Intelligent operation and maintenance system and method based on large language model and feature detection

Abstract

The invention relates to an intelligent operation and maintenance system based on a large language model and feature detection, and the system comprises a data obtaining module which is used for obtaining the voice of operation and maintenance personnel, and the operation and maintenance data and feature data of the system; the data storage module is used for storing operation and maintenance data and constructing an operation and maintenance knowledge base; the big language model module is used for converting and processing a natural language based on a big language model; the processing module is used for cleaning and formatting the voice of the operation and maintenance personnel and the operation and maintenance data; and the output module is used for outputting voices, characters and an operation and maintenance auxiliary system. According to the intelligent operation and maintenance process constructed by the intelligent operation and maintenance system based on the large language model and feature detection, more comprehensive and easy-to-understand operation and maintenance suggestions are provided, the operation and maintenance working efficiency is improved, the system fault prediction accuracy is enhanced, manual intervention is reduced, and the problem that fault early warning and potential risk identification are lacked in the prior art is effectively solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural language processing, and particularly relates to an intelligent operation and maintenance system and method based on a large language model and feature detection. Background Art

[0002] In the context of the rapid development of current information technology, the stability and reliability of information systems are crucial for business processes. The Communication and Information Center of the Yangtze River Three Gorges Navigation Administration undertakes extremely critical navigation operations. The operation and maintenance work of its information system not only affects the smooth progress of daily operations but also directly impacts the operational efficiency and safety of the entire Three Gorges navigation. The currently used equipment and facilities have been in operation for many years, and due to their highly customized nature, with the continuous increase in new requirements and the continuous advancement of technology upgrades, complex and ever-changing problems frequently occur in the operation and maintenance work.

[0003] Traditional integrated operation and maintenance platforms mainly include monitoring systems and alarm systems. However, due to their complex user interfaces and operation processes, the user experience is poor. For the solution of operation and maintenance scenarios, operation and maintenance personnel still need to judge based on the displayed data, lacking flexibility and convenience. Operation and maintenance personnel rely on paper or electronic manuals and need to accumulate a large amount of experience and training to complete various operation and maintenance operations. Summary of the Invention

[0004] The technical problem of the present invention is to: real-time monitor the status of hardware devices through an intelligent operation and maintenance platform, and by setting reasonable thresholds and alarm rules, automatically monitor the faulty devices, time, and current status indicators, and then analyze the cause of the fault for operation and maintenance personnel based on the scope and presented characteristics of the fault impact, and provide professional technical guidance and operation and maintenance methods.

[0005] The technical solution of the present invention is an intelligent operation and maintenance system based on a large language model and feature detection, including: A data acquisition module for acquiring the voice of operation and maintenance personnel, as well as the operation and maintenance data and feature data of the system; A data storage module for storing operation and maintenance data and constructing an operation and maintenance knowledge base; A large language model module for converting and processing natural language based on a large language model; A processing module for cleaning and formatting the voice and operation and maintenance data of operation and maintenance personnel; An output module for outputting voice, text, and an operation and maintenance assistance system.

[0006] Furthermore, the feature data includes statistical data, image processing data, industry standard data, and historical experience data.

[0007] The data acquisition module includes historical operation and maintenance information and system real-time data information obtained, including the information system name, computing resources, storage resources, middleware, database, and interface conditions.

[0008] Furthermore, the data acquisition module includes fault handling methods, system alarm thresholds, system responsible persons, system health indicators, rules and regulations, operation procedures, fault problems, and fault issues.

[0009] Preferably, the data acquisition module structures the question information through a semi-supervised learning algorithm and submits it to the large language model for reply. The semi-supervised learning algorithm learns by establishing an S3VM objective function and maximizes the margin while keeping the labeled points away from the boundary.

[0010] Furthermore, the operation and maintenance knowledge base constructs a local operation and maintenance database and a vector database based on historical operation data and operation and maintenance technical documents.

[0011] In the creation of the knowledge base, different methods are used for data vectorization in the vector database for different formats of data. One-hot encoding is used for text vectorization, a convolutional neural network model is used for image vectorization to extract visual features of images, and Fourier transform is used for audio vectorization to extract audio features.

[0012] Preferably, the large language model module is based on the Ollama open-source large model and introduces deep residual connections to alleviate the vanishing gradient in the deep neural network.

[0013] Preferably, the processing module extracts operation and maintenance information and uses NLP natural language processing methods to convert unstructured data into structured knowledge, establishes relevant indexes and tags, and then uses them for data query, retrieval, data cleaning, and standardization processing.

[0014] Preferably, for data retrieval, relevant document data is vectorized and embedded into the vector database, and then the user's query statement is converted into a vectorized query to recall data with high similarity from the vector database.

[0015] Furthermore, the large language model learns common semantics from different data modalities through pictures, voices, and texts input by operation and maintenance personnel.

[0016] Preferably, the output module is used to output the operation and maintenance management suggestions of the target information system to a preset terminal or page.

