Artificial intelligence operation and maintenance method and system based on large model

Abstract

The application discloses an artificial intelligence operation and maintenance method and system based on a large model, which comprises the following steps: obtaining operation data of a target platform in multiple time periods; determining first operation and maintenance instructions of the multiple time periods by using an operation and maintenance large model based on a preset operation and maintenance instruction set and the operation data, the preset operation and maintenance instruction set being associated with instruction features, the preset operation and maintenance instruction set comprising at least one operation and maintenance instruction, and the first operation and maintenance instruction being determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction features matched with operation data of a corresponding time period of the first operation and maintenance instruction; obtaining simulation operation and maintenance information of the multiple time periods based on the first operation and maintenance instructions; performing reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information to obtain systematic deviations of the multiple time periods; and generating second operation and maintenance instructions of the target platform by using a target large model based on the systematic deviations of the multiple time periods and the operation data, and sending the second operation and maintenance instructions to the target platform, so that the automation efficiency of operation and maintenance and the rationality of an operation and maintenance scheme can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of platform operation and maintenance, and in particular to an artificial intelligence operation and maintenance method and system based on a large model. BACKGROUND

[0002] In related technologies, operation and maintenance solutions for platforms (such as e-commerce platforms, financial systems, or Internet of Things devices) mainly include traditional manual operation and maintenance solutions, DevOps operation and maintenance solutions, and containerized operation and maintenance solutions.

[0003] However, in the above operation and maintenance methods, human intervention is mostly relied on during application, such as manual operation and management, or maintenance and updating of automated tools, thereby having a high dependence on humans, resulting in low efficiency and limited rationality of operation and maintenance solutions due to personnel experience. SUMMARY

[0004] To solve the above technical problems, the present application provides an artificial intelligence operation and maintenance method and system based on a large model, which can improve the automation efficiency of operation and maintenance and the rationality of operation and maintenance solutions.

[0005] In a first aspect, the present application provides an artificial intelligence operation and maintenance method based on a large model, comprising:

[0006] obtaining running data of a target platform in multiple time periods;

[0007] determining, based on at least one preset operation and maintenance instruction set and the running data, a first operation and maintenance instruction corresponding to each of the multiple time periods by using an operation and maintenance large model, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set includes at least one operation and maintenance instruction, the operation and maintenance instruction is suitable for indicating a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matching the running data of the corresponding time period;

[0008] obtaining simulation operation and maintenance information of each of the multiple time periods based on the first operation and maintenance instruction;

[0009] performing reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information to obtain a systematic bias corresponding to each of the multiple time periods, wherein the systematic bias corresponding to each time period represents a systematic error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period;

[0010] generating a second operation and maintenance instruction of the target platform by using a target large model based on the systematic bias corresponding to each of the multiple time periods and the running data, and sending the second operation and maintenance instruction to the target platform.

[0011] Optionally, the reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information comprises:

[0012] Select at least part of the time periods as enhancement time periods;

[0013] For each enhancement time period, perform enhancement processing on the simulation operation and maintenance information of the enhancement time period to obtain enhancement operation and maintenance information of the enhancement time period;

[0014] Based on the enhancement operation and maintenance information and the simulation operation and maintenance information of each of the time periods, perform reinforcement learning on the operation and maintenance large model.

[0015] Optionally, the enhancement processing on the simulation operation and maintenance information of the enhancement time period to obtain enhancement operation and maintenance information of the enhancement time period comprises:

[0016] Perform perturbation processing on the simulation operation and maintenance information of the enhancement time period to obtain perturbation operation and maintenance information;

[0017] Perform random noise adding on the perturbation operation and maintenance information to obtain noise-added operation and maintenance information;

[0018] Determine a first probability distribution of the noise-added operation and maintenance information, and determine a second probability distribution of the simulation operation and maintenance information of the enhancement time period;

[0019] Based on the first probability distribution and the second probability distribution, convert the noise-added operation and maintenance information into enhancement operation and maintenance information of the enhancement time period.

[0020] Optionally, the conversion of the noise-added operation and maintenance information into enhancement operation and maintenance information of the enhancement time period based on the first probability distribution and the second probability distribution comprises:

[0021] Determine a target noise adding strategy by analyzing the difference between the first probability distribution and the second probability distribution;

[0022] According to the target noise adding strategy, perform noise adding processing on the noise-added operation and maintenance information to obtain enhancement operation and maintenance information of the enhancement time period.

[0023] Optionally, the target large model is obtained by fine-tuning of the operation and maintenance large model via reinforcement learning.

[0024] Optionally, the obtaining of the simulation operation and maintenance information of each of the time periods based on the first operation and maintenance instruction comprises:

[0025] For the first operation and maintenance instruction corresponding to each time period, perform simulation based on the first operation and maintenance instruction to simulate running state information of the target platform under the action of the first operation and maintenance instruction, and generate simulation operation and maintenance information of the time period according to the running state information.

[0026] Optionally, in the process of reinforcement learning, the learning weight of the operation and maintenance large model for the simulation operation and maintenance information corresponding to each period is determined by the operation state information corresponding to the period.

