Intelligent operation and maintenance method, device and equipment based on large language model

Abstract

The invention belongs to the technical field of information, and particularly relates to an intelligent operation and maintenance method, device and equipment based on a large language model, and the method comprises the steps: obtaining real-time operation data of a software system; the real-time operation data comprises real-time log data and performance monitoring data; determining an operation and maintenance strategy according to the real-time operation data and a pre-trained large language model; generating an operation and maintenance instruction according to the operation and maintenance strategy; the operation and maintenance instructions are operated in sequence, and whether the system is abnormal or not is recognized after the operation and maintenance instructions are operated every time; and if the system is identified to be abnormal, rolling back a parameter value of operation and maintenance of the operation and maintenance instruction before operation and maintenance, and sending alarm information. The problems that a software system is high in operation and maintenance cost, low in efficiency, poor in adaptive capacity and weak in automatic operation and maintenance capacity are solved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an intelligent operation and maintenance method, device and equipment based on a large language model. Background Art

[0002] Large-scale software systems usually serve a large user base and need to ensure 24 / 7 uninterrupted service, with high requirements for availability and reliability. However, due to their large scale and complex operating logic, these software systems are extremely prone to failures and difficult to detect, accurately locate and diagnose. Once a failure occurs, analyzing and debugging system errors will make it more difficult to locate and repair the fault.

[0003] Currently, there are two main solutions for the operation and maintenance of software systems: the first solution is to use monitoring tools or automated operation and maintenance scripts to assist operation and maintenance personnel in locating system faults. This method requires the pre-writing of complex monitoring scripts and requires operation and maintenance personnel to identify the cause of the fault, resulting in low fault location efficiency. The second solution is to locate system faults through pre-trained machine learning algorithms. This method requires the training of specific machine learning models for specific tasks, has the problem of weak cross-platform capabilities, and requires complex preprocessing and feature engineering to extract training data.

[0004] In addition, during the operation and maintenance process, the existing technology still requires operation and maintenance personnel to provide solutions based on the cause of the failure, and requires operation and maintenance personnel to manually troubleshoot the software system failure. This places high demands on the experience level of the operation and maintenance personnel, further increasing the manpower cost of software system operation and maintenance. Summary of the invention

[0005] The present invention provides an intelligent operation and maintenance method, device and equipment based on a large language model, which are used to solve the problems of high operation and maintenance cost, low efficiency, poor adaptability and weak automated operation and maintenance capability of software systems.

[0006] In order to solve the above technical problems, the first aspect of this document provides an intelligent operation and maintenance method based on a large language model, the method comprising:

[0007] Acquire real-time operation data of the software system; the real-time operation data includes real-time log data and performance monitoring data;

[0008] Determine an operation and maintenance strategy based on the real-time operation data and the pre-trained large language model;

[0009] Generate operation and maintenance instructions according to the operation and maintenance strategy;

[0010] Execute the operation and maintenance instructions in sequence, and identify whether the system is abnormal after each execution of the operation and maintenance instructions;

[0011] If a system abnormality is identified, the parameter values ​​of the operation and maintenance instructions will be rolled back to before the operation and maintenance and an alarm message will be sent;

[0012] Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

[0013] Furthermore, before the real-time operation data is input into the pre-trained large language model, the method further includes:

[0014] Extracting key content data from the real-time log data;

[0015] dividing the key content data into a plurality of overlapping content blocks;

[0016] Inputting each overlapping content block into the large language model respectively to obtain a plurality of refined content blocks;

[0017] Inputting the plurality of refined content blocks into the large language model to obtain refined key content data;

[0018] The real-time log data is adjusted to refined key content data.

[0019] Furthermore, before the real-time operation data is input into the pre-trained large language model, the method further includes:

[0020] Acquire data with structured requirements in the real-time operation data and format requirements corresponding to the data with structured requirements;

[0021] According to the format requirement, adjusting the language token probability of the large language model;

[0022] Inputting the data with structuring requirements into the large language model to obtain structured data;

[0023] The data with structuring requirements is adjusted to the structured data.

[0024] Further, adjusting the language token probability of the large language model according to the format requirement includes:

[0025] The probability of language tokens that do not conform to the format requirements is set to zero.

[0026] Furthermore, an operation and maintenance strategy is determined based on the real-time operation data and the pre-trained large language model, including:

[0027] Obtaining a context example group from a context prompt library; the context example group includes one or more context examples, the context examples are generated based on historical operation and maintenance data of the system, and the context examples include system failure causes and solutions;

[0028] The real-time operation data, the context example group and the preset thought chain prompt are input into the large language model to obtain the operation and maintenance strategy; the preset thought chain prompt is used to guide the analysis of the real-time operation data.

