Operation and maintenance method and device for intelligent agent, equipment and storage medium

Abstract

The invention discloses an operation and maintenance method and device for an intelligent agent, equipment and a storage medium, and relates to the technical field of intelligent agents. The method comprises the steps that an edge sensing layer determines a risk score of an intelligent agent, and the edge sensing layer sends an alarm packet to a cloud decision-making layer when the risk score is greater than a first preset threshold value; the cloud decision-making layer determines the risk probability of the intelligent agent based on the alarm packet, and generates an intervention instruction and a channel construction decision for the intelligent agent when the risk probability is greater than a second preset threshold value; the dynamic execution layer establishes a secure channel based on the channel construction decision, and transmits an intervention instruction to the agent based on the secure channel so as to control the agent to perform operation and maintenance based on the intervention instruction; and when it is detected that the intelligent agent reaches the preset state, releasing the resources of the security channel.

Description

Technical Field

[0001] This application relates to the field of intelligent agent technology, and specifically to an operation and maintenance method, apparatus, device and storage medium for intelligent agents. Background Technology

[0002] An embodied intelligent agent refers to an autonomous intelligent entity with a physical or virtual form that can perceive its surrounding environment using multimodal sensors such as vision, force, position perception, and environmental interaction, and make decisions based on the perceived information to interact with the environment.

[0003] Most mainstream embodied agent management solutions currently adopt a persistent background connection mechanism, which maintains a permanent or long-term connection between the agent and the background, such as through a persistent transmission control protocol / Internet Protocol connection or a fixed application programming interface channel.

[0004] However, this continuous background connection scheme requires the agent to maintain an active connection with the backend for a long period of time, which continuously consumes computing resources, storage space, and network bandwidth. Since the actual effective interaction rate is usually low, resources remain idle most of the time, resulting in resource waste. Summary of the Invention

[0005] The purpose of this application is to provide an operation and maintenance method, apparatus, device, and storage medium for intelligent agents.

[0006] To achieve the above objectives, the first aspect of this application provides an operation and maintenance method for intelligent agents. This method is applied to an operation and maintenance framework for intelligent agents, the framework comprising an edge perception layer, a cloud decision layer, and a dynamic execution layer. The method includes: The edge perception layer determines the risk score of the intelligent agent. If the risk score is greater than the first preset threshold, the edge perception layer sends an alarm packet to the cloud decision layer. The cloud-based decision-making layer determines the risk probability of the intelligent agent based on the alarm package. If the risk probability is greater than the second preset threshold, the cloud-based decision-making layer generates intervention instructions for the intelligent agent and channel construction decisions. The dynamic execution layer establishes a secure channel based on the channel construction decision, and transmits intervention commands to the agent based on the secure channel to control the agent to perform operation and maintenance based on the intervention commands; Release the resources of the safe channel when the agent is detected to have reached a preset state.

[0007] In this embodiment of the application, the edge perception layer determines the risk score of the agent by: the edge perception layer acquiring the original data of multiple modalities of the agent, as well as the weight and standard data corresponding to each modality; for each modality, determining the anomaly coefficient based on the original data and standard data, and summing the product of the anomaly coefficients of all modalities and the corresponding weights to determine the risk score of the agent.

[0008] In this embodiment, the alarm package includes the agent's identifier and abnormal modality. The cloud decision layer determines the agent's risk probability based on the alarm package, including: the cloud decision layer obtaining the agent's historical failure rate and similar agent experience knowledge base based on the identifier; and inputting the abnormal modality, historical failure rate, and similar agent experience knowledge base into the multimodal scenario prediction model to obtain the agent's risk probability output by the multimodal scenario prediction model.

[0009] In this embodiment, the method further includes: a cloud-based decision layer verifying the executability of the intervention instruction based on preset behavioral rules; if the executability verification is successful, verifying the scenario compatibility of the intervention instruction based on the agent's environmental state, task progress, and interactive objects; if the scenario compatibility verification is successful, predicting the intervention instruction's predicted data based on the agent's multiple modal raw data; if the predicted data verification is successful, obtaining the intervention instruction's manual review result; and if the manual review result verification is successful, generating a channel construction decision for the agent.