[0027] Optionally, the operation and maintenance large model is used to determine the first operation and maintenance instruction corresponding to each of the plurality of periods based on at least one preset operation and maintenance instruction set and the operation data, including:

[0028] The at least one preset operation and maintenance instruction set and the operation data of the plurality of periods are input into the operation and maintenance large model;

[0029] For each period, the operation and maintenance large model generates a corresponding first operation and maintenance instruction based on at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction characteristics matched with the operation data of the period under the prompting of the operation and maintenance instructions.

[0030] Optionally, the target large model is used to generate the second operation and maintenance instruction of the target platform based on the systemic bias and operation data corresponding to each of the plurality of periods, including:

[0031] The systemic bias and operation data corresponding to each of the plurality of periods are input into the target large model;

[0032] The target large model determines a period weight corresponding to each period according to the systemic bias corresponding to each period;

[0033] The target large model generates the second operation and maintenance instruction based on the period weight and operation data corresponding to each of the plurality of periods.

[0034] In a second aspect, the embodiments of the present application provide an artificial intelligence operation and maintenance system based on a large model, including:

[0035] A data acquisition module is configured to acquire operation data of a target platform in a plurality of periods;

[0036] A first operation and maintenance instruction module is configured to use an operation and maintenance large model to determine a first operation and maintenance instruction corresponding to each of the plurality of periods based on at least one preset operation and maintenance instruction set and the operation data, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction characteristic, the preset operation and maintenance instruction set includes at least one operation and maintenance instruction, the operation and maintenance instruction is adapted to indicate a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction characteristics matched with the operation data of the corresponding period;

[0037] A simulation module is configured to acquire simulation operation and maintenance information of each of the plurality of periods based on the first operation and maintenance instruction.

[0038] a reinforcement learning module configured to perform reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information, to obtain a system bias corresponding to each of the plurality of time periods, wherein the system bias corresponding to each time period represents a system error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period;

[0039] a second operation and maintenance instruction module configured to generate a second operation and maintenance instruction of the target platform based on the system bias corresponding to each of the plurality of time periods and the running data, and send the second operation and maintenance instruction to the target platform.

[0040] In summary, the embodiments of the present application have at least the following beneficial effects:

[0041] By adopting the embodiments of the present application, the running data of the target platform in a plurality of time periods is obtained; based on at least one preset operation and maintenance instruction set and the running data, an operation and maintenance large model is used to determine a first operation and maintenance instruction corresponding to each of the plurality of time periods, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set includes at least one operation and maintenance instruction, the operation and maintenance instruction is suitable for indicating a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matched with the running data of the corresponding time period; based on the first operation and maintenance instruction, simulation operation and maintenance information of each of the plurality of time periods is obtained; based on the simulation operation and maintenance information, reinforcement learning is performed on the operation and maintenance large model, to obtain a system bias corresponding to each of the plurality of time periods, wherein the system bias corresponding to each time period represents a system error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period; based on the system bias corresponding to each of the plurality of time periods and the running data, a target large model is used to generate a second operation and maintenance instruction of the target platform, and the second operation and maintenance instruction is sent to the target platform, thereby improving the automation efficiency of operation and maintenance and the rationality of the operation and maintenance scheme. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a flowchart of an artificial intelligence operation and maintenance method based on a large model provided by the embodiments of the present application;

[0043] Figure 2 is a structural schematic diagram of an artificial intelligence operation and maintenance system based on a large model provided by the embodiments of the present application;

[0044] Figure 3 is a structural schematic diagram of a computer device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0045] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0046] In the description of the present application, the terms "first", "second", "third" and the like are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specified. In the description of the present application, the term "includes" and its variants are open-ended, that is, "including but not limited to". The term "based on" is "at least partially based on". The term "according to" is "at least partially according to". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments".

[0047] In the description of the present application, it should be noted that, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium; it can be the communication inside two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0048] In the description of the present application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present application have the same meaning as understood by those skilled in the art. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments, and are not intended to limit the present application. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0049] In a first aspect, see Figure 1 , a flow diagram of a large model-based artificial intelligence operation and maintenance method provided by an embodiment of the present application is shown, which includes steps S101-S105, as follows:

[0050] S101, obtaining running data of a target platform in multiple time periods.

[0051] In an example, the multiple time periods can be consecutive time periods.

[0052] In an example, the obtaining the operation data of the target platform in the plurality of time periods can include: according to a target period, obtaining continuous operation data of the target platform in a preset time length in the past, and dividing the continuous operation data into a plurality of sub-data according to time sequence, and selecting the operation data of the plurality of time periods from the plurality of divided sub-data. The target period can be preset, or can be dynamically adjusted according to the last second operation and maintenance instruction.

[0053] S102, based on at least one preset operation and maintenance instruction set and the operation data, a first operation and maintenance instruction corresponding to each of the plurality of time periods is determined by using an operation and maintenance large model, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set includes at least one operation and maintenance instruction, the operation and maintenance instruction is adapted to indicate a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matched with the operation data of the corresponding time period.

[0054] In an example, the operation and maintenance large model described above can be a pre-trained large language model, such as the Deepseek model.

