An AI agent-based enterprise operation and maintenance method, device and medium

Abstract

The application discloses an enterprise operation and maintenance method and device based on an AI agent, and a medium, relates to the field of artificial intelligence, and comprises the following steps: receiving a business processing request of a target business system, obtaining business problem description information, extracting key entity information and intention information; generating business problem context data based on the key entity information and the intention information through a pre-constructed business operation correlation graph; collecting system state data, fusing the business problem context data and the system state data, and generating a system panoramic state vector; generating an operation operation instruction sequence through a pre-trained reinforcement learning agent, sending the operation operation instruction sequence to a safe execution engine, and executing the operation operation; and adjusting model parameters of the reinforcement learning agent according to current system state data of the target business system. The method effectively reduces the interference on the normal operation of the business system in the operation and maintenance process, and improves the timeliness of the operation and maintenance response and the accuracy of the operation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, specifically to an enterprise operation and maintenance method, equipment, and medium based on AI intelligent agents. Background Technology

[0002] Intelligent operation and maintenance is a core component of modern enterprise digital management. It involves comprehensive monitoring, analysis, and optimization of business system operation status, business process efficiency, and business outcome data, aiming to ensure business continuity and promote business growth through technological means. As enterprise-level application software systems become increasingly functional, system complexity increases exponentially, and system operation and maintenance work becomes increasingly complex. Consequently, the application complexity of key operation and maintenance systems also continues to rise.

[0003] Currently, enterprises typically combine data monitoring dashboards with predefined automated processes to monitor and respond to common business scenarios. However, this approach often separates technical indicator monitoring from business operational decision-making. When operational users face sudden alerts or other emergencies, it becomes difficult to accurately and promptly locate, analyze, and handle the issues, leading to a escalating impact. Furthermore, operational decisions made through predefined automated processes lack the ability to perceive and adapt to dynamically changing business environments, making it difficult to generate real-time optimal strategies. In addition, the lack of quantitative assessment methods for the potential business impact and risks of operational actions results in highly unpredictable and risky decision-making processes, failing to support the needs of refined business operations. Summary of the Invention

[0004] To address the aforementioned issues, this application proposes an enterprise operation and maintenance method based on AI intelligent agents, comprising:

[0005] Receive a business processing request from the target business system, obtain business problem description information, parse the business problem description information, and extract key entity information and intent information;

[0006] Business problem context data is generated based on the key entity information and the intent information through a pre-constructed business operation association graph; the business operation association graph contains business entity nodes and operation entity nodes that are associated in a graph structure in the target business system;

[0007] The target business system is monitored in real time, system status data is collected, and the business problem context data and system status data are fused to generate a system panoramic status vector.

[0008] Based on the system's panoramic state vector, an operational operation instruction sequence is generated through a pre-trained reinforcement learning agent. The operational operation instruction sequence is then sent to the security execution engine, which executes the operational operation corresponding to the operational operation instruction sequence.

[0009] After the sequence of operational instructions is executed, the model parameters of the reinforcement learning agent are adjusted according to the current system status data of the target business system.

[0010] On the other hand, this application also proposes an enterprise operation and maintenance equipment based on AI intelligent agents, including:

[0011] At least one processor; and,

[0012] A memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform, for example, an enterprise operation and maintenance method based on an AI intelligent agent as described in the above example.

[0014] On the other hand, this application also proposes a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: an enterprise operation and maintenance method based on an AI intelligent agent as described in the above example.

[0015] The enterprise operation and maintenance method based on AI intelligent agents proposed in this application can bring the following beneficial effects:

[0016] By analyzing business issues to extract key entities and intents, and combining this with a business operation relationship graph to generate contextual data, maintenance actions can be closely aligned with business objectives, avoiding blind operations divorced from the business scenario and ensuring that every maintenance action matches the core business needs of the enterprise. The panoramic state vector formed by real-time fusion of business issue context and system status data allows the reinforcement learning agent to fully grasp the system dynamics, and the generated operational operation instruction sequence is more in line with the actual system operation status. At the same time, the intervention of the security execution engine can ensure the compliance and reliability of maintenance operations, effectively reduce interference with the normal operation of business systems during maintenance, improve the timeliness of maintenance response and the accuracy of operations, and deeply align with the enterprise's business logic and system operation characteristics, providing precise and targeted support for operations and maintenance.