[0010] In this embodiment of the application, the method further includes: the dynamic execution layer acquiring feedback data from multiple modalities of the agent and calculating the deviation value between the feedback data and the predicted data; if the deviation value is greater than a third preset threshold, transmitting a rollback command to the agent based on the safe channel to control the agent to restore the original state based on the rollback command; if the deviation value is less than or equal to the third preset threshold, jumping to the step of releasing the resources of the safe channel when the agent is detected to have reached a preset state.

[0011] In this embodiment of the application, the method further includes: the dynamic execution layer obtaining the core function recovery rate of the agent and the corresponding first weight, the safety indicator value and the corresponding second weight; superimposing the product of the core function recovery rate and the first weight, and the product of the safety indicator value and the second weight to obtain the health index of the agent; and determining that the agent has reached the preset state when the health index is greater than the fourth preset threshold and no new intervention instruction is detected within the preset time period.

[0012] The second aspect of this application provides an operation and maintenance framework for intelligent agents. The framework includes: a risk perception layer, used to determine the risk score of the intelligent agent and send an alarm package to a cloud decision layer when the risk score is greater than a first preset threshold; a cloud decision layer, used to determine the risk probability of the intelligent agent based on the alarm package and generate intervention instructions and channel construction decisions for the intelligent agent when the risk probability is greater than a second preset threshold; and a dynamic execution layer, used to establish a safe channel based on the channel construction decisions and transmit intervention instructions to the intelligent agent based on the safe channel to control the intelligent agent to perform operation and maintenance based on the intervention instructions, and release the resources of the safe channel when the intelligent agent is detected to have reached a preset state.

[0013] A third aspect of this application provides an operation and maintenance apparatus for an intelligent agent, comprising: a memory configured to store instructions; a processor configured to retrieve instructions from the memory and to implement an operation and maintenance method for the intelligent agent when executing the instructions.

[0014] The fourth aspect of this application provides an operation and maintenance device for intelligent agents, including: an operation and maintenance apparatus for intelligent agents.

[0015] The fifth aspect of this application provides a machine-readable storage medium storing instructions that cause a machine to perform an operation and maintenance method for an intelligent agent.

[0016] The aforementioned technical solution breaks through the limitation of continuous channel occupancy in traditional persistent connection architectures, instead implementing an on-demand triggering mechanism based on dynamic risk assessment. The secure channel between the cloud and the intelligent agent is activated only when a high risk is identified, thereby transforming the network resource usage pattern from "continuous idle occupancy" to "instantaneous on-demand access." This paradigm shift effectively solves the problem of long-term bandwidth and computing power idleness, significantly improving resource utilization efficiency.

[0017] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings: Figure 1 The schematic diagram illustrates a flowchart of an operation and maintenance method for an intelligent agent according to an embodiment of this application; Figure 2 The schematic diagram illustrates a flowchart of an operation and maintenance method for an intelligent agent according to an embodiment of this application; Figure 3 This schematic diagram illustrates a structural block diagram of an operation and maintenance device for intelligent agents according to an embodiment of this application; Figure 4 The diagram illustrates the internal structure of a computer device according to an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] Figure 1 The illustration schematically shows a flowchart of an operation and maintenance method for an intelligent agent according to an embodiment of this application. For example... Figure 1 As shown in the figure, this application provides an operation and maintenance method for intelligent agents, applied to an operation and maintenance framework for intelligent agents. The operation and maintenance framework includes an edge perception layer, a cloud decision layer, and a dynamic execution layer. The method may include the following steps: Step 101: The edge perception layer determines the risk score of the intelligent agent. If the risk score is greater than the first preset threshold, the edge perception layer sends an alarm packet to the cloud decision layer.

[0021] The edge perception layer is deployed locally on the intelligent agent, responsible for collecting data in real time and conducting preliminary risk assessments. The cloud decision-making layer is deployed in the cloud, receiving alarm information reported by the edge perception layer, conducting precise risk assessments, generating response strategies, and distributing them to the dynamic execution layer. The dynamic execution layer is responsible for translating cloud decisions into concrete actions.

[0022] Optionally, the intelligent agent acquires real-time sensor data streams and calculates risk scores in real time based on a small, efficient neural network or rule model running internally. Then, the risk score is compared with a first preset threshold (e.g., 0.7): if the risk score is less than or equal to the first preset threshold, it indicates that the agent is in a normal or controllable state, and monitoring continues; if the risk score is greater than the first preset threshold, it means that the agent has initially assessed that there is a risk. At this time, the edge perception layer will send an alarm packet to the cloud decision layer through a secure channel, or directly call its API interface. The specific content may include the agent ID, key situation summary, risk score, abnormal modality, etc. The alarm packet here can use summary generation technology to compress the original data stream into key state feature packets, realizing a qualitative upgrade in data transmission volume from massive raw streams to extremely simple event packets, solving network congestion and latency bottlenecks.