[0055] In an example, the operation and maintenance large model can include:

[0056] An input layer for receiving sample operation data from a platform, including but not limited to system performance indicators (such as CPU utilization, memory usage, etc.), log files, network traffic, etc., using these data as input features of the model;

[0057] A feature extraction layer for processing time series data in sample operation data using convolutional neural networks, recurrent neural networks, long short-term memory networks, or transformer architecture in deep learning, to extract key features to facilitate the model to understand complex patterns and dependencies;

[0058] A decision layer for using a general supervised learning algorithm to decide appropriate operation and maintenance instructions based on the features extracted by the feature extraction layer and output;

[0059] An output layer for generating a first operation and maintenance instruction for a specific time period according to the output of the decision layer, and the first operation and maintenance instruction can be used to indicate at least one of the following: resource allocation adjustment, service restart, load balancing strategy change, etc.

[0060] In an example, for each preset operation and maintenance instruction set, the instructions in the second operation and maintenance instructions determined in the process of generating the second operation and maintenance instructions for the target platform and associated with the preset operation and maintenance instruction set, which meet the screening condition, can be added to the preset operation and maintenance instruction set to complete the update of the preset operation and maintenance instruction set, and the updated preset operation and maintenance instruction set is taken as the current preset operation and maintenance instruction set. Wherein, the system deviation corresponding to the second operation and maintenance instruction can be determined through the embodiments related to determining the corresponding system deviation by using the first operation and maintenance instruction in the present application, so that the screening condition can include that the system deviation corresponding to the second operation and maintenance instruction is less than the deviation threshold.

[0061] In an example, the instruction feature can be obtained by clustering the respective features of the operation and maintenance instructions included in the corresponding preset operation and maintenance instruction set.

[0062] In an example, at least one preset operation and maintenance instruction set is associated with at least one instruction feature.

[0063] In an example, the platform operation and maintenance strategy can include at least one of the following:

[0064] Performance optimization strategy, for example, the overall performance of the platform can be improved by adjusting resource configuration (such as CPU, memory allocation), database optimization, cache strategy adjustment and the like.

[0065] Security protection strategy, for example, the security of the platform system can be enhanced by firewall configuration modification, intrusion detection setting and the like to prevent potential security threats.

[0066] Data backup and recovery strategy, for example, it can indicate to immediately perform data backup and provide data recovery instruction set when needed to protect important data from loss.

[0067] Resource management strategy, for example, the use of platform computing resources is monitored and managed to ensure effective use of resources while avoiding service quality degradation caused by excessive resource consumption.

[0068] Version upgrade and maintenance strategy, for example, software and hardware upgrade plan is made to ensure that the platform technology stack remains up-to-date and can introduce the latest features.

[0069] Log management strategy, for example, how to collect, store and analyze platform system logs is specified to facilitate tracking of problem sources and improving system stability.

[0070] It can be understood that through the above platform operation and maintenance strategy, the operation and maintenance large model and / or the target large model can be prompted to match the corresponding strategy according to the running data, so as to improve the reliability and rationality of the generated corresponding operation and maintenance instructions.

[0071] S103, based on the first operation and maintenance instruction, obtaining the simulation operation and maintenance information of each time period.

[0072] In an example, the simulation operation and maintenance information can be obtained by simulation according to the first operation and maintenance instruction of each time period through operation and maintenance simulation software pre-set according to the relevant parameters of the target platform. The simulation operation and maintenance information can be used to indicate at least one of the following: platform system performance, resource utilization, operation and maintenance cost (such as time and / or resource consumption, etc.) of the target platform under the action of the corresponding first operation and maintenance instruction. The operation and maintenance simulation software can include at least one of the following: Apache JMeter for e-commerce platforms (which can be used for load testing and performance measurement to help evaluate the performance of e-commerce platforms under high concurrency), Simulacron for financial systems (a simulator for Cassandra databases that can be used to test database operations in financial systems), IoT-LAB for Internet of Things devices (which can be used to provide real hardware resources and simulation tools to test Internet of Things applications and services, including protocol stacks, energy consumption analysis, etc.).

[0073] S104, based on the simulation operation and maintenance information, performing reinforcement learning on the operation and maintenance large model to obtain the system bias corresponding to each time period, wherein the system bias corresponding to each time period represents the system error of the target platform with respect to the operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period.

[0074] In an example, under the reinforcement learning framework, the operation and maintenance large model can adjust its model parameters according to the simulation operation and maintenance information. The adjustment direction / adjustment purpose can be to minimize the difference between the operation and maintenance evaluation index and the expected target. In this process, the operation and maintenance large model can try to find the optimal operation and maintenance strategy to make the actual performance of the target platform as close to the ideal state as possible. For each time period, the system bias corresponding to the time period can be calculated based on the simulation operation and maintenance information and the learning result of the operation and maintenance large model. Here, the system bias can represent the system error of the target platform with respect to the operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period. In other words, the system bias can be a measure of the gap between the actual operation and maintenance effect and the expected value. At this time, the system bias can be obtained by calculating the measure value (such as similarity, norm, etc.) for measuring the gap.

[0075] In an example, for the reinforcement learning algorithm, a policy gradient method, DQN (Deep Q-Networks) and the like can be adopted to optimize long-term rewards so as to find the operation and maintenance instructions that can match the operation and maintenance evaluation indicators (such as maximizing system performance and / or minimizing failure rate, which can be realized by regarding different targets as different sub-targets of the optimization algorithm in the "and" relationship) through the rewards. In the reinforcement learning stage, the operation and maintenance large model can observe feedback by using the simulated operation and maintenance information, and adjust the strategy of the operation and maintenance large model according to the observed feedback (such as system stability, performance improvement, etc.) to obtain better long-term effect.

