usage instructions are provided in combination with the language model's understanding ability. At the same time, if the block content includes pictures, the picture information is extracted according to the MD5 value, and the picture information is also associated with the step stepjm.

[0144] 4) When the task scenario is "other manual operation", no processing is performed.

[0145] 5. When stepjm mounts a script, review and modify its operation and maintenance script; when stepjm mounts a tool usage document, review and modify the mounted tool usage document. After the review is passed, store the task workflow in the scene arrangement knowledge base. When there is no other mounting in the step, directly store the task workflow in the scene arrangement knowledge base. Repeat the above process to obtain a scene workflow set<SceneDag1, SceneDag2,…, SceneDagj> .

[0146] The present invention is applied to the field of intelligent operation and maintenance, and aims to realize the automatic conversion from operation and maintenance experience documents to operation and maintenance scenario workflows through the collaborative work of multiple AI (Artificial Intelligence) models. The overall system includes four nodes: construction of a multimodal operation and maintenance knowledge base, generation of task workflows, review and test analysis of task workflows, and release of scenario workflows, so as to solve the complexity of operation and maintenance scenario combing in traditional technical solutions, consistency and maintainability of script writing, lack of monitoring and logging of agents, and poor robustness and fault tolerance of agents. It significantly improves the efficiency and quality of operation and maintenance work, and enhances the stability and reliability of operation and maintenance tasks.

[0147] Specifically, NLP (Natural Language Processing) technology is used to automatically extract the operation steps from the operation and maintenance documents through contextual understanding, thereby effectively solving the high cost and time-consuming problems of manually defining operation and maintenance orchestration. In addition, in-depth analysis of operation steps with the help of NLP technology can assist operation and maintenance personnel in quickly writing operation and maintenance scripts and tool usage documents, significantly reducing the time cost required for manual writing, thereby improving the overall efficiency of task development.

[0148] In order to further ensure the quality and safety of operation and maintenance tasks, the present invention also uses virtualization technology to build a minimum executable environment to support automated testing and log analysis of operation and maintenance scripts, which not only makes the testing and review process of operation and maintenance scripts more standardized, but also greatly improves the safety and maintainability of task operation. In summary, the present invention effectively solves the problems of high workflow definition costs, non-standard script development, and insufficient task stability caused by lack of unified testing in traditional operation and maintenance orchestration by converting the knowledge and experience in operation and maintenance documents into specific scenario solutions and adopting a multi-model collaborative strategy.

[0149] See also Figure 8, showing a schematic diagram of the main modules of an operation and maintenance workflow generation device 800 provided by an embodiment of the present invention, including: a step extraction module 801, used to receive an operation and maintenance document, and use a planner to extract operation steps from the operation and maintenance document to obtain an operation step set; wherein, by fine-tuning the open source large language model, a planner, a scheduler and an executor are obtained; a task workflow generation module 802, used to use a scheduler to identify the task scenario of each operation step, and according to the task workflow generation method pre-configured for each task scenario, call a multimodal operation and maintenance knowledge base to obtain the task workflow of each operation step; a scenario workflow generation module 803, used to use an executor to determine the type of task workflow, and process each task workflow according to the audit test method pre-configured for each type. When the audit test result is passed, an operation and maintenance scenario workflow set is generated based on the task workflow set to be deployed in the production environment.

[0150] In the implementation device of the present invention, the step extraction module 801 is used to: split the operation and maintenance document into a sentence set, calculate the matching results and comprehensive similarity scores of each sentence in the sentence set and the preset step extraction prompt template; filter out matching sentences with comprehensive similarity scores greater than or equal to the preset similarity threshold from the sentence set to obtain a matching sentence set; obtain the context of a preset length of each matching sentence from the matching sentence set to obtain a step context set; determine the corresponding matching sentence for each step context, and construct a prompt statement based on the matching results of the matching sentence and the preset step extraction prompt template; extract operation steps from each step context according to the prompt statement to obtain an operation step set for each step context.

[0151] In the implementation device of the present invention, the process of calculating the comprehensive similarity score is: using a preset string similarity calculation method to calculate the string similarity score of each sentence and the preset step extraction prompt template; using a vector model to vectorize each sentence and the preset step extraction prompt template respectively, and calculating the vector similarity score of the two vectors; and taking the sum of the vector similarity score and the string similarity score as the comprehensive similarity score of each sentence and the step extraction prompt template.

