Automatic operation and maintenance method and system based on intelligent agent

Abstract

The invention discloses an intelligent agent-based automatic operation and maintenance method and system, and relates to the technical field of computer operation and maintenance, and the method comprises the steps: constructing an intelligent agent comprising a data collection layer, a state sensing layer, a decision-making layer and an execution layer, the data collection layer collects IT system multi-source data, the state sensing layer analyzes the state of the data sensing system, and the decision-making layer determines the state of the data sensing system; the decision-making layer generates an operation and maintenance instruction based on a sensing result decision, and the execution layer executes the instruction; relevant knowledge of the IT system is collected and sorted, and a knowledge graph is constructed; the intelligent agent is trained by using historical operation and maintenance data, the intelligent agent is enabled to try and explore in different environments through a reinforcement learning algorithm, the behavior of the intelligent agent is continuously adjusted according to a feedback result, and a decision model and an operation and maintenance strategy of the intelligent agent are optimized; and the intelligent agent monitors the IT system in real time, starts the data acquisition layer, the state sensing layer, the decision-making layer and the execution layer when detecting an abnormal event, and automatically executes corresponding operation and maintenance operation according to a decision-making result. According to the invention, stable and efficient operation of the IT system can be guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of computer operation and maintenance technology, and in particular to an intelligent agent-based automated operation and maintenance method and system. Background Art

[0002] With the rapid development of information technology, the scale of enterprise IT systems has expanded exponentially, and their architectures have become increasingly complex, encompassing multiple layers, from servers and storage devices at the physical layer to switches and routers at the network layer, and finally to various business software at the application layer. Traditional operations and maintenance methods rely primarily on manual operations, requiring operators to manually perform tasks such as troubleshooting, performance monitoring, and resource allocation. This approach is not only inefficient but also prone to human error, making it difficult to meet the high availability, high performance, and rapid response requirements of modern enterprises for IT systems.

[0003] In recent years, AI big models have achieved remarkable results in fields such as natural language processing and image recognition, and are gradually being introduced into the field of operations and maintenance. Some big-model-based operations and maintenance assistance tools can analyze operations and maintenance logs and documents, providing reference suggestions for troubleshooting. However, these applications still have significant limitations. Big models are primarily trained on large-scale, unsupervised data. When faced with highly specialized and scenario-based tasks such as IT operations and maintenance, they lack a deep understanding of specific system architectures and business logic, making it difficult to accurately grasp complex and changing system states. In the face of sudden failures or large-scale system changes, dynamic and accurate decision-making adjustments based on real-time system status and business needs are impossible, resulting in extended system downtime and severe business impacts. Furthermore, the black-box nature of big models makes it difficult for operations and maintenance personnel to understand their decision-making process, increasing uncertainty and risk.

[0004] Therefore, it is of great practical significance to develop an automated operation and maintenance method that has the ability of autonomous learning, accurate decision-making, and deep adaptation to IT operation and maintenance scenarios. Summary of the Invention

[0005] In response to the needs and shortcomings of current technological development, the present invention provides an automated operation and maintenance method and system based on an intelligent agent, which utilizes the autonomous learning and decision-making capabilities of the intelligent agent to achieve intelligent and automated operation and maintenance management of IT systems.

[0006] In the first aspect, the present invention provides an agent-based automated operation and maintenance method, which solves the above-mentioned technical problems using the following technical solutions:

[0007] An agent-based automated operation and maintenance method comprises the following steps:

[0008] S1. Build an intelligent agent based on machine learning and deep learning technologies. The intelligent agent consists of a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from IT systems, the state perception layer analyzes data to perceive system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions.

[0009] S2. Collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning during the decision-making process;

[0010] S3. Use historical operation and maintenance data to train the intelligent agent. Through reinforcement learning algorithms, the intelligent agent is allowed to try and explore in different environments. Based on the feedback, the intelligent agent continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is regularly updated to enable the intelligent agent to adapt to the continuous changes and development of the IT system.

[0011] S4, the intelligent agent monitors the IT system in real time. When an abnormal event is detected, it activates the data collection layer, status perception layer, decision layer, and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results;

[0012] S5. During the execution of operation and maintenance operations, the intelligent agent continuously monitors the effectiveness of the operations and makes dynamic adjustments based on actual conditions.