[0017] By leveraging the mechanism of reinforcement learning agents to dynamically adjust model parameters based on operational performance, these agents can continuously adapt to changes in enterprise business systems over long-term applications. Whether it's business process updates or system architecture upgrades, the agents can gradually optimize decision-making logic, forming operational capabilities that are synchronized with enterprise development. Furthermore, automated operational processes reduce the uncertainty caused by manual intervention, freeing up operational personnel to focus on more important system planning and business support work. At the same time, stable operational performance improves the operational continuity of target business systems, providing a solid guarantee for the smooth operation of enterprise business. This helps enterprises maintain an efficient and stable development trend in digital operations, endowing enterprise operations and maintenance with continuous optimization capabilities and long-term value. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 This is a flowchart illustrating an enterprise operation and maintenance method based on an AI intelligent agent, as described in an embodiment of this application.

[0020] Figure 2 This is a schematic diagram of an enterprise operation and maintenance device based on an AI intelligent agent, as described in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] like Figure 1 As shown in the figure, this application provides an enterprise operation and maintenance method based on AI intelligent agents, including:

[0024] S101. Receive a business processing request from the target business system, obtain business problem description information, parse the business problem description information, and extract key entity information and intent information.

[0025] Specifically, through user input or automatic system capture, business processing requests from the target business system are received, and business problem description information is obtained. Business problem description information refers to natural language text or structured data that describes business anomalies or optimization targets.

[0026] Furthermore, natural language processing is used to parse the business problem description information to extract key entity information and intent information. Key entity information refers to core words with specific business or technical meanings identified from the business problem description information, such as "order", "user", "payment service", etc. Intent information refers to the purpose that the user hopes to achieve or the type of problem that needs to be solved, such as "the reason for the decline in diagnosis success rate" or "improving user activity".

[0027] S102. Based on the key entity information and the intent information, business problem context data is generated by using a pre-constructed business operation association graph. The business operation association graph contains business entity nodes and operation entity nodes that are associated in a graph structure in the target business system.

[0028] Load the pre-built business operation association graph stored in the graph database. The graph stores business entities and operation entities as nodes and the association relationships between entities as edges.

[0029] Based on key entity and intent information, the corresponding nodes are matched in the graph using graph query language. The direct and indirect relationships between nodes are traversed to integrate business problem context data, which includes information such as the business links involved in the problem, related operations, and pre- and post-processes. This data is then stored in a structured format.

[0030] In this embodiment, the specific process for generating business problem context data is as follows: Based on key entity information, the corresponding entity node in the graph is located using graph query language, and a traversal operation is performed. First, the directly related nodes and connecting edge information of the node are obtained; then, the indirectly related nodes are recursively traversed until the traversal depth reaches a preset threshold. All related entity nodes, operation nodes, node attributes, and relationship types are integrated into a related data set and temporarily stored in a graph structure.

[0031] Furthermore, based on the intent information, the relevance of each node and relationship in the associated data set is calculated. Specifically, the intent information and text attributes in the associated data are first converted into vectors, and then the relevance score between the vectors is calculated using a cosine similarity algorithm. Subsequently, low-relevance information is filtered out according to a preset threshold, and the remaining associated data is sorted from high to low score. Finally, the sorted associated entities, operation nodes, and core relationships are reorganized into a subgraph structure, retaining only the business links highly related to the intent.

[0032] Furthermore, the related subgraphs, key entity information, and intent information are integrated into structured data. Specifically, this is encapsulated using a preset format (such as JSON), where the key entity field stores the entities and their attributes extracted in the first step; the intent information field stores the intent type and a detailed description; and the related subgraph field stores the nodes and relationships within the subgraph in a nested structure. After encapsulation, this structured data is output as business problem context data for subsequent integration with system status data.

[0033] In this embodiment, the process of constructing the business operation association graph is as follows: Business entity objects and operation entity objects are identified and obtained from multi-source data of the target business system. Specifically, business entity pairs, such as entities with business attributes like "order," "user," and "inventory," are extracted by scanning the system metadata database (e.g., database table structure, data dictionary), including information such as name, ID, and attribute fields. Operation entity objects, such as operations with execution logic like "create order," "deduct inventory," and "initiate payment," are extracted by parsing system interface documents, service call logs, and process definition files, including information such as operation name, input / output parameters, and calling method.