[0023] In one feasible implementation, the modalities that the intelligent agent needs to monitor are clearly defined, such as simplified visual features, key positions / poses, tactile / force summaries, interaction states, and internal diagnostic codes. Based on this, raw data for each modality is collected in real time via sensors. Simultaneously, modal weights pre-assigned according to business priorities and standard data defining the normal state of each modality based on historical normal data statistics are obtained; these can be in the form of standard values ​​or standard ranges. Then, for each modality, the collected raw data is compared with the set standard data to calculate an anomaly coefficient, which reflects the degree to which the modality deviates from the normal state. Specifically, a threshold comparison method can be used: if the raw data exceeds the standard range, the coefficient is 1; otherwise, it is 0. Alternatively, a deviation ratio method can be used, such as anomaly coefficient = (raw data - standard data) / standard data; no specific limitation is made here. Finally, the anomaly coefficient of each modality is multiplied by its weight, and a weighted sum is obtained to obtain the final comprehensive risk score. The formula is This is to allow for a brief risk assessment in conjunction with a first preset threshold.

[0024] Therefore, by using the edge perception layer deployed locally on the intelligent agent, its risk status can be initially assessed, and behavioral patterns or potential risk signs that deviate significantly from expectations can be identified, such as abnormal stagnation, trajectory jitter, interaction response delay, and sudden increase in energy consumption.

[0025] Step 102: The cloud-based decision layer determines the risk probability of the intelligent agent based on the alarm package. If the risk probability is greater than the second preset threshold, the cloud-based decision layer generates intervention instructions and channel construction decisions for the intelligent agent.

[0026] Upon receiving an alarm packet from the edge perception layer, the cloud-based decision-making layer performs in-depth analysis by integrating multi-source data, including historical operational data, task context, environmental models, and experience knowledge bases from similar intelligent agents. This data is then fed into a high-precision predictive model that incorporates advanced methods such as time series analysis, graph neural networks, and causal inference to accurately assess the risk probability of the intelligent agent. Furthermore, we can further analyze the agent's risk type (such as physical collision, task failure, social conflict, and energy runaway), expected impact range, and optimal intervention window. Among these, the optimal intervention time window is... This refers to the time interval from risk identification to the expected occurrence of the failure, used to guide the timeliness of intervention. .

[0027] In one feasible implementation, the cloud-based decision layer uses the agent identifier in the alarm package as the query key to retrieve the agent's historical failure rate and the experience knowledge base of similar agents. The historical failure rate is calculated based on the number of failures and total runtime within each statistical period. The experience knowledge base of similar agents includes, but is not limited to, information such as operational data, failure cases, and solutions for other agents of the same type or with similar characteristics. Then, the abnormal modality, historical failure rate, and experience knowledge base of similar agents are input into a multimodal scenario prediction model. This multimodal scenario prediction model, based on built-in algorithms and logic, comprehensively considers the correlation and influence between abnormal modal characteristics, historical failure trends, and group experience, and calculates risk probability using machine learning or deep learning algorithms. For example, different sub-modules within the model will process various types of information: the numerical analysis module assesses the danger level of temperature data such as 95℃; the time-series analysis module, combined with the historical failure rate, determines whether the anomaly is a recurring or trending problem; and the knowledge reasoning module activates rules and cases related to "high temperature" and "this type of equipment" in the knowledge base. Finally, the outputs of each module are weighted and integrated at the fusion layer, comprehensively considering the severity of the current anomaly, individual historical performance, and group operational experience, and a comprehensive risk probability is generated through the output layer. In this scheme, the anomaly modality reflects the current operational anomaly, the historical failure rate records the frequency of past failures of individuals, and the experience knowledge base of similar intelligent agents provides operational experience and case references at the group level. The multimodal scenario prediction model can collaboratively process multi-source heterogeneous data, organically integrate them, make full use of various types of information, avoid judgment biases that may be caused by a single data source, more accurately capture potential risk factors, and thus significantly improve the accuracy of risk probability prediction.