[0076] In S105, a second operation and maintenance instruction of the target platform is generated by using a target large model based on the system bias and the operation data of each of the plurality of time periods, and the second operation and maintenance instruction is sent to the target platform.

[0077] In an example, the above-mentioned target large model can be a pre-trained large language model, and its related structure can refer to the embodiments related to the operation and maintenance large model in the present application, which will not be described here. However, it should be understood that the above-mentioned target large model can be an independent large language model.

[0078] It can be understood that the embodiments of the present application first obtain the operation data of the target platform in a plurality of time periods, ensuring a comprehensive understanding of the actual state of the platform system in the target time period (for example, in a certain historical period in the past, which can be used to assist in analyzing the most appropriate second operation and maintenance instruction in a subsequent period of time); by using the operation and maintenance large model, the first operation and maintenance instruction is determined in combination with the preset operation and maintenance instruction set and the operation data, so as to consider the complex / typical factors represented by the preset operation and maintenance instruction set, so as to generate a more reasonable and effective operation and maintenance strategy; then the simulated operation and maintenance information is obtained based on the first operation and maintenance instruction, and the operation and maintenance large model is reinforced to obtain the system bias, so as to evaluate the effectiveness of the decision corresponding to each first operation and maintenance instruction, and identify the system bias, which can be used to represent potential problems; finally, the second operation and maintenance instruction is generated by using the target large model based on the system bias and the operation data, so that the target large model can optimize the operation and maintenance strategy by fully considering the system bias (such as the potential problems represented by the system bias), thereby reducing the possibility of wrong decisions, improving the rationality of the operation and maintenance scheme, and the embodiments of the present application can also reduce the participation of artificial operation and maintenance process to improve the automation efficiency, so as to realize the function of AIOps (Artificial Intelligence for IT Operations, Artificial Intelligence for IT Operations).

[0079] In an optional implementation, the reinforcement learning of the operation and maintenance large model based on the simulated operation and maintenance information comprises:

[0080] select at least part of the plurality of time periods as an enhanced time period;

[0081] for each enhanced time period, perform enhancement processing on the simulation operation and maintenance information of the enhanced time period to obtain enhanced operation and maintenance information of the enhanced time period;

[0082] based on the enhanced operation and maintenance information and the simulation operation and maintenance information of each of the plurality of time periods, perform reinforcement learning on the operation and maintenance large model.

[0083] In the embodiment, the simulation operation and maintenance information of the enhanced time period can be subjected to enhancement processing to increase the amount of operation and maintenance information of the enhanced time period, thereby improving the reliability and accuracy of the systematic bias corresponding to the relevant enhanced time period obtained through reinforcement learning.

[0084] In an example, the above-mentioned enhancement processing can refer to information amount expansion of the simulation operation and maintenance information, for example, using multiple simulation methods to generate corresponding simulation operation and maintenance information according to the first operation and maintenance instruction corresponding to the enhanced time period, and fusing the simulation operation and maintenance information generated by different simulation methods to obtain the enhanced operation and maintenance information of the enhanced time period.

[0085] In an optional implementation, the enhancement processing on the simulation operation and maintenance information of the enhanced time period to obtain the enhanced operation and maintenance information of the enhanced time period includes:

[0086] performing disturbance processing on the simulation operation and maintenance information of the enhanced time period to obtain disturbance operation and maintenance information;

[0087] performing random noise addition on the disturbance operation and maintenance information to obtain noise-added operation and maintenance information;

[0088] determining a first probability distribution of the noise-added operation and maintenance information, and determining a second probability distribution of the simulation operation and maintenance information of the enhanced time period;

[0089] based on the first probability distribution and the second probability distribution, converting the noise-added operation and maintenance information into the enhanced operation and maintenance information of the enhanced time period.

[0090] In the embodiment, the simulation operation and maintenance information can be subjected to disturbance processing and random noise addition in sequence, so that the simulation operation and maintenance information has a large amount of interference and noise, so as to increase the amount of information; then, the noise-added operation and maintenance information can be converted into the enhanced operation and maintenance information of the enhanced time period through the comparison result / difference between the first probability distribution and the second probability distribution.

[0091] In an example, the first probability distribution can include a first cross-attention distribution, and the second probability distribution can include a second cross-attention distribution.

[0092] In an example, if the noise added by the random noise adding is Gaussian noise, the first probability distribution can include a first Gaussian distribution, and the second probability distribution can include a second Gaussian distribution.

[0093] In an example, the first probability distribution and the second probability distribution can be obtained using kernel density estimation.

[0094] In an example, based on the first probability distribution and the second probability distribution, converting the noisy operation and maintenance information into the enhanced operation and maintenance information of the enhanced period can include: inputting the noisy operation and maintenance information, the first probability distribution, and the second probability distribution into a target large model, so that the target large model can convert the noisy operation and maintenance information into the enhanced operation and maintenance information of the enhanced period under the prompting of the difference between the first probability distribution and the second probability distribution. Wherein, the related description of the target large model can refer to the corresponding embodiments of the present application, and will not be repeated here. In addition, it should be noted that in the art, it is not difficult to understand that the target large model can realize the functions described in the present embodiment by performing corresponding fine-tuning on the pre-trained large model.