[0152] In the implementation device of the present invention, the step extraction module 801 is further used to extract summary information from the context of each step, and use the summary information as the scene description information of each operation step in the operation step set of each step context.

[0153] The implementation device of the present invention also includes a knowledge base construction module, which is used to: obtain the operation and maintenance knowledge of the historical time period, determine the knowledge type of each operation and maintenance knowledge; wherein the knowledge type is an operation and maintenance script or a tool usage document; use the knowledge processing method pre-configured for each knowledge type to process each operation and maintenance knowledge, and store the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base; the task workflow generation module 802 is used to: use the instruction learning method and the preset scene intention recognition template to analyze the information of each operation step to determine whether each operation step meets the script coding conditions; if the judgment result is in compliance, calculate The matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is the operation and maintenance script in the multimodal operation and maintenance knowledge base is calculated. According to the relationship between the matching degree and the first preset matching degree threshold, the corresponding script matching mode or script generation mode is used to obtain the script, and the script is associated with each operation step. If the judgment result is that it does not meet the requirements, the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is the tool usage document in the multimodal operation and maintenance knowledge base is calculated. According to the relationship between the matching degree and the second preset matching degree threshold, the tool usage document is obtained by using the corresponding tool task mode or preset mode, and the script is associated with each operation step.

[0154] In the implementation device of the present invention, when the knowledge type is an operation and maintenance script, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the task workflow generation module 802 is used to: use the script understanding model to translate the operation and maintenance script, obtain script translation information, and store the script translation information in the full-text retrieval database; and use the vector model to vectorize the script translation information to obtain a script translation vector, and store the script translation vector in the vector database.

[0155] In the implementation device of the present invention, the task workflow generation module 802 is used to: calculate the first matching degree between each operation step and the script translation information in the full-text retrieval database, and the second matching degree between each operation step and the script translation vector in the vector database; according to the correspondence between the script translation information and the script translation vector, integrate the first matching degree and the second matching degree to obtain the third matching degree between each operation step and each operation and maintenance knowledge.

[0156] In the implementation device of the present invention, the task workflow generation module 802 is used to: determine the maximum matching degree, compare the maximum matching degree with the first preset matching degree threshold; if the comparison result is greater than or equal to, then through the script matching mode, the target operation and maintenance script corresponding to the target operation and maintenance knowledge with the maximum matching degree is associated with each of the operation steps; if the comparison result is less than, then through the script generation mode, use the scheduler's fine-tuning training template to construct a script writing prompt instruction, input the script writing prompt instruction into the scheduler to write the operation and maintenance script code, and associate the written operation and maintenance script with each of the operation steps.

[0157] In the implementation device of the present invention, the scenario workflow generation module 803 is used to: determine the type of mounted object in the task workflow, and when the mounted object is an operation and maintenance script, build a minimum executable image of the operation and maintenance script through container technology, send the operation and maintenance script to the virtual environment for testing, obtain the task execution log and send it to the manual review node, so that the manual review node analyzes the test results according to the task execution log, and modifies the operation and maintenance script according to the test results until the test passes.

[0158] In the implementation device of the present invention, when the knowledge type is a tool usage document, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; the task workflow generation module 802 is used to: extract text content from the tool usage document, use a recursive segmentation method to divide the text content into blocks, obtain the block content and store it in the full-text retrieval database; and vectorize each block content and store the block content vector in the vector database.

[0159] In the implementation device of the present invention, the task workflow generation module 802 is used to: calculate the fourth matching degree between each operation step and the block content in the full-text retrieval database, and the fifth matching degree with the block content vector in the vector database; according to the correspondence between the block content and the block content vector, integrate the fourth matching degree and the fifth matching degree to obtain the sixth matching degree between each operation step and each block content.

[0160] In the implementation device of the present invention, the task workflow generation module 802 is used to: determine the maximum matching degree, compare the maximum matching degree with the second preset matching degree threshold; if the comparison result is greater than or equal to, then through the tool task mode, the target block content corresponding to the maximum matching degree is input into the language model, so that the language model processes the target block content, obtains tool usage information, and associates the tool usage information with each operation step; if the comparison result is less than, the tool usage information is not generated.