[0013] Optionally, the data collection layer uses various monitoring tools and API interfaces to collect performance indicators, log information, and configuration data from multiple sources of IT system servers, network devices, and application components, and stores the data in a time series database;

[0014] The state perception layer uses data analysis technology to process the collected data, perceive the system's operating status in real time, and identify abnormal patterns and potential risks in the system;

[0015] Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models, and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions;

[0016] The execution layer converts the instructions generated by the decision-making layer into actual operation and maintenance operations by calling system management tools and API interfaces to adjust and optimize the IT system.

[0017] Optionally, the data collection layer can integrate a multimodal data collection engine, support Prometheus and Zabbix monitoring protocols, implement structured / unstructured data cleaning through ETL tools, and establish a time series database to store real-time data.

[0018] The state perception layer is built on a graph neural network and uses a knowledge graph inference engine to implement multi-source data association analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction.

[0019] The decision-making layer adopts a dual-engine decision-making mechanism. Priority strategies are predefined through the rule engine, and an operation and maintenance decision model based on the PPO algorithm is constructed through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient, and resource optimization gain.

[0020] Adaptive operation adapters are designed for the execution layer to support infrastructure-as-code operations generated by Ansible Playbooks, cloud-native Kubernetes API calls, and application-layer hot fixes.

[0021] Optionally, execute step S2, first collect and organize relevant documents, operation and maintenance manuals and fault reports of the IT system, extract key information from them, then use the knowledge graph construction tool to convert the key information into nodes and edges, build an initial knowledge graph, and then verify the accuracy and completeness of the knowledge graph through manual review, and continuously update and improve the knowledge graph to enable it to reflect the latest status and changes of the IT system.

[0022] Optionally, step S3 is executed, and the training process of the agent includes:

[0023] Prepare historical operation and maintenance data, including system status data, operation and maintenance event data, and corresponding processing results;

[0024] Divide the prepared historical operation and maintenance data into a training set and a test set, and use the training set to train the agent. During the training process, set hyperparameters and adjust the structure and parameters of the machine learning model to improve the performance of the agent.

[0025] After training, the agent is evaluated using the test set and further optimized based on the evaluation results;

[0026] At the same time, new operation and maintenance data are collected regularly to conduct incremental training on the intelligent agent, enabling it to continuously learn and adapt to new situations.

[0027] In a second aspect, the present invention provides an agent-based automated operation and maintenance system, which solves the above-mentioned technical problems using the following technical solutions:

[0028] An agent-based automated operation and maintenance system, comprising:

[0029] The intelligent agent building module is used to build intelligent agents based on machine learning and deep learning technologies. The intelligent agent consists of a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from IT systems, the state perception layer analyzes data to perceive system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions.

[0030] The knowledge graph construction module is used to collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning in the decision-making process;

[0031] The training optimization module uses historical operation and maintenance data to train the intelligent agent. Using reinforcement learning algorithms, the intelligent agent is allowed to experiment and explore in different environments. Based on the feedback, the intelligent agent continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is regularly updated to enable the intelligent agent to adapt to the continuous changes and development of the IT system.

[0032] The intelligent agent is used to monitor IT systems in real time. When an abnormal event is detected, it starts the data collection layer, status perception layer, decision layer and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results. During the execution of the operation and maintenance operation, the intelligent agent continuously monitors the effect of the operation and dynamically adjusts it according to the actual situation.

[0033] Optionally, the data collection layer uses various monitoring tools and API interfaces to collect performance indicators, log information, and configuration data from multiple sources of IT system servers, network devices, and application components, and stores the data in a time series database;

[0034] The state perception layer uses data analysis technology to process the collected data, perceive the system's operating status in real time, and identify abnormal patterns and potential risks in the system;

[0035] Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models, and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions;

[0036] The execution layer converts the instructions generated by the decision-making layer into actual operation and maintenance operations by calling system management tools and API interfaces to adjust and optimize the IT system.

[0037] Optionally, the data collection layer can integrate a multimodal data collection engine, support Prometheus and Zabbix monitoring protocols, implement structured / unstructured data cleaning through ETL tools, and establish a time series database to store real-time data.

[0038] The state perception layer is built on a graph neural network and uses a knowledge graph inference engine to implement multi-source data association analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction.