[0095] In an alternative embodiment, based on the first probability distribution and the second probability distribution, converting the noisy operation and maintenance information into the enhanced operation and maintenance information of the enhanced period includes:

[0096] By analyzing the difference between the first probability distribution and the second probability distribution, a target noise adding strategy is determined;

[0097] According to the target noise adding strategy, the noisy operation and maintenance information is added to obtain the enhanced operation and maintenance information of the enhanced period.

[0098] In the present embodiment, the target noise adding strategy can adapt to the difference between the first probability distribution and the second probability distribution, so that subsequent noise adding can be targeted according to the difference, thereby expanding the amount of information by adding noise, and the added noise can adapt to the difference, so that the probability distribution of the enhanced operation and maintenance information obtained by adding noise can also tend to the second probability distribution of the simulation operation and maintenance information (i.e., the target noise adding strategy can be used to indicate that the probability distribution of the enhanced operation and maintenance information tends to the second probability distribution of the simulation operation and maintenance information). In this way, in addition to the expansion of the amount of information, the present embodiment can also improve the adaptability of the operation and maintenance large model after reinforcement learning to interference and / or noise, so that in the next use, the operation and maintenance large model with stronger adaptability to interference and / or noise can be called.

[0099] In an example, the target noise adding strategy described above can be used to add target noise to the noisy operation and maintenance information according to the difference between the first probability distribution and the second probability distribution, so that the probability distribution of the enhanced operation and maintenance information converges to the second probability distribution of the simulated operation and maintenance information. For example, for a certain information region, if the difference represents that the gap between the two probability distributions at this information region is small, then more noise in the added target noise needs to be added to this information region. Correspondingly, at another information region, if the difference represents that the gap between the two probability distributions at this information region is large, then less noise in the added target noise needs to be added to this information region. In this way, the gap between the two probability distributions at different information regions can be pulled to a similar level. Since the probability distribution is a relative quantity, it represents the probability at each point, so at this time the probability distribution of the enhanced operation and maintenance information can actually converge to the second probability distribution of the simulated operation and maintenance information.

[0100] In an optional implementation, the target large model is obtained by fine-tuning the operation and maintenance large model via reinforcement learning.

[0101] In an optional implementation, the obtaining of the simulated operation and maintenance information of each time period based on the first operation and maintenance instruction comprises:

[0102] For each first operation and maintenance instruction corresponding to a time period, simulation is performed based on the first operation and maintenance instruction to simulate the running state information of the target platform under the action of the first operation and maintenance instruction, and the simulated operation and maintenance information of the time period is generated according to the running state information.

[0103] In an example, simulation can be performed by using an operation and maintenance simulation software that is pre-set according to relevant parameters of the target platform. The operation and maintenance simulation software can include at least one of the following: Apache JMeter for e-commerce platforms (which can be used for load testing and performance measurement to help evaluate the performance of e-commerce platforms under high concurrency), Simulacron for financial systems (a simulator for Cassandra databases, which can be used to test database operations in financial systems), IoT-LAB for Internet of Things devices (which can be used to provide real hardware resources and simulation tools, and can be used to test Internet of Things applications and services, including protocol stacks, energy consumption analysis, etc.).

[0104] In an optional implementation, in the process of reinforcement learning, the learning weight of the operation and maintenance large model for the simulated operation and maintenance information corresponding to each time period is determined by the running state information corresponding to the time period.

[0105] In an example, the learning weight described above can include at least one of the following: a step size (i.e., learning rate) used when updating the model parameters, an importance weight assigned to different simulated operation and maintenance information.

[0106] In an example, the running state information described above can include at least one of the following:

[0107] Performance indicators, for example, can include CPU utilization, memory usage (including physical memory usage and virtual memory usage), disk I / O (used to measure the speed and frequency of read and write operations), and / or network traffic (used to indicate the number and speed of data packets in and out of the platform system);

[0108] Resource availability, for example, can include service availability (characterizing whether a particular service is online and accessible) and / or bandwidth usage (characterizing the maximum capacity of a network connection and the current usage ratio);

[0109] Application-specific indicators (which can vary depending on different platforms), for example, can include: for financial systems, transaction success rate and / or latency time, etc.; for Internet of Things devices, sensor data accuracy and / or response time, etc.

[0110] In an optional implementation, the determination of the first operation and maintenance instruction corresponding to each of the plurality of time periods by the operation and maintenance large model based on the at least one preset operation and maintenance instruction set and the running data includes:

[0111] inputting the at least one preset operation and maintenance instruction set and the running data of the plurality of time periods into the operation and maintenance large model;

[0112] for each time period, generating a corresponding first operation and maintenance instruction based on the running data of the time period under the prompt of at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction features matching the running data of the time period by the operation and maintenance large model.