[0161] In the implementation device of the present invention, the multimodal operation and maintenance knowledge base also includes file storage, and the knowledge base construction module is also used to: when the tool usage document includes pictures, extract the pictures from the tool usage document, determine the block content where the pictures are located, calculate the hash value of the text content in the block content, and store the hash value as the identifier of the picture in the file storage; the task workflow generation module 802 is used to: calculate the target hash value of the text content in the target block content, and when the target hash value exists in the file storage, extract the target picture corresponding to the target hash value from the file storage; generate a tool usage document based on the tool usage information and the target picture, and associate the generated tool usage document with each of the operation steps.

[0162] In the implementation device of the present invention, the scenario workflow generation module 803 is used to: determine the type of mounting object in the task workflow, and when the mounting object is a tool usage document, send the tool usage document to the manual review node for review and modification; the device also includes: determining the label of the target block content, and using the label as the label of the generated tool usage document.

[0163] In addition, the specific implementation content of the device described in the embodiment of the present invention has been described in detail in the method described above, so the repeated content will not be described again here.

[0164] Fig. 9 An exemplary system architecture 900 to which embodiments of the present invention may be applied is shown, including terminal devices 901 , 902 , 903 , a network 904 and a server 905 (only an example).

[0165] Terminal devices 901, 902, 903 can be various electronic devices with display screens and supporting web browsing, and various communication client applications are installed. Users can use terminal devices 901, 902, 903 to interact with server 905 through network 904 to receive or send messages, etc.

[0166] The network 904 is used to provide a medium for communication links between the terminal devices 901, 902, 903 and the server 905. The network 904 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0167] The server 905 may be a server that provides various services, such as a backend management server that provides support for shopping websites browsed by users using terminal devices 501, 502, and 503 (for example only). The backend management server may analyze and process the received data such as product information query requests, and feed back the processing results (such as target push information, product information - for example only) to the terminal device. It should be noted that the method provided in the embodiment of the present invention is generally executed by the server 905, and accordingly, the device is generally set in the server 905.

[0168] It should be understood that Fig. 9 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0169] Reference below Fig.10 , which shows a schematic diagram of the structure of a computer system 1000 of a terminal device suitable for implementing an embodiment of the present invention. Fig.10 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0170] like Fig.10 As shown, the computer system 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage part 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the system 1000 are also stored. The CPU 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0171] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.

[0172] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, the above-mentioned functions defined in the system of the present invention are executed.

[0173] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0174] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0175] The modules involved in the embodiments of the present invention may be implemented in software or hardware. The modules described may also be set in a processor. For example, they may be described as: a processor including a step extraction module, a task workflow generation module, and a scenario workflow generation module. The names of these modules do not, in some cases, constitute limitations on the modules themselves. For example, the scenario workflow generation module may also be described as an "operation and maintenance scenario workflow generation module."

[0176] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device executes any of the above operation and maintenance workflow generation methods.

[0177] The computer program product of the present invention includes a computer program, and when the computer program is executed by a processor, the operation and maintenance workflow generation method in the embodiment of the present invention is implemented.

[0178] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for generating an operation and maintenance workflow, It is characterized in that include: Receive an operation and maintenance document, and use a planner to extract operation steps from the operation and maintenance document to obtain an operation step set; wherein a planner, a scheduler, and an executor are obtained by fine-tuning an open source large language model; Use the scheduler to identify the task scenario of each operation step, and call the multimodal operation and maintenance knowledge base according to the task workflow generation method pre-configured for each task scenario to obtain the task workflow of each operation step; Use the executor to determine the type of task workflow, process each task workflow according to the audit test method pre-configured for each type, and if the audit test result is passed, generate an operation and maintenance scenario workflow set based on the task workflow set for deployment to the production environment.