[0029] Further, a context example group is obtained from the context prompt library, including:

[0030] Obtaining the success rates of all context example groups in the context prompt library;

[0031] Determining the sampling probability of all context example groups in the context prompt library according to the success rate;

[0032] According to the sampling probability, a context example group is obtained from a context prompt library.

[0033] Furthermore, the method further comprises:

[0034] Based on the operation and maintenance results, update the success rate of the context example group used this time.

[0035] Furthermore, the preset thought chain prompt is obtained according to the following method:

[0036] Obtaining initial thought chain prompts, a preset context example group, and historical operation data of the system;

[0037] Inputting the initial thought chain prompt, the preset context example group and the historical operation data of the system into the large language model to obtain an output result;

[0038] Determine whether the output result meets a preset condition;

[0039] If the judgment is no, adjust the thought chain prompt corresponding to the output result until the preset condition is met;

[0040] If the judgment is yes, the thinking chain prompt corresponding to the output result is determined as the preset thinking chain prompt.

[0041] Furthermore, while generating the operation and maintenance instructions according to the operation and maintenance strategy, it also includes:

[0042] generating hardware fault indication information;

[0043] Based on the hardware fault indication information, the staff is prompted to handle the hardware level problem.

[0044] Furthermore, the method further comprises:

[0045] Obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction;

[0046] Determine the system operation and maintenance result according to the instruction operation and maintenance result;

[0047] The system operation and maintenance results are used as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results are used as feature data to optimize the large language model.

[0048] Furthermore, the method further comprises:

[0049] Obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction, and the hardware operation and maintenance result corresponding to the hardware fault indication information;

[0050] Determine the system operation and maintenance result according to the instruction operation and maintenance result and the hardware operation and maintenance result;

[0051] The system operation and maintenance results are used as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results are used as feature data to optimize the large language model.

[0052] A second aspect of the present invention provides an intelligent operation and maintenance device based on a large language model, the device comprising:

[0053] An acquisition module is used to acquire real-time operation data of the software system; the real-time operation data includes real-time log data and performance monitoring data;

[0054] A first input module, used to determine an operation and maintenance strategy based on the real-time operation data and the pre-trained large language model;

[0055] A second input module is used to generate an operation and maintenance instruction according to the operation and maintenance strategy;

[0056] An operation module, used to sequentially execute the operation and maintenance instructions, and identify whether the system is abnormal after each operation and maintenance instruction is executed;

[0057] The control module is used to roll back the parameter values ​​of the operation and maintenance instructions to the values ​​before the operation and maintenance and send an alarm message if a system abnormality is identified;

[0058] Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

[0059] The third aspect of this document provides a computer device, including a memory, a processor, and a computer program stored on the memory, wherein when the computer program is executed by the processor, the computer program executes instructions of the intelligent operation and maintenance method based on a large language model described in any of the aforementioned embodiments.

[0060] The fourth aspect of this document provides a computer storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer device, the computer program executes the instructions of the intelligent operation and maintenance method based on a large language model described in any of the aforementioned embodiments.

[0061] A fifth aspect of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor of a computer device, the computer program executes instructions of the intelligent operation and maintenance method based on a large language model described in any of the aforementioned embodiments.

[0062] The intelligent operation and maintenance method, device and equipment based on the large language model provided in this article take into account that the training of the large language model does not require complex feature engineering and the large language model has the ability to process tasks across platforms. The real-time operation data of the software system is identified through the pre-trained large language model, and the operation and maintenance strategy is determined, thereby improving the adaptability of the software system operation and maintenance; according to the operation and maintenance strategy determined by the large language model, the operation and maintenance instructions are generated, and the troubleshooting operations are automatically performed according to the operation and maintenance instructions, thereby reducing the manpower cost of software operation and maintenance and improving the operation and maintenance efficiency; at the same time, the results of each execution of the operation and maintenance instructions are monitored. When a system abnormality is identified, it is rolled back in time and an alarm message is sent, thereby ensuring the system security of the operation and maintenance process, further reducing the manpower cost required for monitoring, and improving the operation and maintenance efficiency.

[0063] In order to make the above and other purposes, features and advantages of this article more obvious and easy to understand, the following specifically cites preferred embodiments and describes them in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of this article or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of this article. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0065] Figure 1 A first flow chart of the intelligent operation and maintenance method based on a large language model according to an embodiment of this invention is shown;

[0066] Figure 2 A flowchart of extracting key content data according to an embodiment of this invention is shown;

[0067] Figure 3 A flow chart showing structured data in an embodiment of this invention is shown;

[0068] Figure 4 A flow chart showing the operation and maintenance strategy determined in the embodiment of this article is shown;

[0069] Figure 5 A flowchart of obtaining a context example group according to an embodiment of this invention is shown;

[0070] Figure 6 A flowchart of obtaining preset thought chain prompts in the embodiment of this article is shown;

[0071] Figure 7 A flowchart of generating hardware fault indication information according to an embodiment of the present invention is shown;

[0072] Figure 8 A second flow chart of the intelligent operation and maintenance method based on a large language model according to an embodiment of this invention is shown;

[0073] Fig. 9 A third flow chart of the intelligent operation and maintenance method based on a large language model in the embodiment of this article is shown;

[0074] Fig.10 The structure diagram of the intelligent operation and maintenance device based on the large language model in the embodiment of this article is shown;

[0075] Fig.11 A structural diagram of a computer device according to an embodiment of this invention is shown.