[0028] Furthermore, the calculated risk probability is compared with a second preset threshold (e.g., 0.8): if the risk probability > the second preset threshold, the intervention process is initiated; otherwise, it is simply recorded and continuously monitored. During the intervention phase, the cloud-based decision layer generates intervention instructions for the intelligent agent, i.e., selecting or dynamically generating the most suitable intervention instructions from the contingency plan library based on the risk probability and / or risk type. The intervention instructions referred to here are semantic cognitive-level instructions, such as "adjust obstacle avoidance sensitivity parameters to X", "switch to safe interaction mode Y", "replan the path to bypass area Z", and "trigger self-check and repair of specific cognitive modules". These instructions act at the cognitive level, rather than being direct underlying hardware control commands. In addition, the cloud-based decision layer also formulates channel construction strategies, i.e., dynamically allocating channel resources (bandwidth, priority) based on the risk probability and / or risk type, and selecting the corresponding security protocol level, such as reserving dedicated bandwidth and enabling stronger encryption for high-risk scenarios.

[0029] Step 103: The dynamic execution layer establishes a secure channel based on the channel construction decision, and transmits intervention commands to the agent based on the secure channel to control the agent to perform operation and maintenance based on the intervention commands.

[0030] The dynamic execution layer proactively establishes a secure channel with the agent according to the channel construction decision requirements, and issues intervention commands through this channel. Correspondingly, after receiving the intervention commands through the secure channel, the agent verifies the legality and timeliness of the source of the intervention commands, and interprets and executes the corresponding operation and maintenance operations at the cognitive level, such as adjusting decision parameters, loading new behavior strategies, and initiating self-repair processes.

[0031] Step 104: Release the resources of the secure channel when the agent is detected to have reached a preset state.

[0032] It is important to note that after the intervention command is sent, the dynamic execution layer does not immediately disconnect. Instead, it keeps the channel open and listens for information returned by the agent in real time to monitor the progress of operations and maintenance. Only when the agent reaches a preset state, such as when data backup is complete or fault repair has been verified, will the resources of the secure channel be released to ensure the successful execution of the operations and maintenance task and that the relevant risks have been completely eliminated.

[0033] In one feasible implementation, during the design phase, the core functional indicators and their corresponding primary weights are defined based on the design goals and application scenarios of the intelligent agent. Taking an intelligent logistics robot as an example, core functional indicators may include the accuracy and speed of cargo handling. When the intelligent agent is in its optimal operating state, the baseline values ​​of these indicators are measured and recorded as a reference for subsequent evaluation of the degree of functional recovery. In the practical application phase, after the intelligent agent executes the intervention command, the current core functional indicator values ​​are measured in real time, and the core functional recovery rate is calculated according to the formula: Core Functional Recovery Rate = Current Core Functional Indicator Value / Core Functional Baseline Value. Similarly, reasonable upper and lower thresholds and corresponding secondary weights are pre-set for each safety indicator, and in practical applications, the formula can be used: Safety Indicator Value = (Upper Threshold ... (Current security indicator value) / (Upper threshold) The system calculates a safety indicator value using a lower threshold. Then, it calculates the agent's health indicator using the formula: Health Indicator = Core Function Recovery Rate × First Weight + Safety Indicator Value × Second Weight. This health indicator is used to quantify the agent's overall recovery to a stable operating state. Each time a new health indicator is calculated, it is compared to a preset fourth threshold (e.g., 0.9). If the health indicator is not higher than this threshold, the agent is considered not yet recovered, and monitoring continues. If it is higher than the threshold, the system enters the next stage of stability observation, which involves a preset observation period (e.g., 10 minutes) to monitor for new intervention commands. Only when the health indicator consistently exceeds the fourth preset threshold and no new intervention commands are triggered during the observation period is the agent considered to have reached the preset state. At this point, the resources of the safety channel are released, and the agent regains full autonomy.

[0034] In this embodiment, the edge perception layer first performs a preliminary risk assessment of the agent based on risk scoring. If an alarm is triggered, the cloud decision layer intervenes to complete a precise assessment and decision based on risk probability. The dynamic execution layer then establishes a secure channel between the cloud and the agent as needed based on the assessment results and releases channel resources in a timely manner based on the agent's real-time status. This solution replaces long-term active connections with an event-driven model. Through a closed loop of preliminary assessment, precise assessment, and on-demand execution, the cloud establishes a secure channel with the agent only when intervention is needed, and releases resources immediately after the agent's operation stabilizes. This mechanism realizes a paradigm shift in network resource utilization from continuous idle occupation to instantaneous on-demand access, greatly reducing ineffective bandwidth and computing power occupation, thereby ensuring efficient resource utilization.