[0113] In an optional implementation, the generation of the second operation and maintenance instruction of the target platform by the target large model based on the systemic bias and running data corresponding to each of the plurality of time periods includes:

[0114] inputting the systemic bias and running data corresponding to each of the plurality of time periods into the target large model;

[0115] determining, by the target large model, a time period weight corresponding to each time period according to the systemic bias corresponding to each time period;

[0116] generating, by the target large model, the second operation and maintenance instruction based on the time period weight and running data corresponding to each of the plurality of time periods.

[0117] In a second aspect, correspondingly, the embodiments of the present application also provide an artificial intelligence operation and maintenance system based on a large model, which can implement all processes of the artificial intelligence operation and maintenance method based on a large model provided by the above embodiments.

[0118] Referring to Figure 2 , a structure schematic diagram of an artificial intelligence operation and maintenance system based on a large model provided by the embodiments of the present application is shown, the artificial intelligence operation and maintenance system based on a large model comprises:

[0119] The data acquisition module 201 is configured to acquire running data of a target platform in a plurality of time periods.

[0120] The first operation and maintenance instruction module 202 is configured to determine, based on at least one preset operation and maintenance instruction set and the running data, a first operation and maintenance instruction corresponding to each of the plurality of time periods by using an operation and maintenance large model, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set comprises at least one operation and maintenance instruction, the operation and maintenance instruction is adapted to indicate a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matched with the running data of the corresponding time period.

[0121] The simulation module 203 is configured to acquire simulation operation and maintenance information of each of the plurality of time periods based on the first operation and maintenance instruction.

[0122] The reinforcement learning module 204 is configured to perform reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information, to obtain a systematic bias corresponding to each of the plurality of time periods, wherein the systematic bias corresponding to each time period represents a systematic error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period.

[0123] The second operation and maintenance instruction module 205 is configured to generate a second operation and maintenance instruction of the target platform by using a target large model based on the systematic bias corresponding to each of the plurality of time periods and the running data, and send the second operation and maintenance instruction to the target platform.

[0124] In an optional implementation, the reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information comprises:

[0125] Selecting at least part of the plurality of time periods as enhancement time periods;

[0126] For each enhancement time period, performing enhancement processing on the simulation operation and maintenance information of the enhancement time period to obtain enhancement operation and maintenance information of the enhancement time period;

[0127] Performing reinforcement learning on the operation and maintenance large model based on the enhancement operation and maintenance information and the simulation operation and maintenance information of each of the plurality of time periods.

[0128] In an optional implementation, the simulation operation and maintenance information of the enhanced period is enhanced to obtain enhanced operation and maintenance information of the enhanced period, including:

[0129] The simulation operation and maintenance information of the enhanced period is disturbed to obtain disturbed operation and maintenance information;

[0130] The disturbed operation and maintenance information is randomly added with noise to obtain noise-added operation and maintenance information;

[0131] The first probability distribution of the noise-added operation and maintenance information is determined, and the second probability distribution of the simulation operation and maintenance information of the enhanced period is determined;

[0132] Based on the first probability distribution and the second probability distribution, the noise-added operation and maintenance information is converted into enhanced operation and maintenance information of the enhanced period.

[0133] In an optional implementation, the noise-added operation and maintenance information is converted into enhanced operation and maintenance information of the enhanced period based on the first probability distribution and the second probability distribution, including:

[0134] By analyzing the difference between the first probability distribution and the second probability distribution, a target noise-adding strategy is determined;

[0135] The noise-added operation and maintenance information is added with noise according to the target noise-adding strategy to obtain enhanced operation and maintenance information of the enhanced period.

[0136] In an optional implementation, the target large model is obtained by fine-tuning the operation and maintenance large model via reinforcement learning.

[0137] In an optional implementation, the simulation operation and maintenance information of each period is obtained based on the first operation and maintenance instruction, including:

[0138] For the first operation and maintenance instruction corresponding to each period, simulation is performed based on the first operation and maintenance instruction to simulate the running state information of the target platform under the action of the first operation and maintenance instruction, and the simulation operation and maintenance information of the period is generated according to the running state information.

[0139] In an optional implementation, in the process of reinforcement learning, the learning weight of the simulation operation and maintenance information of each period by the operation and maintenance large model is determined by the running state information corresponding to the period.

[0140] In an optional implementation, the first operation and maintenance instruction corresponding to each period is determined by the operation and maintenance large model based on at least one preset operation and maintenance instruction set and the running data, including:

[0141] inputting the at least one preset operation and maintenance instruction set and the operation data of the plurality of time periods into the operation and maintenance large model;

[0142] For each time period, the operation and maintenance large model generates a corresponding first operation and maintenance instruction based on the operation data of the time period under the prompt of at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction features matching the operation data of the time period.

[0143] In an optional implementation, the generating, by the target large model, of the second operation and maintenance instruction of the target platform based on the system bias and the operation data of each of the plurality of time periods comprises:

[0144] inputting the system bias and the operation data of each of the plurality of time periods into the target large model;

[0145] determining, by the target large model, a time period weight corresponding to each time period according to the system bias corresponding to the time period;

[0146] generating, by the target large model, the second operation and maintenance instruction based on the time period weight and the operation data of each of the plurality of time periods.

[0147] In a third aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps of the artificial intelligence operation and maintenance method based on a large model.