[0017] An intelligent operation and maintenance method based on a large language model and feature detection, characterized by including the following steps: S1: The data acquisition module obtains the voice and text information of operation and maintenance personnel, converts it into system operation and maintenance data, and then extracts the data feature values in the operation and maintenance data; S2: Preprocess the data eigenvalue and basic data collected in step S1 by the processing module; S3: Input the data preprocessed in step S2 into the storage module; S4: Retrieve the local operation and maintenance database and the vector database, and call the large language model for natural language understanding and reasoning, and then generate a reply answer; S5: Output the reply answer through the output module to reply to the operation and maintenance personnel.

[0018] Compared with the prior art, the beneficial effects of the present invention include: 1) The operation and maintenance intelligent process constructed by the intelligent operation and maintenance system based on the large language model and feature detection of the present invention provides more comprehensive and easy-to-understand operation and maintenance suggestions, improves the operation and maintenance work efficiency, enhances the accuracy of system fault prediction, and reduces manual intervention, effectively solving the lack of fault warning and potential risk identification in the prior art.

[0019] 2) The intelligent operation and maintenance system based on the large language model and feature detection of the present invention is trained through the operation and maintenance knowledge base and historical operation and maintenance data, can understand and process specific concepts and languages in specific operation and maintenance scenarios, improves the interaction ability of the system, and enables users to communicate with the system more effectively. Description of the Drawings

[0020] The present invention will be further described below with reference to the drawings and embodiments.

[0021] Figure 1 It is the overall flowchart of the intelligent operation and maintenance system based on the large language model and feature detection of the embodiment of the present invention; Figure 2 It is the flowchart of the intelligent operation and maintenance system based on the large language model and feature detection of the embodiment of the present invention; Figure 3 It is the interface diagram of the intelligent operation and maintenance system based on the large language model and feature detection of the embodiment of the present invention. Detailed Embodiments

[0022] As Figure 1 shown, an intelligent operation and maintenance system based on a large language model and feature detection includes: A data acquisition module, used to acquire the voice of operation and maintenance personnel, as well as the operation and maintenance data and feature data of the system; A data storage module, used to store operation and maintenance data and build an operation and maintenance knowledge base; A large language model module, which converts and processes natural language based on the large language model; A processing module, used to clean and format the voice of operation and maintenance personnel and operation and maintenance data; An output module, used for voice, text, and operation and maintenance assistance system output.

[0023] Furthermore, the feature data includes statistical data, image processing data, industry standard data, and historical experience data.

[0024] The data acquisition module includes the acquired historical operation and maintenance information and system real-time data information, including the information system name, computing resources, storage resources, middleware, database, and interface conditions.

[0025] The data acquisition module includes fault handling methods, system alarm thresholds, system responsible persons, system health indicators, rules and regulations, operation procedures, fault problems, and fault issues.

[0026] The data acquisition module structures the question information through a semi-supervised learning algorithm and submits it to the large language model for response. The semi-supervised learning algorithm learns by establishing an S3VM objective function and maximizes the margin while keeping the labeled points away from the boundary.

[0027] Furthermore, the operation and maintenance knowledge base constructs a local operation and maintenance database and a vector database based on historical operation data and operation and maintenance technical documents.

[0028] In the creation of the knowledge base, different methods are used for data vectorization in the vector database for different formats of data. One-hot encoding is used for text vectorization, a convolutional neural network model is used for image vectorization to extract visual features of images, and Fourier transform is used for audio vectorization to extract audio features.

[0029] The large language model module is based on the Ollama open-source large model and introduces deep residual connections to alleviate the vanishing gradient in deep neural networks.

[0030] The processing module extracts operation and maintenance information and uses NLP natural language processing methods to convert unstructured data into structured knowledge, establishes relevant indexes and tags, and then uses them for data querying, retrieval, data cleaning, and standardization processing.

[0031] The data processing module also includes a deduplication operation to delete duplicate samples in the original data to ensure that each sample appears only once in the training set, thereby improving the diversity of data and the generalization ability of the model; at the same time, irrelevant information and error data in the text, such as non-text content, garbled characters, etc., are identified and removed through denoising to ensure the purity and effectiveness of the training data. Data processing uses word segmentation and part-of-speech tagging to split continuous text into independent lexical units and assigns a grammatical role label, such as noun, verb, adjective, to each lexical unit.

[0032] The retrieval of data vectorizes relevant document data and embeds it into the vector database, and then converts the user's query statement into a vectorized query to recall data with high similarity from the vector database.

[0033] The creation of a vector database involves the following steps: 1) Create a model: Determine the machine learning model to generate vector embeddings; 2) Information embedding: Embed information such as text, images, and audio into vectors; 3) Data vectorization: Convert data into vector representations through the model; 4) Store metadata: Store additional metadata together with the vectors to assist in searching; 5) Create indexes: Create indexes for the vectors and metadata respectively; 6) Data storage: Store the vector data, its indexes, and metadata in the database; 7) Query preparation: Conduct searches and metadata queries through an ANN neural network; 8) Data filtering: Exclude vectors that do not meet the conditions based on the metadata; 9) Vector search: Use ANN search to find similar vectors; 10) Similarity evaluation: Evaluate the similarity between vectors through methods such as cosine similarity.