[0034] The analysis examines the relationships between two types of entities, including data flow relationships, functional dependencies, business affiliation relationships, and performance impact relationships. Specifically, data flow relationships are determined by tracking data transmission logs; functional dependencies are determined by parsing operation dependency configurations; business affiliation relationships are determined by matching against a business rule base; and performance impact relationships are determined by performance monitoring data. The identified relationships are then categorized and labeled according to these four types.

[0035] Based on business entity objects and operational entity objects, they are transformed into nodes in a graph. A unique node is created for each entity object, and node attributes include entity type, identifier ID, core attributes, etc. According to the relationship type, connection edges are added between nodes: data flow relationship corresponds to "data flow" edge; functional dependency relationship corresponds to "dependency" edge; business affiliation relationship corresponds to "affiliation" edge; and performance impact relationship corresponds to "impact" edge.

[0036] The complete structure of nodes and edges is stored in a graph database. The indexing mechanism of the graph database is used to optimize query efficiency and form a dynamically updatable business operation relationship graph, providing structured relationship support for subsequent business problem analysis.

[0037] S103. Monitor the target business system in real time, collect system status data, and fuse the business problem context data and the system status data to generate a system panoramic status vector.

[0038] Specifically, monitoring agents deployed in the target business system collect system status data in real time. Hardware metrics include server CPU utilization, memory usage, and disk I / O; software metrics include application response time, database connection count, and API call success rate. The data is stored in a time-series database in time-series format.

[0039] The business problem context data and the real-time collected system status data are aligned by timestamp and then fused into a one-dimensional vector through feature splicing, weight allocation, and other methods to obtain a system panoramic status vector. The vector dimension covers key features at both the business and technical levels.

[0040] S104. Based on the system panoramic state vector, an operational operation instruction sequence is generated through a pre-trained reinforcement learning agent. The operational operation instruction sequence is then sent to the security execution engine, which executes the operational operation corresponding to the operational operation instruction sequence.

[0041] Specifically, the generated system panoramic state vector is input into a pre-trained reinforcement learning agent model. The reinforcement learning agent first calls its internal action space loading module to read the definitions of all executable operations from a pre-set operational operation metadata database. The metadata database stores standardized operations in enterprise operation and maintenance scenarios, with each operation containing metadata such as a unique identifier, operation type, and applicable scenario. The agent then extracts a set of these operations through its interface to form the operational operation action space, serving as the range of options for subsequent decisions.

[0042] Furthermore, the policy network (such as a Transformer- or CNN-based neural network) in the reinforcement learning agent is invoked to perform forward computation using the system's panoramic state vector as input. The policy network, through multi-layer feature extraction and non-linear transformation, outputs an expected value score for each executable operation in the action space. This expected value score, based on historical training experience, reflects the probability and effectiveness of solving the problem after performing the operation in the current system state. The scores of all operations are compared, and the operation with the highest expected value score is selected as the candidate operation for the current step.

[0043] Furthermore, based on the unique identifier of the candidate operational operation, a precise query is performed in the preset operational operation metadata database to obtain the complete instruction definition and required execution parameters corresponding to the operation. Following the interface specifications of the target business system, the instruction definition and parameters are combined into a directly executable structured instruction, which is then added to the end of the operational operation instruction sequence being constructed.

[0044] The latest status data of the target business system is collected through real-time monitoring tools. Combined with the current business problem context, a new system panoramic status vector is regenerated. The new vector is then input into the reinforcement learning agent, and the above steps are repeated to generate new candidate operational instructions and add them to the sequence. Simultaneously, termination conditions are continuously monitored: if the system status data shows that the core indicators have converged to the normal range, or if the instruction sequence length reaches the preset maximum threshold, the iteration stops, and the final operational instruction sequence is output.

[0045] In this embodiment, the construction process of the reinforcement learning agent is as follows: determine the core business objective corresponding to the business processing request, assign corresponding weight coefficients to the core business objective, construct a multi-objective linear weighted reward function, construct an action space based on all executable operation operations in the preset operation metadata database, initialize the reinforcement learning agent architecture based on a deep neural network, add the multi-objective linear weighted reward function and the action space to the reinforcement learning agent architecture, and construct the reinforcement learning agent.