[0148] In a fourth aspect, an embodiment of the present application provides a computer program product, which includes computer instructions. The computer instructions are executed by a processor to implement the steps of the artificial intelligence operation and maintenance method based on a large model.

[0149] In a fifth aspect, an embodiment of the present application provides a computer device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. The processor executes the computer program to implement the steps of the artificial intelligence operation and maintenance method based on a large model.

[0150] Referring to Figure 3 The computer device of the embodiment includes a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301, such as an artificial intelligence operation and maintenance program based on a large model. The processor 301 executes the computer program to implement the steps in each of the artificial intelligence operation and maintenance methods based on a large model, such as the steps S101-S105 shown in the figure. Figure 1

[0151] ​The computer program can be divided into one or more modules / units for example, which are stored in the memory 302 and executed by the processor 301 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the computer device.

[0152] The computer device can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The computer device can include, but is not limited to, the processor 301 and the memory 302. Those skilled in the art can understand that the schematic diagram is only an example of the computer device, and does not constitute a limitation on the computer device, and can include more or fewer components than the diagram, or combine certain components, or different components, for example, the computer device can also include an input / output device, a network access device, a bus and the like.

[0153] The processor 301 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor 301 can also be any conventional processor and the like, which is the control center of the computer device, and connects all parts of the computer device through various interfaces and lines.

[0154] The memory 302 can be used to store the computer programs and / or modules, and the processor 301 realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 302, and calling the data stored in the memory 302. The memory 302 can mainly include a program storage area and a data storage area, wherein the program storage area can store operating systems, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory 302 can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0155] The modules / units integrated in the computer device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor 301 executes the computer program, the steps of the above-mentioned various method embodiments can be realized. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording media, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0156] In summary, the embodiments of the present application have at least the following beneficial effects:

[0157] By adopting the embodiment of the application, running data of a target platform in multiple time periods is acquired; based on at least one preset operation and maintenance instruction set and the running data, a first operation and maintenance instruction corresponding to each of the multiple time periods is determined by using an operation and maintenance large model, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set includes at least one operation and maintenance instruction, the operation and maintenance instruction is adapted to indicate a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matched with the running data of the corresponding time period; based on the first operation and maintenance instruction, simulation operation and maintenance information of each of the multiple time periods is acquired; based on the simulation operation and maintenance information, the operation and maintenance large model is subjected to reinforcement learning to obtain a systematic bias corresponding to each of the multiple time periods, wherein the systematic bias corresponding to each time period represents a systematic error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period; based on the systematic bias corresponding to each of the multiple time periods and the running data, a second operation and maintenance instruction of the target platform is generated by using a target large model, and the second operation and maintenance instruction is sent to the target platform, so as to improve the automation efficiency of operation and maintenance and the rationality of an operation and maintenance scheme.

[0158] From the above description of the embodiments, those skilled in the art can clearly understand that the application can be implemented by means of software and a necessary hardware platform, and of course, can also be implemented entirely by hardware. Based on such an understanding, all or part of the technical solutions of the application that make contributions to the background art can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0159] The above is the preferred embodiment of the application. It should be noted that, for those skilled in the art, without departing from the principles of the application, a number of improvements and refinements can be made, which are also considered within the protection scope of the application.

Claims

1. A large model-based artificial intelligence operation and maintenance method, characterized in that, The method comprises: obtaining running data of a target platform at multiple time periods; based on at least one preset operation and maintenance instruction set and the running data, determining a first operation and maintenance instruction corresponding to each of the multiple time periods by using an operation and maintenance large model, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction feature, the preset operation and maintenance instruction set comprises at least one operation and maintenance instruction, the operation and maintenance instruction is adapted to indicate a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction feature matched with the running data of the corresponding time period; based on the first operation and maintenance instruction, obtaining simulation operation and maintenance information of each of the multiple time periods, comprising: for the first operation and maintenance instruction corresponding to each time period, based on the first operation and maintenance instruction, simulating by using an operation and maintenance simulation software previously set according to related parameters of the target platform to simulate running state information of the target platform under the action of the first operation and maintenance instruction, and generating simulation operation and maintenance information of the time period according to the running state information; based on the simulation operation and maintenance information, performing reinforcement learning on the operation and maintenance large model to obtain a systematic deviation corresponding to each of the multiple time periods, wherein the systematic deviation corresponding to each time period represents a systematic error of the target platform with respect to an operation and maintenance evaluation index under the action of the first operation and maintenance instruction corresponding to the time period; based on the systematic deviation corresponding to each of the multiple time periods and the running data, generating a second operation and maintenance instruction of the target platform by using a target large model, and sending the second operation and maintenance instruction to the target platform, wherein the target large model is obtained by fine-tuning the operation and maintenance large model through reinforcement learning; wherein, in the process of reinforcement learning, the learning weight of the operation and maintenance large model for the simulation operation and maintenance information corresponding to each time period is determined by the running state information corresponding to the time period; wherein, in the reinforcement learning stage, the operation and maintenance large model can observe feedback by using the simulation operation and maintenance information, and the observed feedback includes system stability and performance improvement; wherein, the learning weight includes an importance weight assigned to different simulation operation and maintenance information, and the running state information includes an application-specific index determined by the target platform; in the case that the target platform comprises a financial system, the application-specific index comprises a transaction success rate and / or a delay time; in the case that the target platform comprises an Internet of Things device, the application-specific index comprises sensor data accuracy and / or response time.