[0076] Description of the accompanying symbols:

[0077] 1010. Obtain module;

[0078] 1020, a first input module;

[0079] 1030, second input module;

[0080] 1040, operation module;

[0081] 1050, control module;

[0082] 1102. Computer equipment;

[0083] 1104. Processor;

[0084] 1106. Memory;

[0085] 1108, driving mechanism;

[0086] 1110, input / output module;

[0087] 1112. Input device;

[0088] 1114. Output device;

[0089] 1116. Presentation equipment;

[0090] 1118. Graphical user interface;

[0091] 1120, network interface;

[0092] 1122. Communication link;

[0093] 1124. Communication bus. DETAILED DESCRIPTION

[0094] The following will be combined with the drawings in the embodiments of this article to clearly and completely describe the technical solutions in the embodiments of this article. Obviously, the described embodiments are only part of the embodiments of this article, not all of the embodiments. Based on the embodiments of this article, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this article.

[0095] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0096] This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on routine or non-creative work. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel.

[0097] It should be noted that the acquisition, storage, use, and processing of data in the technical solutions of the embodiments of this specification comply with the relevant provisions of national laws and regulations.

[0098] 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.

[0099] In one embodiment of the present invention, an intelligent operation and maintenance method based on a large language model is provided to solve the problems of high operation and maintenance cost, low efficiency, poor adaptability and weak automated operation and maintenance capability of software systems.

[0100] Specifically, Figure 1 As shown in the figure, the intelligent operation and maintenance methods based on the large language model include:

[0101] Step 110, obtaining real-time operation data of the software system;

[0102] Wherein, the real-time operation data includes real-time log data and performance monitoring data;

[0103] Step 120, determining an operation and maintenance strategy according to the real-time operation data and the pre-trained large language model;

[0104] Step 130, generating an operation and maintenance instruction according to the operation and maintenance strategy;

[0105] Step 140, executing the operation and maintenance instructions in sequence, and identifying whether the system is abnormal after each execution of the operation and maintenance instructions;

[0106] Step 150: If a system abnormality is identified, the parameter value of the operation and maintenance instruction is rolled back to before the operation and maintenance and an alarm message is sent;

[0107] Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

[0108] This embodiment uses a pre-trained large language model to identify the real-time operating data of the software system, and then determines the operation and maintenance strategy, thereby improving the adaptability of the software system operation and maintenance; generates operation and maintenance instructions based on the operation and maintenance strategy determined by the large language model, and automatically troubleshoots according to the operation and maintenance instructions, thereby reducing the manpower cost of software operation and maintenance and improving the operation and maintenance efficiency; at the same time, monitors the results of each execution of the operation and maintenance instructions, and when a system abnormality is identified, rolls back in time and sends an alarm message, thereby ensuring the system security of the operation and maintenance process, further reducing the manpower cost required for monitoring, and improving the operation and maintenance efficiency.

[0109] In some embodiments of this document, log data includes log information recorded in log files, and performance monitoring data includes CPU usage, memory consumption, network bandwidth, etc.; after obtaining the real-time operation data of the software system, the real-time operation data and related prompt information are input into a pre-trained large language model to determine the operation and maintenance strategy, and related prompt information such as "analyze the operation data and provide an operation and maintenance strategy."

[0110] In some embodiments of the present invention, after determining the operation and maintenance strategy, in order to generate code that can be recognized by the computer, it is necessary to generate operation and maintenance instructions according to the operation and maintenance strategy. A matching list of preset operation and maintenance strategies and operation and maintenance instructions can be selected, or the operation and maintenance strategy can be input into a large language model, and prompt information can be attached to require the large language model to generate operation and maintenance instructions according to the operation and maintenance strategy, thereby rationally utilizing the large language model to speed up processing efficiency; the operation and maintenance instructions are code instructions that can be recognized by the computer. For example, if the operation and maintenance instructions need to be output in Java language, a prompt information of "output operation and maintenance instructions written in Java language according to the operation and maintenance strategy" can be attached.