[0046] This involves retrieving all executable operational operations from the preset operational operation metadata database, creating corresponding operation identifiers for each executable operational operation, and adding metadata descriptions. The metadata descriptions include the operation name, operation type, target system interface, call parameter template, required permissions, estimated execution time, a list of expected impact business metrics, and risk level labels.

[0047] The pre-training process of the reinforcement learning agent is as follows: An experience pool is constructed using structured experience data (including fault scenarios, operation sequences, and execution effects) transformed from historical operation and maintenance logs, and diverse test scenarios (including typical problems such as high load, abnormal links, and data inconsistency) generated by the business system simulator. Training is advanced iteratively. In each iteration, the agent samples input from the state space based on the current policy, selects actions using an ε-greedy policy, and executes them in the simulation environment or test system. The agent obtains the new state and reward value after execution and stores the current state, action, reward, and new state in the experience pool. When the experience pool reaches a certain data volume, batches of experience are randomly sampled. The prediction error of the value network is calculated using a temporal difference algorithm, and the parameters are updated using gradient descent. The policy network parameters are then optimized based on the policy gradient theorem. Simultaneously, the parameters of the target network and the current value network are periodically synchronized to avoid training oscillations. Indicators such as average reward value and problem resolution rate are continuously monitored during the iteration process. When the indicators converge to a preset threshold for multiple consecutive rounds or the number of iterations reaches the upper limit, training stops, and the optimal model parameters are saved, forming a reinforcement learning agent adaptable to actual operation and maintenance scenarios.

[0048] Furthermore, the operational operations corresponding to the sequence of operational operation instructions are executed through the secure execution engine.

[0049] The secure execution engine receives a sequence of operational instructions generated by the reinforcement learning agent through a pre-defined interface and reads the instructions sequentially. For each instruction, it performs multi-dimensional verification based on a pre-defined operational metadata database. This verification includes checking whether the instruction format conforms to the specifications defined in the metadata database; whether the operation permissions match the access control policies of the target business system; and whether the execution conditions meet the preconditions recorded in the metadata database. If all verification items pass, the instruction is marked as "executable." If it fails, the reason for the verification failure is recorded, the execution of the instruction is terminated, and an exception message is sent back to the agent.

[0050] For verified operational instructions, the security execution engine invokes the mapping rule engine to read the corresponding mapping rule from the metadata database. The rule defines the conversion logic between standardized instructions and target application programming interfaces (APIs). Based on the mapping rule, the engine fills the parameters from the instruction into the API request template, generating a structured call request that conforms to the API protocol, including the API address, request headers, and request body. The engine then sends this request to the corresponding API via the network module.

[0051] After receiving a call request, the application programming interface (API) executes the corresponding operation according to its built-in logic. The secure execution engine monitors the operation execution process in real time through an API callback mechanism or polling: it continuously acquires the execution status codes returned by the API and parses the returned result data. The engine records the status codes and result data to the execution log with timestamps, and simultaneously checks for exceptions. If an exception occurs, a preset fault tolerance mechanism is immediately triggered; if execution is successful, it continues processing the next instruction in the instruction sequence until all instructions have been executed.

[0052] S105: After the sequence of operational instructions is executed, the model parameters of the reinforcement learning agent are adjusted according to the current system status data of the target business system.

[0053] Specifically, monitoring components deployed in the target business system collect current system status data after the execution of operational command sequences. Operational result data focuses on the business level, including whether business problems have been resolved, whether business processes have returned to normal, and key business metric values; performance result data focuses on the technical level, covering hardware resource metrics and software performance metrics. The collected data is stored categorized by timestamp and metric type, forming a structured status dataset.

[0054] The current system status data is aligned with the system status data recorded before execution, and the numerical difference of specific indicators under each dimension is calculated. For example, if the "number of abnormal order statuses" was 5 before execution and 0 after execution, the change in this indicator in the business dimension is -5; if the "number of database connections" was 1000 (exceeding the threshold) before execution and 300 after execution, the change in this indicator in the technical dimension is -700. The changes in all dimensions are integrated into an actual change matrix to quantitatively characterize the impact of the operation sequence on the system status.