2. The method of claim 1, wherein, The reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information comprises: selecting at least part of the multiple time periods as enhancement time periods; for each enhancement time period, performing enhancement processing on the simulation operation and maintenance information of the enhancement time period to obtain enhancement operation and maintenance information of the enhancement time period; based on the enhancement operation and maintenance information and the simulation operation and maintenance information of the multiple time periods, performing reinforcement learning on the operation and maintenance large model.

3. The method of claim 2, wherein, The enhancement processing on the simulation operation and maintenance information of the enhancement time period to obtain the enhancement operation and maintenance information of the enhancement time period comprises: performing perturbation processing on the simulation operation and maintenance information of the enhancement time period to obtain perturbation operation and maintenance information; Randomly add noise to the disturbance operation and maintenance information to obtain noisy operation and maintenance information; Determine a first probability distribution of the noisy operation and maintenance information, and determine a second probability distribution of the simulation operation and maintenance information of the enhanced time period; Convert the noisy operation and maintenance information into enhanced operation and maintenance information of the enhanced time period based on the first probability distribution and the second probability distribution.

4. The method of claim 3, wherein, The conversion of the noisy operation and maintenance information into enhanced operation and maintenance information of the enhanced time period based on the first probability distribution and the second probability distribution comprises: Determine a target noise adding strategy by analyzing the difference between the first probability distribution and the second probability distribution; According to the target noise adding strategy, the noisy operation and maintenance information is added to obtain the enhanced operation and maintenance information of the enhanced time period.

5. The method of claim 1, wherein, The determination of the first operation and maintenance instruction corresponding to each time period of the plurality of time periods based on at least one preset operation and maintenance instruction set and the running data comprises: Input the at least one preset operation and maintenance instruction set and the running data of the plurality of time periods into the operation and maintenance large model; For each time period, the operation and maintenance large model generates a corresponding first operation and maintenance instruction based on at least part of the operation and maintenance instructions associated with the instruction characteristics matching the running data of the time period in the preset operation and maintenance instruction set.

6. The method of claim 1, wherein, The generation of the second operation and maintenance instruction of the target platform based on the systemic bias and running data corresponding to each time period of the plurality of time periods by using a target large model comprises: Input the systemic bias and running data corresponding to each time period of the plurality of time periods into the target large model; Determine the time period weight corresponding to each time period by the target large model according to the systemic bias corresponding to each time period; Generate the second operation and maintenance instruction based on the time period weight and running data corresponding to each time period of the plurality of time periods by the target large model.

7. A large model-based artificial intelligence operation and maintenance system, characterized in that, Comprise: The data acquisition module is used for acquiring the running data of the target platform in a plurality of time periods; The first operation and maintenance instruction module is used for determining the first operation and maintenance instruction corresponding to each time period of the plurality of time periods by using an operation and maintenance large model based on at least one preset operation and maintenance instruction set and the running data, wherein the at least one preset operation and maintenance instruction set is associated with at least one instruction characteristic, the preset operation and maintenance instruction set comprises at least one operation and maintenance instruction, the operation and maintenance instruction is suitable for indicating a platform operation and maintenance strategy, and the first operation and maintenance instruction is determined by at least part of the operation and maintenance instructions in the preset operation and maintenance instruction set associated with the instruction characteristics matching the running data of the corresponding time period; The simulation module is used for acquiring the simulation operation and maintenance information of each time period based on the first operation and maintenance instruction, comprising: for the first operation and maintenance instruction corresponding to each time period, simulating the running state information of the target platform under the action of the first operation and maintenance instruction by using the operation and maintenance simulation software previously set according to the related parameters of the target platform, and generating the simulation operation and maintenance information of the time period according to the running state information; The reinforcement learning module is configured to perform reinforcement learning on the operation and maintenance large model based on the simulation operation and maintenance information, so as to obtain a system bias corresponding to each time period, wherein the system bias corresponding to each time period represents a system error of the target platform with respect to an operation and maintenance evaluation index under the action of a first operation and maintenance instruction corresponding to the time period. The second operation and maintenance instruction module is configured to generate a second operation and maintenance instruction of the target platform based on the system bias corresponding to each time period and running data, and send the second operation and maintenance instruction to the target platform, wherein the target large model is obtained by fine-tuning the operation and maintenance large model via reinforcement learning. In the process of reinforcement learning, the learning weight of the operation and maintenance large model for the simulation operation and maintenance information corresponding to each time period is determined by running state information corresponding to the time period. In the reinforcement learning stage, the operation and maintenance large model can observe feedback by using the simulation operation and maintenance information, and the observed feedback includes system stability and performance improvement. The learning weight includes an importance weight assigned to different simulation operation and maintenance information, and the running state information includes application-specific indicators determined by the target platform. In the case where the target platform includes a financial system, the application-specific indicators include transaction success rate and / or delay time. In the case where the target platform includes an Internet of Things device, the application-specific indicators include sensor data accuracy and / or response time.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-6.

9. A computer program product comprising computer instructions, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1-6.

10. A computer device, comprising: The computer program is executed by the processor to implement the method of any one of claims 1-6.

Images