[0111] The generated operation and maintenance instruction may be one or more. In some embodiments of the present invention, after the operation and maintenance instruction is generated, the generated operation and maintenance instructions are executed in sequence to realize automatic operation and maintenance of the system. In order to prevent the system from being unable to handle abnormalities after the operation and maintenance instructions are executed, an automatic monitoring method is adopted. After an operation and maintenance instruction is executed, if a system abnormality is identified (for example, an application or a function of the system reports an error or fails), the parameter value of the operation and maintenance instruction is rolled back to the value before the operation and maintenance, and an alarm message is sent to the system administrator. For example, the operation and maintenance strategy includes the strategy of "reducing the bandwidth of application A by half", and a code (operation and maintenance instruction) is generated to execute the strategy. One of the parameter values ​​of the operation and maintenance instruction is the bandwidth of application A. After the operation and maintenance, the bandwidth of application A is reduced to 1 / 2 of the original value. At this time, if the system abnormality is monitored, a rollback is executed to restore the bandwidth of application A to the value before the operation and maintenance, and an alarm message is sent to remind the operation and maintenance management personnel.

[0112] In some embodiments of this invention, when multiple operation and maintenance instructions are generated, the real-time status of the system, the executed operation and maintenance instructions, and the corresponding operation and maintenance strategies can be input into the large language model again after each operation and maintenance instruction is executed, and additional prompt information can be attached, requiring the large language model to output a new operation and maintenance strategy after analysis.

[0113] In the embodiments of this article, the large language model used is trained based on the historical operation data, historical operation and maintenance strategies, and historical operation and maintenance results of the software system. The historical operation data and historical operation and maintenance strategies serve as feature data that the large language model needs to learn, and the historical operation and maintenance results serve as labels for training the large language model; historical operation data refers to the log data and performance monitoring data of the past time period recorded by the system, the historical operation and maintenance strategy refers to the operation and maintenance strategy corresponding to the historical operation data of the past time period recorded by the system, and the historical operation and maintenance results refer to the operation and maintenance success or failure results of the operation and maintenance strategy corresponding to the historical operation and maintenance strategy of the past time period recorded by the system. The large language model can adopt the BERT series, GPT series, Switch Transformer series, Tongyi Qianwen, etc., which are not limited in this article.

[0114] In one embodiment of this invention, Figure 2As shown, before the real-time running data is input into the pre-trained large language model, it also includes:

[0115] Step 210, extracting key content data from the real-time log data;

[0116] Step 220, dividing the key content data into a plurality of overlapping content blocks;

[0117] Step 230, inputting each overlapping content block into the large language model to obtain a plurality of refined content blocks;

[0118] Step 240, inputting the plurality of refined content blocks into the large language model to obtain refined key content data;

[0119] Step 250: Adjust the real-time log data into refined key content data.

[0120] This embodiment takes into account that the content of the running data may be too long, causing it to exceed the context length window limit of the large language model, affecting the analysis and processing of the running data; therefore, the amount of information input to the large language model is reduced by refining real-time log data, ensuring that the large language model can correctly analyze the log data, while improving the processing efficiency of the large language model.

[0121] In some embodiments of the present invention, key content data in real-time log data is extracted through regular expressions or log parsing tools, and the key content data includes exception type, occurrence time, impact scope, etc.

[0122] The extracted key content data may still be too long, thus affecting the analysis of the large language model. Therefore, in some embodiments of this invention, the key content data is divided into multiple overlapping content blocks, each overlapping content block is input into the large language model, and the prompt information of the refined content is attached. Each overlapping content block is processed separately by the large language model, and the information of each overlapping content block is refined to obtain multiple refined content blocks; then the multiple refined content blocks are input into the large language model, and the prompt information of the integrated content is attached. Since each refined content block has overlapping content, the large language model can identify the previous and next association relationship of each refined content block according to the overlapping content when integrating the content, thereby obtaining the key content data after integrating multiple refined content blocks. The method of dividing multiple overlapping content blocks can use a sliding window, divide according to a fixed number of rows, timestamps, etc., or other methods, which are not limited in this invention.

[0123] In one embodiment of this invention, Figure 3 As shown, before the real-time running data is input into the pre-trained large language model, it also includes:

[0124] Step 310, obtaining data with structured requirements in the real-time operation data and format requirements corresponding to the data with structured requirements;

[0125] Step 320, adjusting the language token probability of the large language model according to the format requirement;

[0126] Step 330, inputting the data with structuring requirements into the large language model to obtain structured data;

[0127] Step 340: Adjust the data with structuring requirements into the structured data.

[0128] This embodiment takes into account that some data have format requirements. In order to ensure that the large language model can correctly output structured data that meets the requirements, the language token probability of the large language model is adjusted to increase the probability of the large language model selecting the correct language token, so that more accurate content can be output.