[0055] Based on the reinforcement learning sample format, relevant data is encapsulated into new training samples. Specifically, the samples contain three core pieces of information: first, a system panoramic state vector (i.e., the overall system state before execution, serving as the "initial state"); second, an operational instruction sequence (the "action sequence" executed by the agent); and third, an actual change matrix. Simultaneously, metadata such as execution time and business scenario labels are appended to the samples to enhance their scenario adaptability, ultimately forming structured training sample data points that are stored in the agent's experience pool.

[0056] Newly constructed training sample data points are extracted from the experience pool. For the policy network, the matching degree between action sequences and actual feedback is calculated using policy gradient algorithms (such as PPO), and network weights are adjusted to strengthen the selection probability of high-feedback action sequences. For the value network, state value estimation is optimized based on temporal difference algorithms (such as TD-learning) to make the predicted value closer to the cumulative reward of actual feedback. During parameter updates, training stability is ensured by setting mechanisms such as learning rate decay and gradient pruning. After the update is completed, the new model parameters are saved, enabling the agent to generate operation instruction sequences that better meet actual needs in subsequent similar scenarios, achieving continuous iterative optimization.

[0057] By analyzing business issues to extract key entities and intents, and combining this with a business operation relationship graph to generate contextual data, maintenance actions can be closely aligned with business objectives, avoiding blind operations divorced from the business scenario and ensuring that every maintenance action matches the core business needs of the enterprise. The panoramic state vector formed by real-time fusion of business issue context and system status data allows the reinforcement learning agent to fully grasp the system dynamics, and the generated operational operation instruction sequence is more in line with the actual system operation status. At the same time, the intervention of the security execution engine can ensure the compliance and reliability of maintenance operations, effectively reduce interference with the normal operation of business systems during maintenance, improve the timeliness of maintenance response and the accuracy of operations, and deeply align with the enterprise's business logic and system operation characteristics, providing precise and targeted support for operations and maintenance.

[0058] By leveraging the mechanism of reinforcement learning agents to dynamically adjust model parameters based on operational performance, these agents can continuously adapt to changes in enterprise business systems over long-term applications. Whether it's business process updates or system architecture upgrades, the agents can gradually optimize decision-making logic, forming operational capabilities that are synchronized with enterprise development. Furthermore, automated operational processes reduce the uncertainty caused by manual intervention, freeing up operational personnel to focus on more important system planning and business support work. At the same time, stable operational performance improves the operational continuity of target business systems, providing a solid guarantee for the smooth operation of enterprise business. This helps enterprises maintain an efficient and stable development trend in digital operations, endowing enterprise operations and maintenance with continuous optimization capabilities and long-term value.

[0059] like Figure 2 As shown in the embodiments of this application, an enterprise operation and maintenance equipment based on an AI intelligent agent is also proposed, including:

[0060] At least one processor; and,

[0061] A memory communicatively connected to the at least one processor; wherein,

[0062] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform an enterprise operation and maintenance method based on an AI intelligent agent as described in any of the above embodiments.

[0063] This application also provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: an enterprise operation and maintenance method based on an AI intelligent agent as described in any of the above embodiments.

[0064] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0065] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0071] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0072] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0073] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for enterprise operation and maintenance based on AI intelligent agents, characterized in that, include: Receive a business processing request from the target business system, obtain business problem description information, parse the business problem description information, and extract key entity information and intent information; Based on the key entity information and the intent information, business problem context data is generated by using a pre-built business operation association graph. The business operation association graph includes business entity nodes and operation entity nodes that are associated in a graph structure in the target business system; The target business system is monitored in real time, system status data is collected, and the business problem context data and system status data are fused to generate a system panoramic status vector. Based on the system's panoramic state vector, an operational operation instruction sequence is generated through a pre-trained reinforcement learning agent. The operational operation instruction sequence is then sent to the security execution engine, which executes the operational operation corresponding to the operational operation instruction sequence. After the sequence of operational instructions is executed, the model parameters of the reinforcement learning agent are adjusted according to the current system status data of the target business system.

2. The enterprise operation and maintenance method based on AI intelligent agents according to claim 1, characterized in that, The step of generating business problem context data based on the key entity information and the intent information through a pre-constructed business operation association graph specifically includes: The pre-built business operation association graph is traversed and queried to obtain the association data of the key entity information; Calculate the relevance between the intent information and the associated data, filter and sort the associated data based on the relevance, and obtain an association subgraph; The associated subgraph, the key entity information, and the intent information are structured and encapsulated to generate business problem context data.