[0039] The decision-making layer adopts a dual-engine decision-making mechanism. Priority strategies are predefined through the rule engine, and an operation and maintenance decision model based on the PPO algorithm is constructed through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient, and resource optimization gain.

[0040] Adaptive operation adapters are designed for the execution layer to support infrastructure-as-code operations generated by Ansible Playbooks, cloud-native Kubernetes API calls, and application-layer hot fixes.

[0041] Optionally, the knowledge graph construction modules involved include:

[0042] Information collection and extraction unit, used to collect and organize relevant documents, operation and maintenance manuals, and fault reports of IT systems and extract key information from them;

[0043] The knowledge graph construction unit is used to use the knowledge graph construction tool to convert key information into nodes and edges, build an initial knowledge graph, and manually review and verify the accuracy and completeness of the knowledge graph;

[0044] The knowledge graph update unit is used to continuously update and improve the knowledge graph so that it can reflect the latest status and changes of the IT system.

[0045] Optionally, the training optimization modules involved specifically include:

[0046] Data preparation unit, used to prepare historical operation and maintenance data, including system status data, operation and maintenance event data and corresponding processing results;

[0047] A data partitioning unit is used to divide the prepared historical operation and maintenance data into a training set and a test set;

[0048] The agent training unit is used to train the agent using the training set. During the training process, hyperparameters are set to adjust the structure and parameters of the machine learning model to improve the performance of the agent.

[0049] Evaluation and optimization unit, used to evaluate the agent using the test set, further optimize the agent based on the evaluation results, and output;

[0050] The periodic collection unit is used to regularly collect new operation and maintenance data and perform incremental training on the output intelligent agent so that it can continuously learn and adapt to new situations.

[0051] The present invention provides an agent-based automated operation and maintenance method and system, which has the following beneficial effects compared with the prior art:

[0052] 1. This invention utilizes the autonomous learning and decision-making capabilities of intelligent agents to achieve intelligent and automated operation and maintenance management of IT systems. It can perceive system status in real time, quickly and accurately handle various operation and maintenance events, improve operation and maintenance efficiency and quality, reduce operation and maintenance costs, and ensure the stable and efficient operation of IT systems.

[0053] 2. The intelligent agent of the present invention can monitor IT systems in real time, quickly respond to various operation and maintenance events, and automatically perform operation and maintenance operations, greatly shortening troubleshooting time and improving operation and maintenance efficiency. Through knowledge graphs and machine learning models, the intelligent agent can fully utilize historical data and domain knowledge to make decisions, avoiding the subjectivity and one-sidedness of human decision-making. When faced with complex operation and maintenance problems, the intelligent agent can comprehensively consider multiple factors and formulate more reasonable and effective solutions, thereby improving the accuracy and scientific nature of operation and maintenance decisions. The intelligent agent can promptly discover and resolve potential problems, avoid business losses caused by system failures, and reduce the company's operational risks and costs.

[0054] 3. The present invention can improve operation and maintenance efficiency, enhance decision-making accuracy, reduce operation and maintenance costs, improve system stability, and effectively solve the shortcomings of traditional operation and maintenance methods. It is not only suitable for the daily operation and maintenance management of various servers, network equipment and application software in traditional enterprise data centers, but also has broad application potential in distributed system operation and maintenance scenarios under emerging computing modes such as cloud computing and edge computing. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Attachment Figure 1 is a flow chart of a method according to embodiment 1 of the present invention;

[0056] Attachment Figure 2 This is a module connection block diagram of the second embodiment of the present invention;

[0057] Attachment Figure 3 is an architectural diagram of the intelligent agent in embodiments 1 and 2 of the present invention;

[0058] Attachment Figure 4 This is a workflow diagram of the intelligent agent in Examples 1 and 2 of the present invention. DETAILED DESCRIPTION

[0059] In order to make the technical solution, the technical problems solved and the technical effects of the present invention more clear, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0060] Example 1:

[0061] Combined with attachment Figure 1 、 34. This embodiment proposes an agent-based automated operation and maintenance method, which includes the following steps:

[0062] S1. Build an intelligent agent based on machine learning and deep learning technologies. The intelligent agent includes a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from the IT system, the state perception layer analyzes the data to perceive the system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions.