[0129] In some embodiments of the present invention, taking into account that some types of data may have multiple representation forms, for example, the month can be represented by Chinese characters, English, symbols, numbers, etc., in order to reduce the error probability of the large language model, it is stipulated that the month can only be represented in a certain form. For example, only the English abbreviation JAN can be used to represent January, and FEB can be used to represent February. At this time, by adjusting the language token probability of the large language model, the large language model can only select the correct month from the months with English abbreviations.

[0130] For ease of operation, some embodiments of the present invention set the probability of language tokens that do not meet format requirements to zero to adjust the language token probability of the large language model so that the large language model can only select the correct month from the months abbreviated in English.

[0131] In one embodiment of this invention, Figure 4 As shown, according to the real-time operation data and the pre-trained large language model, an operation and maintenance strategy is determined, including:

[0132] Step 410, obtaining a context example group from a context prompt library;

[0133] The context example group includes one or more context examples, the context examples are generated according to the historical operation and maintenance data of the system, and the context examples include the cause and solution of the system failure;

[0134] Step 420, inputting the real-time operation data, the context example group and the preset thought chain prompt into the large language model to obtain an operation and maintenance strategy;

[0135] Among them, the preset thinking chain prompt is used to guide the analysis of the real-time operation data.

[0136] This embodiment assists the large language model in determining the operation and maintenance strategy through context example groups and preset thought chain prompts, so that the large language model can output a more accurate operation and maintenance solution.

[0137] In the embodiments of this article, a preset thinking chain prompt is used, such as "You are reading the server log, and your goal is to understand the system status and problem-solving process. Please follow the following steps to analyze: Step 1, check each log entry; Step 2, identify and analyze errors; Step 3, infer the system status." Combined with the preset thinking chain prompt, the large language model analyzes the real-time operation data. When an abnormal system status is identified, combined with the system failure causes and solutions in the context example group, it is inferred whether the cause of the current system failure is similar to the system failure cause in the context example group, and a similar solution is given. System failure causes in the context example group, such as "network delay" and "database timeout", all have corresponding historical solutions.

[0138] In addition, when it comes to complex problems, the output of operation and maintenance strategies can be achieved by alternating between preset thinking chain prompts and context example groups, making the output of operation and maintenance strategies more accurate.

[0139] In one embodiment of this invention, Figure 5 As shown, a context example group is obtained from the context prompt library, including:

[0140] Step 510, obtaining the success rates of all context example groups in the context prompt library;

[0141] Step 520, determining the sampling probability of all context example groups in the context prompt library according to the success rate;

[0142] Step 530: Obtain a context example group from a context prompt library according to the sampling probability.

[0143] This embodiment presets sampling probabilities for all context example groups according to the success rates of all context example groups in the context prompt library, so that context example groups with higher success rates are more easily selected, while ensuring randomness and taking into account the use of other context example groups.

[0144] In some embodiments of the present invention, the sampling probability can be preset based on a random number. For ease of understanding, it is assumed that there are four context example groups A, B, C, and D in the context prompt library, among which B has the highest success rate, and the success rates of other context example groups are the same. At this time, the sampling probability of B can be set to 0.4, and the sampling probabilities of A, C, and D are all set to 0.2. At this time, a random number is generated between 0 and 1. If the generated random number is in the range of 0-0.2, A is determined as the acquired context example group. If the generated random number is in the range of 0.2-0.6, B is determined as the acquired context example group. If the generated random number is in the range of 0.6-0.8, C is determined as the acquired context example group. If the generated random number is in the range of 0.8-1, D is determined as the acquired context example group.

[0145] The ratio of sampling probabilities may be determined according to the ratio of success rates of each context example group, or by adding weights based on the success rate ratios, or by using a custom rule to determine the sampling probability, or by other feasible methods.

[0146] In some embodiments of the present invention, after the operation and maintenance is completed, the success rate of the context example group used this time can be adjusted according to whether the operation and maintenance result is success or failure, thereby updating the sampling probability of the context example group in the context prompt library, so that the sampling probability of the context example group can change in real time with the change of the system status, so that the context example group that is more beneficial to the current operation of the system can be obtained with a higher probability.

[0147] In one embodiment of this invention, Figure 6 As shown, the preset thought chain prompt is obtained in the following manner:

[0148] Step 610, obtaining an initial thought chain prompt, a preset context example group and historical operation data of the system;

[0149] Step 620, inputting the initial thought chain prompt, the preset context example group and the historical operation data of the system into the large language model to obtain an output result;

[0150] Step 630, determining whether the output result satisfies a preset condition, if the determination is no, executing step 640, if the determination is yes, executing step 650;

[0151] Step 640, adjusting the thought chain prompt corresponding to the output result until a preset condition is met;

[0152] Step 650: determine the thought chain prompt corresponding to the output result as the preset thought chain prompt.