3. The enterprise operation and maintenance method based on AI intelligent agents according to claim 1, characterized in that, The step of generating an operational instruction sequence based on the system's panoramic state vector, using a pre-trained reinforcement learning agent, specifically includes: The system's panoramic state vector is input into a pre-trained reinforcement learning agent to obtain the operational action space of the reinforcement learning agent; the action space consists of all executable operations defined in a preset operational action metadata database. Using the policy network in the reinforcement learning agent, based on the system's panoramic state vector, the expected value score of each executable operation in the operation action space is calculated, and the candidate operation with the highest expected value score is determined. The system queries the preset operation metadata database to obtain the complete instruction definition and required execution parameters corresponding to the candidate operation, generates candidate operation instructions, and adds them to the operation instruction sequence. The system state data is updated in real time to generate a new system panoramic state vector. The new system panoramic state vector is iteratively input into the reinforcement learning agent to generate new candidate operation instructions, which are added to the operation instruction sequence until the system state data converges or the operation instruction sequence reaches its maximum length.

4. The enterprise operation and maintenance method based on AI intelligent agents according to claim 3, characterized in that, Before generating the operational instruction sequence based on the system panoramic state vector using a pre-trained reinforcement learning agent, the method further includes: Determine the core business objective corresponding to the business processing request, assign corresponding weight coefficients to the core business objective, and construct a multi-objective linear weighted reward function; Based on all executable operational operations in the preset operational operation metadata database, an action space is constructed; The reinforcement learning agent architecture is initialized based on a deep neural network, and the multi-objective linear weighted reward function and the action space are added to the reinforcement learning agent architecture to construct the reinforcement learning agent.

5. The enterprise operation and maintenance method based on AI intelligent agents according to claim 4, characterized in that, Before constructing the action space based on all executable operational operations in the preset operational operation metadata database, the method further includes: Obtain all executable operation operations from the preset operation metadata database, create corresponding operation identifiers for each executable operation, and add metadata descriptions. The metadata description includes the operation name, operation type, target system interface, call parameter template, required permissions, estimated execution time, a list of expected impact business metrics, and risk level label.

6. The enterprise operation and maintenance method based on AI intelligent agents according to claim 3, characterized in that, The execution of the operational operation corresponding to the operational operation instruction sequence through the secure execution engine specifically includes: The secure execution engine receives the sequence of operational operation instructions, sequentially reads the operational operation instructions in the sequence, and verifies the operational operation instructions through the preset operational operation metadata database. Once the verification is successful, an interface call operation request is generated according to the mapping rules corresponding to the operation operation instruction and sent to the corresponding application interface. The application programming interface (API) receives and executes the API call operation request, and monitors the execution status code and return result data of the API call operation request in real time.

7. The enterprise operation and maintenance method based on AI intelligent agents according to claim 1, characterized in that, The step of adjusting the model parameters of the reinforcement learning agent based on the current system state data of the target business system specifically includes: Collect the current system status data of the target business system after the execution of the operational operation instruction sequence; the current system status data includes operational result data and performance result data. The current system state data is compared with the system state data before execution to determine the actual amount of change; Based on the system panoramic state vector, the operation instruction sequence, and the actual change, new training sample data points are constructed. The model parameters of the reinforcement learning agent are updated using the new training sample data points to iteratively optimize the reinforcement learning agent.

8. The enterprise operation and maintenance method based on AI intelligent agents according to claim 1, characterized in that, The process of constructing the business operation association graph specifically includes: Obtain the business entity objects and operation entity objects of the target business system, and determine the type of association between the business entity objects and the operation entity objects; the type of association includes data flow relationship, functional dependency relationship, business affiliation relationship and performance impact relationship. Based on the aforementioned relationship type, the business entity object and the operation entity object are connected to construct a business operation relationship graph.

9. An enterprise operation and maintenance equipment based on AI intelligent agents, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform an enterprise operation and maintenance method based on an AI intelligent agent as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to execute an enterprise operation and maintenance method based on an AI intelligent agent as described in any one of claims 1 to 8.

Images