[0035] Figure 2 The illustration schematically shows a flowchart of an operation and maintenance method for an intelligent agent according to an embodiment of this application. For example... Figure 2 As shown in the embodiments of this application, the method further includes: Step 201: The cloud-based decision-making layer verifies the executability of the intervention instructions based on preset behavioral rules.

[0036] Step 202: If the executability verification passes, verify the scenario compatibility of the intervention instructions based on the agent's environmental state, task progress, and interactive objects.

[0037] Step 203: If the scenario compatibility verification is passed, predict the intervention command based on the raw data of multiple modalities of the agent.

[0038] Step 204: If the prediction data is verified, obtain the manual review results of the intervention instructions.

[0039] Step 205: If the manual review results are verified, the cloud-based decision layer generates a channel construction decision for the intelligent agent.

[0040] First, the cloud-based decision-making layer retrieves relevant behavioral rules for the agent from a pre-defined behavioral rule library. It then compares the intervention command with each of these rules, checking for violations of security rules, ethical guidelines, or access restrictions. If the intervention command violates a behavioral rule, the executability verification fails, an alarm is generated, or alternative strategies are attempted, terminating the subsequent process. If the verification passes, the process proceeds to the scenario compatibility verification stage.

[0041] Furthermore, the current environmental state information of the agent is acquired through various sensors installed on the agent, and the progress information of the agent's current task is obtained from the task management system to clarify the stage of the task and the remaining workload. Additionally, the agent's current interaction object is identified, and relevant information about the interaction object, such as its type, state, and behavior pattern, is obtained. A comprehensive analysis is then performed to determine the compatibility between the intervention command and the current environmental state, task progress, and interaction object. For example, it is determined whether the command is suitable for execution under the current environmental conditions, whether it will interfere with the normal progress of the task, and whether it will conflict with the interaction object. If incompatibility exists, the scenario compatibility verification is deemed unsuccessful, an alarm is generated, or other strategies are tried, and the process is terminated. If the verification passes, the process proceeds to the prediction data generation stage.

[0042] Furthermore, raw data from multiple modalities of the intelligent agent is acquired, and intervention commands are simulated and executed in a sandbox, such as parameter adjustments, path replanning, cognitive module resets, and interaction strategy optimizations. Predictive data on the impact of these intervention commands on the agent's behavior, task objectives, environment, and collaborating objects is generated. This predicted data is then compared with preset reasonable ranges and expected results. If the data exceeds reasonable ranges or does not match expectations, the prediction data verification is deemed unsuccessful, and the subsequent process is terminated. If verification passes, the process proceeds to the manual review stage.

[0043] Furthermore, for intervention instructions with extremely high risks or ethical sensitivities, the instructions and related verification results can be compiled into a detailed report and submitted to reviewers with professional knowledge. Reviewers will conduct a comprehensive review of the intervention instructions based on their experience and judgment. If the reviewers deem the instructions risky or unreasonable, the manual review will be rejected, an alert will be generated, or other strategies will be attempted. If the review is successful, the process will proceed to the channel construction decision generation stage.

[0044] The above solution uses a cognitive behavior sandbox system to construct a four-level security verification system: "rule review → context compatibility → digital twin simulation → manual double signature". This progressively verifies rule compliance, environmental adaptability, simulation feasibility, and manual confirmation of high-risk operations, forming an operational risk firewall and effectively preventing the risk of direct control by low-level instructions.