[0153] This embodiment experiments with thought chain prompts through preset context example groups and historical operation data of the system, and determines the best thought chain prompt as the preset thought chain prompt through iterative adjustment, thereby providing assistance for the subsequent large language model output operation and maintenance strategy, and improving the accuracy of the operation and maintenance strategy output of the large language model.

[0154] The initial thinking chain prompt can be a combination of some prompt information, the output result can be an operation and maintenance strategy for historical operation data, and the preset condition can be the accuracy of the output operation and maintenance strategy based on the thinking chain prompt. When the preset condition is not met, other prompt information combinations can be replaced to adjust the thinking chain prompt, so as to complete the iteration of the thinking chain prompt.

[0155] In one embodiment of this invention, Figure 7 As shown, while generating the operation and maintenance instructions according to the operation and maintenance strategy, it also includes:

[0156] Step 710, generating hardware failure indication information;

[0157] Step 720, prompting staff to handle hardware-level problems based on the hardware fault indication information.

[0158] This embodiment generates operation and maintenance instructions according to the operation and maintenance strategy and also generates hardware fault indication information. The hardware fault indication information prompts the staff to deal with hardware-level problems, thereby performing comprehensive troubleshooting.

[0159] The hardware fault indication information may include analyzed faulty hardware device information, such as a physical connection of a certain device being disconnected, or a facility being damaged, etc. Based on the hardware fault indication information, the staff is prompted to handle the equipment problem.

[0160] In one embodiment of this invention, Figure 8 As shown in the figure, the intelligent operation and maintenance method based on the large language model also includes:

[0161] Step 810, obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction;

[0162] Step 820, determining a system operation and maintenance result according to the instruction operation and maintenance result;

[0163] Step 830, using the system operation and maintenance results as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results as feature data, to optimize the large language model.

[0164] After each operation and maintenance is completed, this embodiment uses the system operation and maintenance results determined by the instruction operation and maintenance results as training labels, and uses the real-time operation data and operation and maintenance strategies as feature data to continuously optimize the large language model to ensure that the large language model can adapt to changes in the system operation conditions, thereby improving the adaptability of the large model and the accuracy of the operation and maintenance strategy output.

[0165] The instruction operation and maintenance results include operation and maintenance success or operation and maintenance failure. Operation and maintenance success means that the system returns to a normal state after the operation and maintenance instructions are run. Operation and maintenance failure means that the system fails to return to a normal state after the operation and maintenance instructions are run, or the system encounters an abnormality and rolls back during the operation and maintenance instructions. When the instruction operation and maintenance result is operation and maintenance success, it can be determined that the system operation and maintenance result is operation and maintenance success. When the instruction operation and maintenance result is operation and maintenance failure, it can be determined that the system operation and maintenance result is operation and maintenance failure.

[0166] In one embodiment of this invention, Fig. 9 As shown in the figure, the intelligent operation and maintenance method based on the large language model also includes:

[0167] Step 910, obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction, and the hardware operation and maintenance result corresponding to the hardware fault indication information;

[0168] Step 920, determining a system operation and maintenance result according to the instruction operation and maintenance result and the hardware operation and maintenance result;

[0169] Step 930: Use the system operation and maintenance results as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results as feature data to optimize the large language model.

[0170] After each operation and maintenance is completed, this embodiment uses the system operation and maintenance results determined by the instruction operation and maintenance results and the hardware operation and maintenance results as training labels, and uses the real-time operation data and operation and maintenance strategies as feature data to continuously optimize the large language model to ensure that the large language model can adapt to changes in the system operation conditions, thereby improving the adaptability of the large model and the accuracy of the operation and maintenance strategy output.

[0171] When the large language model outputs hardware fault indication information, it means that there may be a problem with the hardware and the staff needs to handle it. After the staff handles it, the hardware operation and maintenance result is given as successful operation or failed operation. The successful operation here means that the faulty hardware device is the same as the hardware fault indication information, and the failed operation here means that the faulty hardware device is different from the hardware fault indication information, or no hardware device is faulty. If the large language model does not output an operation and maintenance instruction, it means that only the hardware problem is identified, and the system operation and maintenance result can be determined based on the hardware operation and maintenance result; if the large language model outputs an operation and maintenance instruction, the system operation and maintenance result needs to be determined based on the instruction operation and maintenance result and the hardware operation and maintenance result. Only when the instruction operation and maintenance result and the hardware operation and maintenance result are both successful, the system operation and maintenance result is successful.

[0172] Based on the same inventive concept, this article also provides an intelligent operation and maintenance device based on a large language model, as described in the following embodiments. Since the principle of solving the problem by the intelligent operation and maintenance device based on a large language model is similar to that of the intelligent operation and maintenance method based on a large language model, the implementation of the intelligent operation and maintenance device based on a large language model can refer to the intelligent operation and maintenance method based on a large language model, and the repeated parts will not be repeated.