[0034] The large language model learns common semantics from different data modalities through pictures, voices, and texts input by the operation and maintenance personnel.

[0035] The output module is used to output the operation and maintenance management suggestions of the target information system to a preset terminal or page.

[0036] The output module system can be intelligent operation and maintenance systems such as ZABBIX and ITM, and the terminal is devices such as computers, mobile phones, and tablets used to receive push information.

[0037] As Figure 2 shown, an intelligent operation and maintenance method based on a large language model and feature detection includes the following steps: S1: The data acquisition module acquires the voice and text information of the operation and maintenance personnel, converts it into system operation and maintenance data, and then extracts the data feature values in the operation and maintenance data; S2: Input the data feature values and basic data collected in step S1 into the processing module for preprocessing; S3: Input the data preprocessed in step S2 into the storage module; S4: Retrieve the local operation and maintenance database and the vector database, and call the large language model for natural language understanding and reasoning, and then generate a reply answer; S5: Output the reply answer through the output module to reply to the operation and maintenance personnel.

[0038] As Figure 3As shown, by applying the present invention, an intelligent operation and maintenance system based on a large language model and feature detection has issued 320 alarms since the system deployment was completed in October 2024, given 320 solutions, with a response rate of 100%. The number of accurate fault handling suggestions is 290, with an accuracy rate of 91%. At the same time, 3 subsequent faults were successfully predicted during the system deployment, and 2 fault phenomena were accurately predicted, significantly improving the operation and maintenance efficiency and quality.

[0039] The above embodiments are only the preferred technical solutions of the present invention and should not be regarded as limitations on the present invention. The protection scope of the present invention should be the technical solutions recorded in the claims, including the equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. An intelligent operation and maintenance system based on a large language model and feature detection, characterized in that: include: The data acquisition module is used to acquire the voice of the operation and maintenance personnel, and the operation and maintenance data and feature data of the system; Data storage module, used to store operation and maintenance data and build an operation and maintenance knowledge base; Large language model module, which converts and processes natural language based on the large language model; The processing module is used to clean and format the voice and operation and maintenance data of the operation and maintenance personnel; Output module, used for voice, text and operation and maintenance auxiliary system output.

2. According to claim 1, an intelligent operation and maintenance system based on a large language model and feature detection is characterized in that: The characteristic data includes statistical data, image processing data, industry standard data and historical experience data.

3. According to claim 1, an intelligent operation and maintenance system based on a large language model and feature detection is characterized in that: The data acquisition module includes the acquired historical operation and maintenance information and system real-time data information, including the information system name, computing resources, storage resources, middleware, database and interface status.

4. According to claim 3, an intelligent operation and maintenance system based on a large language model and feature detection is characterized in that: The data acquisition module structures the question information through a semi-supervised learning algorithm and submits it to the large language model for reply, including fault handling methods, system alarm thresholds, system responsible persons, system health indicators, rules and regulations, operating procedures, fault problems and fault issues.

5. According to claim 1, an intelligent operation and maintenance system based on a large language model and feature detection is characterized in that: The operation and maintenance knowledge base builds a local operation and maintenance database and a vector database based on historical operation data and operation and maintenance technical documents; the vector database data vectorization adopts different methods for data in different formats, text vectorization adopts one-hot encoding, image vectorization adopts a convolutional neural network model to extract the visual features of the image, and audio vectorization uses Fourier transform to extract audio features.

6. According to claim 1, an intelligent operation and maintenance system based on a large language model and feature detection is characterized in that: The large language model module is based on the Ollama open source large model and introduces deep residual connections to alleviate the gradient disappearance in deep neural networks.

7. The intelligent operation and maintenance system based on large language model and feature detection according to claim 1, characterized in that: The processing module extracts operation and maintenance information and uses NLP natural language processing methods to convert unstructured data into structured knowledge, establishes relevant indexes and tags, and then uses them for data query, retrieval, data cleaning and standardization processing.

8. The intelligent operation and maintenance system based on large language model and feature detection according to claim 1, characterized in that: The large language model learns common semantics from different data modalities through pictures, voices and texts input by operation and maintenance personnel.

9. The intelligent operation and maintenance system based on large language model and feature detection according to claim 1, characterized in that: The output module is used to output the operation and maintenance management suggestions of the target information system to a preset terminal or page.

10. An intelligent operation and maintenance method based on a large language model and feature detection as described in claims 1-7, characterized in that: The following steps are involved: S1: The data acquisition module acquires the voice and text information of the operation and maintenance personnel, converts it into system operation and maintenance data, and then extracts the data feature values ​​in the operation and maintenance data; S2: input the data characteristic values ​​and basic data collected in step S1 into the processing module for preprocessing; S3: input the data pre-processed in step S2 into the storage module; S4: Retrieve the local operation and maintenance database and vector database, and call the large language model for natural language understanding and reasoning to generate a reply answer; S5: The reply answer is output through the output module to reply to the operation and maintenance personnel.

Images