[0063] In this step, the data collection layer uses Prometheus and Zabbix monitoring tools and APIs to collect performance metrics, log information, and configuration data from multiple sources across IT system servers, network devices, and application components, and stores this data in the InfluxDB distributed database. Specifically, the data collection layer integrates a multimodal data collection engine, supports the Prometheus and Zabbix monitoring protocols, cleans structured and unstructured data using ETL tools, and establishes an InfluxDB distributed database to store real-time data.

[0064] The state perception layer uses data analysis techniques (such as cluster analysis and anomaly detection algorithms) to process collected data, perceive the system's operating status in real time, and identify abnormal patterns and potential risks. Specifically, the state perception layer is built on a graph neural network and uses a knowledge graph inference engine to perform multi-source data correlation analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction.

[0065] Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models (such as neural networks and decision trees), and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions. Specifically, the decision-making layer adopts a dual-engine decision-making mechanism, pre-defining priority strategies through the rule engine and building an operation and maintenance decision model based on the PPO algorithm through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient, and resource optimization gain.

[0066] The execution layer converts the instructions generated by the decision layer into actual operations and maintenance by invoking system management tools (such as Ansible and SaltStack) and API interfaces, adjusting and optimizing IT systems. Specifically, the execution layer designs adaptive operation adapters to support infrastructure-as-code operations generated by Ansible Playbooks, cloud-native Kubernetes API calls, and application-layer hotfixes.

[0067] S2. Collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning during the decision-making process. This process specifically includes:

[0068] First, collect and organize relevant documents, operation and maintenance manuals, and fault reports of the IT system, and extract key information (such as component names, attributes, relationships, etc.);

[0069] Then, a knowledge graph construction tool (such as Neo4j) is used to convert key information into nodes and edges to construct an initial knowledge graph, which is then manually reviewed to verify its accuracy and completeness.

[0070] Finally, by continuously updating and improving the knowledge graph, it can reflect the latest status and changes of the IT system.

[0071] S3. Use historical operation and maintenance data to train the intelligent agent. Through reinforcement learning algorithms, the intelligent agent is allowed to try and explore in different environments. Based on the feedback results, it continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is updated regularly so that the intelligent agent can adapt to the continuous changes and development of the IT system.

[0072] The training process of the agent includes:

[0073] Prepare historical operation and maintenance data, including system status data, operation and maintenance event data, and corresponding processing results;

[0074] Divide the prepared historical operation and maintenance data into a training set and a test set, and use the training set to train the agent. During the training process, set hyperparameters and adjust the structure and parameters of the machine learning model to improve the performance of the agent.

[0075] After training, the agent is evaluated using the test set and further optimized based on the evaluation results;

[0076] At the same time, new operation and maintenance data are collected regularly to conduct incremental training on the intelligent agent, enabling it to continuously learn and adapt to new situations.

[0077] S4, the intelligent agent monitors the IT system in real time. When an abnormal event is detected, it activates the data collection layer, status perception layer, decision layer, and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results;

[0078] S5. During the execution of operation and maintenance operations, the intelligent agent continuously monitors the effectiveness of the operations and makes dynamic adjustments based on actual conditions.

[0079] Suppose the intelligent agent monitors the IT system in real time and detects that the memory usage of a server continues to exceed 80% and the application responds slowly. It starts the data collection layer, state perception layer, decision layer and execution layer in sequence. The data collection layer quickly collects the relevant performance indicators and log information of the server. The state perception layer analyzes this data and determines the possible memory leak problem of the server. The decision layer makes a decision based on the predefined operation and maintenance strategy, deep learning model and knowledge graph, and decides to try to clear the server cache first to free up memory. The execution layer calls the corresponding command line tool or API interface to perform the cache clearing operation.

[0080] After the operation is complete, the agent continues to monitor the server's status. If memory usage decreases and application responsiveness returns to normal, the operation is considered successful. If the problem persists, the agent will further analyze the issue and try other solutions, such as adjusting the application's memory allocation parameters or restarting the application.

[0081] Example 2:

[0082] Combined with attachment Figure 2 、 3 4. This embodiment proposes an agent-based automated operation and maintenance system, which includes:

[0083] The intelligent agent building module is used to build intelligent agents based on machine learning and deep learning technologies. The intelligent agent consists of a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from IT systems, the state perception layer analyzes data to perceive system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions.