[0173] Specifically, Fig.10 As shown, the intelligent operation and maintenance device based on the large language model includes:

[0174] The acquisition module 1010 is used to acquire real-time operation data of the software system; the real-time operation data includes real-time log data and performance monitoring data;

[0175] A first input module 1020, configured to determine an operation and maintenance strategy based on the real-time operation data and the pre-trained large language model;

[0176] The second input module 1030 is used to generate an operation and maintenance instruction according to the operation and maintenance strategy;

[0177] An operation module 1040 is used to sequentially execute the operation and maintenance instructions, and identify whether the system is abnormal after each operation and maintenance instruction is executed;

[0178] The control module 1050 is used to roll back the parameter values ​​of the operation and maintenance instruction to the values ​​before the operation and maintenance and send an alarm message if a system abnormality is identified;

[0179] Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

[0180] This embodiment uses a pre-trained large language model to identify the real-time operating data of the software system, and then determines the operation and maintenance strategy, thereby improving the adaptability of the software system operation and maintenance; generates operation and maintenance instructions based on the operation and maintenance strategy determined by the large language model, and automatically troubleshoots according to the operation and maintenance instructions, thereby reducing the manpower cost of software operation and maintenance and improving the operation and maintenance efficiency; at the same time, monitors the results of each execution of the operation and maintenance instructions, and when a system abnormality is identified, rolls back in time and sends an alarm message, thereby ensuring the system security of the operation and maintenance process, further reducing the manpower cost required for monitoring, and improving the operation and maintenance efficiency.

[0181] The intelligent operation and maintenance method, device and equipment based on the large language model provided in this article take into account that the training of the large language model does not require complex feature engineering and the large language model has the ability to process tasks across platforms. The real-time operation data of the software system is identified through the pre-trained large language model, and the operation and maintenance strategy is determined, thereby improving the adaptability of the software system operation and maintenance; according to the operation and maintenance strategy determined by the large language model, the operation and maintenance instructions are generated, and the troubleshooting operations are automatically performed according to the operation and maintenance instructions, thereby reducing the manpower cost of software operation and maintenance and improving the operation and maintenance efficiency; at the same time, the results of each execution of the operation and maintenance instructions are monitored. When a system abnormality is identified, it is rolled back in time and an alarm message is sent, thereby ensuring the system security of the operation and maintenance process, further reducing the manpower cost required for monitoring, and improving the operation and maintenance efficiency.

[0182] In one embodiment of the present invention, a computer device is also provided for implementing the method described in any of the above embodiments, such as Fig.11 The structure diagram of the computer device of the embodiment of this article is shown, and the computer device 1102 may include one or more processors 1104, such as one or more central processing units (CPUs), and each processing unit may implement one or more hardware threads. The computer device 1102 may also include any memory 1106, which is used to store any kind of information such as code, settings, data, etc. Non-limiting, for example, the memory 1106 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 1102. In one case, when the processor 1104 executes an associated instruction stored in any memory or a combination of memories, the computer device 1102 can perform any operation of the associated instruction. The computer device 1102 also includes one or more drive mechanisms 1108 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0183] The computer device 1102 may also include an input / output module 1110 (I / O) for receiving various inputs (via input devices 1112) and for providing various outputs (via output devices 1114). A specific output mechanism may include a presentation device 1116 and an associated graphical user interface (GUI) 1118. In other embodiments, the input / output module 1110 (I / O), input device 1111, and output device 1114 may not be included, and the computer device 1102 may be used as a computer device in a network. The computer device 1102 may also include one or more network interfaces 1120 for exchanging data with other devices via one or more communication links 1122. One or more communication buses 1124 couple the components described above together.

[0184] The communication link 1122 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 1122 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0185] Corresponds to Figures 1 to 9 The method in this article also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.

[0186] The embodiment of the present invention also provides a computer readable instruction, wherein when the processor executes the instruction, the program therein causes the processor to execute the following Figures 1 to 9 The method shown.

[0187] It should be understood that in the various embodiments of this document, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0188] It should also be understood that in the embodiments of this article, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0189] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this article.

[0190] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0191] In the several embodiments provided herein, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0192] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this article.

[0193] In addition, each functional unit in each embodiment of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

[0194] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this article. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0195] Specific embodiments are used in this article to illustrate the principles and implementation methods of this article. The description of the above embodiments is only used to help understand the methods and core ideas of this article. At the same time, for general technicians in this field, according to the ideas of this article, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on this article.