2. The method according to claim 1, It is characterized in that The use planner extracts operation steps from the operation and maintenance document to obtain an operation step set, including: Splitting the operation and maintenance document into a sentence set, and calculating a matching result and a comprehensive similarity score between each sentence in the sentence set and a preset step extraction prompt template; From the sentence set, select matching sentences whose comprehensive similarity scores are greater than or equal to a preset similarity threshold to obtain a matching sentence set; Obtaining a context of a preset length of each matching sentence from the matching sentence set to obtain a step context set; Determine the corresponding matching sentence for each step context, extract the matching result of the prompt template according to the matching sentence and the preset step, and construct the prompt sentence; Operation steps are extracted from each step context according to the prompt statement to obtain an operation step set of each step context.

3. The method according to claim 2, It is characterized in that The process of calculating the comprehensive similarity score is: Using a preset string similarity calculation method, calculating a string similarity score between each sentence and the preset step extraction prompt template; Using a vector model to vectorize each of the sentences and the preset step extraction prompt template, respectively, and calculate the vector similarity score of the two vectors; The sum of the vector similarity score and the string similarity score is used as the comprehensive similarity score between each sentence and the step extraction prompt template.

4. The method according to claim 2, It is characterized in that The method further comprises: Abstract information is extracted from each step context, and the abstract information is used as scene description information of each operation step in the operation step set of each step context.

5. The method according to claim 1, It is characterized in that The method further comprises: Obtain the operation and maintenance knowledge of the historical time period and determine the knowledge type of each operation and maintenance knowledge; wherein the knowledge type is an operation and maintenance script or a tool usage document; Using the knowledge processing method pre-configured for each knowledge type, processing each operation and maintenance knowledge, and storing the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base; The use of the scheduler to identify the task scenario of each operation step, calling the multimodal operation and maintenance knowledge base according to the task workflow generation method pre-configured for each task scenario, and obtaining the task workflow of each operation step includes: Using the instruction learning method and the preset scene intention recognition template, the information of each operation step is analyzed to determine whether each operation step meets the script coding conditions; If the result of the judgment is compliance, the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is an operation and maintenance script in the multimodal operation and maintenance knowledge base is calculated, and according to the relationship between the matching degree and the first preset matching degree threshold, the corresponding script matching mode or script generation mode is used to obtain the script, and the script is associated with each operation step; If the judgment result is non-compliance, the matching degree of each operation step and each operation and maintenance knowledge whose knowledge type is tool usage document in the multimodal operation and maintenance knowledge base is calculated, and according to the relationship between the matching degree and the second preset matching degree threshold, the corresponding tool task mode or preset mode is used to obtain the tool usage document and associate it with each operation step.

6. The method according to claim 5, It is characterized in that When the knowledge type is operation and maintenance script, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; The method of using the knowledge processing method pre-configured for each knowledge type to process each operation and maintenance knowledge, and storing the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base, includes: Using the script understanding model to translate the operation and maintenance script, obtaining script translation information, and storing the script translation information in a full-text retrieval database; and The script translation information is vectorized using a vector model to obtain a script translation vector, and the script translation vector is stored in a vector database.

7. The method according to claim 6, It is characterized in that The calculating the matching degree between each operation step and each operation and maintenance knowledge whose knowledge type is an operation and maintenance script in the multimodal operation and maintenance knowledge base includes: Calculating a first matching degree between each operation step and the script translation information in the full-text search database, and a second matching degree between each operation step and the script translation vector in the vector database; According to the correspondence between the script translation information and the script translation vector, the first matching degree and the second matching degree are integrated to obtain a third matching degree between each operation step and each operation and maintenance knowledge.

8. The method according to any one of claims 5 to 7, It is characterized in that The step of obtaining a script using a corresponding script matching mode or a script generation mode according to the relationship between the matching degree and the first preset matching degree threshold and associating the script with each of the operation steps includes: Determine a maximum matching degree, and compare the maximum matching degree with a first preset matching degree threshold; If the comparison result is greater than or equal to, then the target operation and maintenance script corresponding to the target operation and maintenance knowledge with the maximum matching degree is associated with each of the operation steps through the script matching mode; If the comparison result is less than, then through the script generation mode, use the scheduler's fine-tuning training template to build a script writing prompt instruction, input the script writing prompt instruction into the scheduler to write the operation and maintenance script code, and associate the written operation and maintenance script with each of the operation steps.