[0084] The knowledge graph construction module is used to collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning in the decision-making process;

[0085] The training optimization module uses historical operation and maintenance data to train the intelligent agent. Using reinforcement learning algorithms, the intelligent agent is allowed to experiment and explore in different environments. Based on the feedback, the intelligent agent continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is regularly updated to enable the intelligent agent to adapt to the continuous changes and development of the IT system.

[0086] The intelligent agent is used to monitor IT systems in real time. When an abnormal event is detected, it starts the data collection layer, status perception layer, decision layer and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results. During the execution of the operation and maintenance operation, the intelligent agent continuously monitors the effect of the operation and dynamically adjusts it according to the actual situation.

[0087] In this embodiment, the data collection layer uses Prometheus and Zabbix monitoring tools and APIs to collect multi-source data, including performance metrics, log information, and configuration data, from servers, network devices, and application components in the IT system. This data is then stored in the InfluxDB distributed database. Specifically, the data collection layer integrates a multimodal data collection engine, supports the Prometheus and Zabbix monitoring protocols, cleans structured and unstructured data using ETL tools, and establishes an InfluxDB distributed database to store real-time data.

[0088] The state perception layer uses data analysis techniques (such as cluster analysis and anomaly detection algorithms) to process collected data, perceive the system's operating status in real time, and identify abnormal patterns and potential risks. Specifically, the state perception layer is built on a graph neural network and uses a knowledge graph inference engine to perform multi-source data correlation analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction.

[0089] Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models (such as neural networks and decision trees), and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions. Specifically, the decision-making layer adopts a dual-engine decision-making mechanism, pre-defining priority strategies through the rule engine and building an operation and maintenance decision model based on the PPO algorithm through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient, and resource optimization gain.

[0090] The execution layer converts the instructions generated by the decision layer into actual operations and maintenance by invoking system management tools (such as Ansible and SaltStack) and API interfaces, adjusting and optimizing IT systems. Specifically, the execution layer designs adaptive operation adapters to support infrastructure-as-code operations generated by Ansible Playbooks, cloud-native Kubernetes API calls, and application-layer hotfixes.

[0091] In this embodiment, the knowledge graph construction modules involved specifically include:

[0092] Information collection and extraction unit, used to collect and organize relevant documents, operation and maintenance manuals, and fault reports of IT systems, and extract key information (such as component names, attributes, relationships, etc.);

[0093] The knowledge graph construction unit is used to convert key information into nodes and edges using knowledge graph construction tools (such as Neo4j), build an initial knowledge graph, and manually review and verify the accuracy and completeness of the knowledge graph;

[0094] The knowledge graph update unit is used to continuously update and improve the knowledge graph so that it can reflect the latest status and changes of the IT system.

[0095] In this embodiment, the training optimization module involved specifically includes:

[0096] Data preparation unit, used to prepare historical operation and maintenance data, including system status data, operation and maintenance event data and corresponding processing results;

[0097] A data partitioning unit is used to divide the prepared historical operation and maintenance data into a training set and a test set;

[0098] The agent training unit is used to train the agent using the training set. During the training process, hyperparameters are set to adjust the structure and parameters of the machine learning model to improve the performance of the agent.

[0099] Evaluation and optimization unit, used to evaluate the agent using the test set, further optimize the agent based on the evaluation results, and output;

[0100] The periodic collection unit is used to regularly collect new operation and maintenance data and perform incremental training on the output intelligent agent so that it can continuously learn and adapt to new situations.

[0101] Based on the automated operation and maintenance system of this embodiment, it is assumed that the intelligent agent monitors the IT system in real time and detects that the memory usage of a server continues to exceed 80% and the application responds slowly. The data collection layer, state perception layer, decision layer and execution layer are started in sequence. The data collection layer quickly collects the relevant performance indicators and log information of the server. The state perception layer analyzes this data to determine the possible memory leak problem of the server. The decision layer makes a decision based on the predefined operation and maintenance strategy, deep learning model and knowledge graph, and decides to try to clear the server's cache first to release memory. The execution layer calls the corresponding command line tool or API interface to perform the cache clearing operation.