Claims

1. An intelligent operation and maintenance method based on a large language model, characterized in that: The method comprises: Acquire real-time operation data of the software system; the real-time operation data includes real-time log data and performance monitoring data; Determine an operation and maintenance strategy based on the real-time operation data and the pre-trained large language model; Generate operation and maintenance instructions according to the operation and maintenance strategy; Execute the operation and maintenance instructions in sequence, and identify whether the system is abnormal after each execution of the operation and maintenance instructions; If a system abnormality is identified, the parameter values ​​of the operation and maintenance instructions will be rolled back to before the operation and maintenance and an alarm message will be sent; Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

2. The method according to claim 1, characterized in that Before inputting the real-time running data into the pre-trained large language model, the method further includes: Extracting key content data from the real-time log data; dividing the key content data into a plurality of overlapping content blocks; Inputting each overlapping content block into the large language model respectively to obtain a plurality of refined content blocks; Inputting the plurality of refined content blocks into the large language model to obtain refined key content data; The real-time log data is adjusted to refined key content data.

3. The method according to claim 1, characterized in that Before inputting the real-time running data into the pre-trained large language model, the method further includes: Acquire data with structured requirements in the real-time operation data and format requirements corresponding to the data with structured requirements; According to the format requirement, adjusting the language token probability of the large language model; Inputting the data with structuring requirements into the large language model to obtain structured data; The data with structuring requirements is adjusted to the structured data.

4. The method according to claim 3, characterized in that According to the format requirement, adjusting the language token probability of the large language model includes: The probability of language tokens that do not conform to the format requirements is set to zero.

5. The method according to claim 1, characterized in that Determine the operation and maintenance strategy based on the real-time operation data and the pre-trained large language model, including: Obtaining a context example group from a context prompt library; the context example group includes one or more context examples, the context examples are generated based on historical operation and maintenance data of the system, and the context examples include system failure causes and solutions; The real-time operation data, the context example group and the preset thought chain prompt are input into the large language model to obtain the operation and maintenance strategy; the preset thought chain prompt is used to guide the analysis of the real-time operation data.

6. The method according to claim 5, characterized in that Get groups of contextual examples from the contextual prompts library, including: Obtaining the success rates of all context example groups in the context prompt library; Determining the sampling probability of all context example groups in the context prompt library according to the success rate; According to the sampling probability, a context example group is obtained from a context prompt library.

7. The method according to claim 6, characterized in that The method further comprises: Based on the operation and maintenance results, update the success rate of the context example group used this time.

8. The method according to claim 5, characterized in that Obtain the preset thought chain prompt in the following manner: Obtaining initial thought chain prompts, a preset context example group, and historical operation data of the system; Inputting the initial thought chain prompt, the preset context example group and the historical operation data of the system into the large language model to obtain an output result; Determine whether the output result meets a preset condition; If the judgment is no, adjust the thought chain prompt corresponding to the output result until the preset condition is met; If the judgment is yes, the thinking chain prompt corresponding to the output result is determined as the preset thinking chain prompt.

9. The method according to claim 1, characterized in that While generating the operation and maintenance instructions according to the operation and maintenance strategy, it also includes: generating hardware fault indication information; Based on the hardware fault indication information, the staff is prompted to handle the hardware level problem.

10. The method according to claim 1, characterized in that The method further comprises: Obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction; Determine the system operation and maintenance result according to the instruction operation and maintenance result; The system operation and maintenance results are used as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results are used as feature data to optimize the large language model.

11. The method according to claim 9, characterized in that The method further comprises: Obtaining the instruction operation and maintenance result corresponding to the operation and maintenance instruction, and the hardware operation and maintenance result corresponding to the hardware fault indication information; Determine the system operation and maintenance result according to the instruction operation and maintenance result and the hardware operation and maintenance result; The system operation and maintenance results are used as training labels, and the real-time operation data and operation and maintenance strategies corresponding to the system operation and maintenance results are used as feature data to optimize the large language model.

12. An intelligent operation and maintenance device based on a large language model, characterized in that: The device comprises: An acquisition module is used to acquire real-time operation data of the software system; the real-time operation data includes real-time log data and performance monitoring data; A first input module, used to determine an operation and maintenance strategy based on the real-time operation data and the pre-trained large language model; A second input module is used to generate an operation and maintenance instruction according to the operation and maintenance strategy; An operation module, used to sequentially execute the operation and maintenance instructions, and identify whether the system is abnormal after each operation and maintenance instruction is executed; The control module is used to roll back the parameter values ​​of the operation and maintenance instructions to the values ​​before the operation and maintenance and send an alarm message if a system abnormality is identified; Among them, the feature data used to train the large language model includes historical operation data and historical operation and maintenance strategies of the software system, and the training labels corresponding to the feature data include historical operation and maintenance results of the software system.

13. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 11 is implemented.

14. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor of a computer device, the method according to any one of claims 1 to 11 is implemented.

15. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor of a computer device, the method according to any one of claims 1 to 11 is implemented.

Images