9. The method according to any one of claims 5 to 7, It is characterized in that The use of the executor to determine the type of task workflow and process each task workflow according to the audit test method pre-configured for each type includes: Determine the type of mounted object in the task workflow. If the mounted object is an operation and maintenance script, use container technology to build the minimum executable image of the operation and maintenance script, send the operation and maintenance script to the virtual environment for testing, obtain the task execution log and send it to the manual review node, so that the manual review node analyzes the test results based on the task execution log, and modifies the operation and maintenance script according to the test results until the test passes.

10. The method according to claim 5, It is characterized in that In the case where the knowledge type is tool usage documents, the multimodal operation and maintenance knowledge base includes a full-text retrieval database and a vector database; The method of using the knowledge processing method pre-configured for each knowledge type to process each operation and maintenance knowledge, and storing the processed operation and maintenance knowledge in a multimodal operation and maintenance knowledge base, includes: Extracting text content from the tool usage document, dividing the text content into blocks using a recursive segmentation method, obtaining the block content and storing it in a full-text retrieval database; and Each block content is vectorized and the block content vector is stored in a vector database.

11. The method according to claim 10, It is characterized in that The calculating the matching degree between each operation step and each operation and maintenance knowledge whose knowledge type is tool usage document in the multimodal operation and maintenance knowledge base includes: Calculating a fourth matching degree between each operation step and the block content in the full-text search database, and a fifth matching degree between each operation step and the block content vector in the vector database; According to the correspondence between the block content and the block content vector, the fourth matching degree and the fifth matching degree are integrated to obtain the sixth matching degree between each operation step and each block content.

12. The method according to claim 10 or 11, It is characterized in that The method of obtaining a tool usage document based on the relationship between the matching degree and the second preset matching degree threshold adopts a corresponding tool task mode or a preset mode and associates the document with each operation step, including: Determine a maximum matching degree, and compare the maximum matching degree with a second preset matching degree threshold; If the comparison result is greater than or equal to, then through the tool task mode, the target block content corresponding to the maximum match is input into the language model, so that the language model processes the target block content, obtains tool usage information, and associates the tool usage information with each operation step; If the comparison result is less than, no tool usage information is generated.

13. The method according to claim 12, It is characterized in that The multimodal operation and maintenance knowledge base also includes file storage, and the method further includes: In the case where the tool usage document includes an image, extract the image from the tool usage document, determine the block content where the image is located, calculate the hash value of the text content in the block content, and use the hash value as an identifier of the image to store in the file storage; The associating the tool usage information with each of the operation steps further includes: Calculating a target hash value of the text content in the target block content, and extracting a target image corresponding to the target hash value from the file storage if the target hash value exists in the file storage; A tool usage document is generated based on the tool usage information and the target image, and the generated tool usage document is associated with each of the operation steps.

14. The method according to claim 5, It is characterized in that The use of the executor to determine the type of task workflow and process each task workflow according to the audit test method pre-configured for each type includes: Determine the type of the mounted object in the task workflow. If the mounted object is a tool usage document, send the tool usage document to the manual review node for review and modification. The method further includes: determining a label of the target segment content, and using the label as a label of the generated tool usage document.

15. A device for generating operation and maintenance workflow, It is characterized in that include: A step extraction module is used to receive an operation and maintenance document, and use a planner to extract operation steps from the operation and maintenance document to obtain an operation step set; wherein the planner, scheduler and executor are obtained by fine-tuning the open source large language model; The task workflow generation module is used to use the scheduler to identify the task scenario of each operation step, and call the multimodal operation and maintenance knowledge base according to the task workflow generation method pre-configured for each task scenario to obtain the task workflow of each operation step; The scenario workflow generation module is used to use the executor to determine the type of task workflow, process each task workflow according to the audit test method pre-configured for each type, and generate an operation and maintenance scenario workflow set based on the task workflow set for deployment to the production environment if the audit test result is passed.

16. An electronic device, It is characterized in that include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 14.

17. A computer readable medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the method according to any one of claims 1 to 14 is implemented.

18. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 14 is implemented.

Citation Information

Cited By

  • Picture auditing method, computer equipment, readable storage medium and program product

    CN121010807A

  • Operation and maintenance process optimization method and system applied to operation and maintenance technology service

    CN121614350A