[0102] After the operation is complete, the agent continues to monitor the server's status. If memory usage decreases and application responsiveness returns to normal, the operation is considered successful. If the problem persists, the agent will further analyze the issue and try other solutions, such as adjusting the application's memory allocation parameters or restarting the application.

[0103] In summary, the intelligent agent-based automated operation and maintenance method and system of the present invention effectively solves many problems existing in traditional operation and maintenance methods by constructing intelligent agents, knowledge graphs, and perfect training optimization and operation and maintenance processes. It has significant technical advantages and practical application value, can provide efficient, intelligent and reliable solutions for the enterprise's IT system operation and maintenance, promote the development of the field of computer operation and maintenance technology, and has broad application prospects and promotion value.

[0104] The above specific examples are used to illustrate the principles and implementation methods of the present invention in detail. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by those skilled in the art without departing from the principles of the present invention should fall within the scope of patent protection of the present invention.

Claims

1. An automated operation and maintenance method based on an intelligent agent, characterized in that: The steps include: S1. Build an intelligent agent based on machine learning and deep learning technologies. The intelligent agent consists of a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from IT systems, the state perception layer analyzes data to perceive system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions. S2. Collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning during the decision-making process; S3. Use historical operation and maintenance data to train the intelligent agent. Through reinforcement learning algorithms, the intelligent agent is allowed to try and explore in different environments. Based on the feedback, the intelligent agent continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is regularly updated to enable the intelligent agent to adapt to the continuous changes and development of the IT system. S4, the intelligent agent monitors the IT system in real time. When an abnormal event is detected, it activates the data collection layer, status perception layer, decision layer, and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results; S5. During the execution of operation and maintenance operations, the intelligent agent continuously monitors the effectiveness of the operations and makes dynamic adjustments based on actual conditions.

2. The agent-based automated operation and maintenance method according to claim 1, characterized in that: The data collection layer uses various monitoring tools and API interfaces to collect performance indicators, log information and configuration data from multiple sources of IT system servers, network devices and application components, and stores the data in a time series database; The state perception layer uses data analysis technology to process the collected data, perceive the operating status of the system in real time, and identify abnormal patterns and potential risks in the system; Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions; The execution layer converts the instructions generated by the decision-making layer into actual operation and maintenance operations by calling system management tools and API interfaces, and adjusts and optimizes the IT system.

3. The agent-based automated operation and maintenance method according to claim 2, characterized in that: The data collection layer integrates a multimodal data collection engine, supports Prometheus and Zabbix monitoring protocols, implements structured / unstructured data cleaning through ETL tools, and establishes a time series database to store real-time data; The state perception layer is built based on a graph neural network and uses a knowledge graph reasoning engine to implement multi-source data association analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction. The decision-making layer adopts a dual-engine decision-making mechanism, pre-defining priority strategies through the rule engine and building an operation and maintenance decision model based on the PPO algorithm through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient and resource optimization gain; The execution layer designs adaptive operation adapters to support infrastructure-as-code operations generated by Ansible Playbook, cloud-native Kubernetes API calls, and application-layer hot fixes.

4. The agent-based automated operation and maintenance method according to claim 1, characterized in that: Execute step S2. First, collect and organize relevant documents, operation and maintenance manuals, and fault reports of the IT system, extract key information from them, and then use the knowledge graph construction tool to convert the key information into nodes and edges to build an initial knowledge graph. Then, verify the accuracy and completeness of the knowledge graph through manual review, and continuously update and improve the knowledge graph to enable it to reflect the latest status and changes of the IT system.

5. The agent-based automated operation and maintenance method according to claim 1, characterized in that: Execute step S3, the training process of the agent includes: Prepare historical operation and maintenance data, including system status data, operation and maintenance event data, and corresponding processing results; Divide the prepared historical operation and maintenance data into a training set and a test set, and use the training set to train the agent. During the training process, set hyperparameters and adjust the structure and parameters of the machine learning model to improve the performance of the agent. After training, the agent is evaluated using the test set and further optimized based on the evaluation results; At the same time, new operation and maintenance data are collected regularly to conduct incremental training on the intelligent agent, enabling it to continuously learn and adapt to new situations.