[0045] It is important to note that, based on the predicted data generated through simulation during the verification phase, this solution can also perform a closed-loop evaluation of the intervention effect after actual execution. The specific process is as follows: The dynamic execution layer acquires feedback data from each modality of the agent in real time, including but not limited to sensor readings, behavioral trajectories, task progress, and self-diagnostic results. Based on the data type and evaluation requirements, an appropriate deviation calculation method is selected, and the deviation between the feedback data and the predicted data is calculated. If the deviation value is greater than a third preset threshold (e.g., 30%), it indicates a significant deviation between the agent's actual state and the expected state, potentially indicating an execution anomaly. In this case, the dynamic execution layer will send a rollback command to the agent through a safety channel, restoring it to its original state before intervention to mitigate risk. If the deviation value is less than or equal to the third preset threshold, it indicates that the agent's current state meets expectations and is within the normal operating range. Under this condition, a long-term observation mechanism is further used to determine whether the agent has reached a preset stable state, and based on this, a decision is made on whether to release the safety channel resources. This solution employs a dual mechanism of "short-term verification + long-term observation" to ensure the correctness of the intervention direction and verify the thoroughness and stability of the recovery process. While ensuring execution efficiency, it also achieves the reliability and security of closed-loop management of the agent's state.

[0046] Figure 1 This is a flowchart illustrating an operation and maintenance method for an intelligent agent in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0047] In one embodiment, an operation and maintenance framework for an intelligent agent is provided, comprising: an edge perception layer for determining the risk score of the intelligent agent and sending an alarm package to a cloud decision layer when the risk score is greater than a first preset threshold; a cloud decision layer for determining the risk probability of the intelligent agent based on the alarm package and generating intervention instructions and channel construction decisions for the intelligent agent when the risk probability is greater than a second preset threshold; and a dynamic execution layer for establishing a secure channel based on the channel construction decisions and transmitting intervention instructions to the intelligent agent based on the secure channel to control the intelligent agent to perform operation and maintenance based on the intervention instructions, and releasing the resources of the secure channel when the intelligent agent is detected to have reached a preset state.

[0048] In one embodiment, such as Figure 3 As shown, an operation and maintenance device 300 for intelligent agents is provided, including a risk scoring determination module, a risk probability determination module, a secure channel establishment module, and a secure channel release module, wherein: The risk scoring determination module 301 determines the risk score of the intelligent agent by the edge perception layer. If the risk score is greater than the first preset threshold, the edge perception layer sends an alarm packet to the cloud decision layer.

[0049] The risk probability determination module 302 determines the risk probability of the intelligent agent based on the alarm package. If the risk probability is greater than the second preset threshold, the cloud decision layer generates intervention instructions and channel construction decisions for the intelligent agent.

[0050] The security channel establishment module 303 establishes a security channel based on the channel construction decision of the dynamic execution layer, and transmits intervention commands to the intelligent agent based on the security channel to control the intelligent agent to perform operation and maintenance based on the intervention commands.

[0051] The safe channel release module 304 releases the resources of the safe channel when it detects that the agent has reached a preset state.

[0052] The operation and maintenance device for intelligent agents includes a processor and a memory. The aforementioned risk scoring determination module, risk probability determination module, safe channel establishment module, and safe channel release module are all stored as program units in the memory. The processor executes the aforementioned program modules stored in the memory to implement the corresponding functions.

[0053] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and the operational methods for the intelligent agent can be implemented by adjusting kernel parameters.

[0054] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0055] This application provides a storage medium on which a program is stored. When the program is executed by a processor, it implements the above-described operation and maintenance method for intelligent agents.

[0056] This application provides a processor for running a program, wherein the program executes the above-described operation and maintenance method for intelligent agents during runtime.

[0057] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor A01, a network interface A02, memory (not shown), and a database (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A04. The non-volatile storage medium A04 stores an operating system B01, a computer program B02, and a database (not shown). The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A04. The database is used to store data. The network interface A02 is used for communication with external terminals via a network connection. When executed by the processor A01, the computer program B02 implements an operation and maintenance method for an intelligent agent.

[0058] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0059] This application provides a computer (electronic) device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of any of the above-mentioned operation and maintenance methods for intelligent agents.

[0060] This application also provides a computer program product that, when executed on a data processing device, is suitable for performing initialization steps of an operation and maintenance method for an intelligent agent.

[0061] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0062] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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 process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0063] 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.

[0064] 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.

[0065] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0066] Memory may include non-persistent memory 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.

[0067] Computer-readable media includes 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.

[0068] 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 process, method, article, or apparatus. Unless otherwise specified, 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 that element.