6. An automated operation and maintenance system based on an intelligent agent, characterized in that: It includes: The intelligent agent building module is used to build intelligent agents based on machine learning and deep learning technologies. The intelligent agent consists of a data acquisition layer, a state perception layer, a decision-making layer, and an execution layer. The data acquisition layer collects multi-source data from IT systems, the state perception layer analyzes data to perceive system status, the decision-making layer generates operation and maintenance instructions based on the perception results, and the execution layer executes the instructions. The knowledge graph construction module is used to collect and organize IT system-related knowledge, build a knowledge graph, and provide background knowledge for intelligent agents to make decisions, enabling them to make more comprehensive and in-depth reasoning in the decision-making process; The training optimization module uses historical operation and maintenance data to train the intelligent agent. Using reinforcement learning algorithms, the intelligent agent is allowed to experiment and explore in different environments. Based on the feedback, the intelligent agent continuously adjusts its behavior and optimizes its decision-making model and operation and maintenance strategy. At the same time, the training data is regularly updated to enable the intelligent agent to adapt to the continuous changes and development of the IT system. The intelligent agent is used to monitor IT systems in real time. When an abnormal event is detected, it starts the data collection layer, status perception layer, decision layer and execution layer in sequence, and automatically performs corresponding operation and maintenance operations based on the decision results. During the execution of the operation and maintenance operation, the intelligent agent continuously monitors the effect of the operation and dynamically adjusts it according to the actual situation.

7. The agent-based automated operation and maintenance system according to claim 6, characterized in that: The data collection layer uses various monitoring tools and API interfaces to collect performance indicators, log information and configuration data from multiple sources of IT system servers, network devices and application components, and stores the data in a time series database; The state perception layer uses data analysis technology to process the collected data, perceive the operating status of the system in real time, and identify abnormal patterns and potential risks in the system; Based on the recognition results of the state perception layer, the decision-making layer uses predefined operation and maintenance strategies, deep learning models and knowledge graphs to make decisions and generate corresponding operation and maintenance instructions; The execution layer converts the instructions generated by the decision-making layer into actual operation and maintenance operations by calling system management tools and API interfaces, and adjusts and optimizes the IT system.

8. The agent-based automated operation and maintenance system according to claim 7, characterized in that: The data collection layer integrates a multimodal data collection engine, supports Prometheus and Zabbix monitoring protocols, implements structured / unstructured data cleaning through ETL tools, and establishes a time series database to store real-time data; The state perception layer is built based on a graph neural network and uses a knowledge graph reasoning engine to implement multi-source data association analysis, time series anomaly detection based on a hidden Markov model, and fault propagation path prediction. The decision-making layer adopts a dual-engine decision-making mechanism, pre-defining priority strategies through the rule engine and building an operation and maintenance decision model based on the PPO algorithm through the reinforcement learning engine. The reward function of the operation and maintenance decision model includes fault recovery time, operation risk coefficient and resource optimization gain; The execution layer designs adaptive operation adapters to support infrastructure-as-code operations generated by Ansible Playbook, cloud-native Kubernetes API calls, and application-layer hot fixes.

9. The agent-based automated operation and maintenance system according to claim 6, characterized in that: The knowledge graph construction module specifically includes: Information collection and extraction unit, used to collect and organize relevant documents, operation and maintenance manuals, and fault reports of IT systems and extract key information from them; The knowledge graph construction unit is used to use the knowledge graph construction tool to convert key information into nodes and edges, build an initial knowledge graph, and manually review and verify the accuracy and completeness of the knowledge graph; The knowledge graph update unit is used to continuously update and improve the knowledge graph so that it can reflect the latest status and changes of the IT system.

10. The agent-based automated operation and maintenance system according to claim 6, characterized in that: The training optimization module specifically includes: Data preparation unit, used to prepare historical operation and maintenance data, including system status data, operation and maintenance event data and corresponding processing results; A data partitioning unit is used to divide the prepared historical operation and maintenance data into a training set and a test set; The agent training unit is used to train the agent using the training set. During the training process, hyperparameters are set to adjust the structure and parameters of the machine learning model to improve the performance of the agent. Evaluation and optimization unit, used to evaluate the agent using the test set, further optimize the agent based on the evaluation results, and output; The periodic collection unit is used to regularly collect new operation and maintenance data and perform incremental training on the output intelligent agent so that it can continuously learn and adapt to new situations.

Images