[0069] The above are merely embodiments of this application and are 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 the operation and maintenance of intelligent agents, characterized in that, The method is applied to an operational framework for intelligent agents, the operational framework including an edge perception layer, a cloud decision-making layer, and a dynamic execution layer, and the method includes: The edge perception layer determines the risk score of the intelligent agent, and if the risk score is greater than a first preset threshold, the edge perception layer sends an alarm packet to the cloud decision layer. The cloud-based decision layer determines the risk probability of the intelligent agent based on the alarm package. If the risk probability is greater than a second preset threshold, the cloud-based decision layer generates an intervention command and a channel construction decision for the intelligent agent. The dynamic execution layer establishes a secure channel based on the channel construction decision, and transmits the intervention command to the agent based on the secure channel to control the agent to perform operation and maintenance based on the intervention command; If the agent is detected to have reached a preset state, the resources of the secure channel are released.

2. The operation and maintenance method for intelligent agents according to claim 1, characterized in that, The edge perception layer determines the risk score of the agent, including: The edge perception layer acquires the raw data of multiple modalities of the agent, as well as the weights and standard data corresponding to each modality; For each modality, an anomaly coefficient is determined based on the original data and the standard data, and the risk score of the agent is determined by summing the anomaly coefficients of all modalities and their corresponding weights.

3. The operation and maintenance method for intelligent agents according to claim 1, characterized in that, The alarm package includes the agent's identifier and abnormal modality. The cloud decision layer determines the risk probability of the agent based on the alarm package, including: The cloud-based decision layer obtains the historical failure rate of the intelligent agent and the experience knowledge base of similar intelligent agents based on the identifier. The abnormal modality, the historical failure rate, and the experience knowledge base of similar intelligent agents are input into the multimodal scenario prediction model to obtain the risk probability of the intelligent agent output by the multimodal scenario prediction model.

4. The operation and maintenance method for intelligent agents according to claim 1, characterized in that, The method further includes: The cloud-based decision-making layer verifies the executability of the intervention instructions based on preset behavioral rules; If the executability verification passes, the scenario compatibility of the intervention command is verified based on the environmental state, task progress, and interaction objects of the intelligent agent. If the scenario compatibility verification is passed, the predicted data of the intervention instruction is predicted based on the raw data of multiple modalities of the agent; If the predicted data is verified, obtain the results of the manual review of the intervention instructions; If the manual review result is verified as successful, the cloud-based decision layer generates a channel construction decision for the intelligent agent.

5. The operation and maintenance method for intelligent agents according to claim 4, characterized in that, The method further includes: The dynamic execution layer acquires feedback data from multiple modalities of the agent and calculates the deviation between the feedback data and the predicted data; If the deviation value is greater than a third preset threshold, a rollback command is transmitted to the agent through the security channel to control the agent to restore the original state based on the rollback command. If the deviation value is less than or equal to the third preset threshold, the process jumps to the step of releasing the resources of the secure channel when the agent is detected to have reached a preset state.

6. The operation and maintenance method for intelligent agents according to any one of claims 1-5, characterized in that, The method further includes: The dynamic execution layer obtains the core function recovery rate of the agent and the corresponding first weight, security index value and corresponding second weight; The health index of the agent is obtained by superimposing the product of the core function recovery rate and the first weight, and the product of the security index value and the second weight. If the health index is greater than the fourth preset threshold and no new intervention instructions are detected within the preset time period, the agent is determined to have reached the preset state.

7. An operation and maintenance framework for intelligent agents, characterized in that, The operation and maintenance framework includes: An edge perception layer is used to determine the risk score of the intelligent agent and send an alarm package to the cloud decision layer if the risk score is greater than a first preset threshold. The cloud-based decision layer is used to determine the risk probability of the intelligent agent based on the alarm package, and generate intervention instructions and channel construction decisions for the intelligent agent when the risk probability is greater than a second preset threshold. A dynamic execution layer is used to establish a secure channel based on the channel construction decision, and to transmit the intervention command to the agent based on the secure channel, so as to control the agent to perform operation and maintenance based on the intervention command, and to release the resources of the secure channel when the agent is detected to have reached a preset state.

8. A maintenance device for intelligent agents, characterized in that, include: The memory is configured to store instructions; The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the operation and maintenance method for an intelligent agent according to any one of claims 1 to 6.

9. An operation and maintenance device for intelligent agents, characterized in that, include: The operation and maintenance device for intelligent agents according to claim 8.

10. A machine-readable storage medium storing instructions thereon, characterized in that, When executed by a processor, this instruction causes the processor to be configured to perform the operation and maintenance method for an agent according to any one of claims 1 to 6.

Images