Intelligent computing cloud platform providing computing power based on multi-agent intelligent operation and maintenance method and device

Abstract

This invention provides an intelligent computing cloud platform that provides computing power, based on a multi-agent intelligent operation and maintenance method and apparatus. It relates to fields such as intelligent computing centers, smart computing centers, computing power infrastructure, and smart computing cloud technology. The method includes: the intelligent computing cloud platform constructing interactive agents such as a perception agent, a diagnostic agent, a planning agent, a scheduling agent, an evaluation agent, a graph database, and a decision engine; the perception agent acquires computing power operation data, searches the graph database for graph matching, and calculates information entropy when no matching graph is found; when the information entropy exceeds a threshold, the diagnostic agent performs root cause analysis based on the computing power operation data to determine the cause of the anomaly; the planning agent generates an intelligent operation and maintenance process based on the cause of the anomaly; the scheduling agent calls operation and maintenance tools to perform operation and maintenance on the abnormal computing power according to the intelligent operation and maintenance process; and the evaluation agent evaluates the operation and maintenance results to execute subsequent operation and maintenance processing operations. This invention significantly improves the autonomy and comprehensiveness of operation and maintenance technology.

Description

Technical Field

[0001] This invention relates to the fields of intelligent computing centers, smart computing centers, computing infrastructure, and smart cloud technologies, specifically to an intelligent computing cloud platform that provides computing power based on a multi-agent intelligent operation and maintenance method and device. Background Technology

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "smart computing centers" have emerged.

[0003] An "intelligent computing center" refers to a facility that provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models) by utilizing large-scale heterogeneous computing resources, including general-purpose and intelligent computing power. Intelligent computing centers encompass facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] "Intelligent computing center" includes, but is not limited to, "intelligent computing center".

[0005] "Intelligent computing center" or artificial intelligence computing center is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing center" and "smart computing center". It is the ability of computer equipment or computing / data center to process data. It is the ability of computer hardware and software to work together to perform a certain computing requirement. It is the computing power to achieve the target output by processing data. It is a new type of productivity that integrates parameter computing power, network carrying capacity and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] Currently, the operation and maintenance technologies of intelligent computing cloud platforms mainly include rule-driven operation and maintenance (O&M) and simple model-assisted O&M. Rule-driven O&M automates processing through pre-written fixed anomaly handling rules and scripts (such as restarting the service if CPU utilization exceeds a threshold). However, this method can only handle known anomalies under preset conditions and cannot identify and handle unknown anomalies. Simple model-assisted O&M uses statistical models or shallow machine learning models to analyze server operating data to achieve anomaly detection and alerts. However, these models typically cannot understand the deep semantics of anomalies, lack autonomous learning and planning capabilities, and cannot effectively identify and handle unknown anomalies.

[0008] It is evident that since the emergence of intelligent computing centers, how to address the shortcomings of intelligent computing cloud platform operation and maintenance technology in terms of autonomy and comprehensiveness has been a pressing issue in this field. Summary of the Invention

[0009] This invention provides a multi-agent intelligent operation and maintenance method and device for an intelligent computing cloud platform that provides computing power, in order to solve the problem of insufficient autonomy and comprehensiveness in the operation and maintenance technology of intelligent computing center cloud platforms in the prior art.

[0010] To solve the above problems, the present invention is implemented as follows:

[0011] In a first aspect, the present invention provides an intelligent computing cloud platform that provides computing power based on a multi-agent intelligent operation and maintenance method, comprising: Step S1: The intelligent computing cloud platform constructs a perceptual intelligent agent, a diagnostic intelligent agent, a planning intelligent agent, a scheduling intelligent agent, an evaluation intelligent agent, a graph database, and a decision engine with interactive relationships; Step S2: The sensing agent acquires computing power operation data within the current time window, and based on the computing power operation data, searches the operation and maintenance causal knowledge graph in the graph database for graph matching. If no matching causal association graph is found, the information entropy is calculated based on the computing power operation data. The information entropy is used to characterize the degree of computing power operation abnormality. Step S3: If the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power. Step S4: The planning agent generates an intelligent operation and maintenance process for the abnormal computing power based on the cause of the abnormality. Step S5: The scheduling agent calls the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process; Step S6: The evaluation agent acquires and evaluates the intelligent operation and maintenance results of the abnormal computing power. When the evaluation results meet different preset conditions, the agent performs the corresponding subsequent operation and maintenance operations in a differentiated manner.

[0012] In one embodiment, step S4 includes: Step S4.1: The planning agent sends the cause of the anomaly to the decision engine, so that the decision engine generates and returns intelligent operation and maintenance process generation and verification decisions based on the cause of the anomaly. Step S4.2: The planning agent generates multiple candidate intelligent operation and maintenance processes for the abnormal computing power based on the intelligent operation and maintenance process and the cause of the anomaly. Step S4.3: The planning agent performs simulation verification on each candidate intelligent operation and maintenance process based on the verification decision to obtain the intelligent operation and maintenance process of the abnormal computing power. The intelligent operation and maintenance process includes at least one of the following operations: isolating the abnormal computing power, scheduling computing tasks on the abnormal computing power, generating maintenance work orders, and rolling back the driver of the abnormal computing power.

[0013] In one embodiment, step S4.3 includes: Step S4.3.1: For each candidate intelligent operation and maintenance process of the abnormal computing power, the planning agent simulates the execution and scores the simulation results of the candidate intelligent operation and maintenance process based on a preset scoring standard. Step S4.3.2: Based on the verification decision, the planning intelligent agent verifies at least one of the following: permissions, operation scope, risk level, and compliance of the candidate intelligent operation and maintenance process with the highest score, and obtains the verification result. Step S4.3.3: If the verification result indicates that the candidate intelligent operation and maintenance process with the highest score has failed the verification, the planning intelligent agent determines the reason for the verification failure based on the verification result. Step S4.3.4: Based on the reason for the verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the intelligent operation and maintenance process for the abnormal computing power.

[0014] In one embodiment, step S4.3.4 includes: Step S4.3.4.1: Based on the reasons for the verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the modified candidate intelligent operation and maintenance process. Step S4.3.4.2: The planning agent iteratively executes process verification on the corrected candidate intelligent operation and maintenance process, and accumulates the number of iterations until the verification is passed and the number of iterations is less than the preset maximum number of iterations, and then obtains the intelligent operation and maintenance process of the abnormal computing power. Step S4.3.4.3: If the number of iterations is greater than or equal to the maximum number of iterations, terminate the current process correction process and execute the downgrade processing mechanism to obtain the intelligent operation and maintenance process of the abnormal computing power.

[0015] In one embodiment, step S2 includes: Step S2.1: The perceptual agent performs semantic parsing on the text information in the computing power operation data, extracts key performance indicators from the performance indicators in the computing power operation data, and obtains parsing results and extraction results. The text information includes at least one of the computing task logs and alarm logs of heterogeneous computing power, and the performance indicators include at least one of hardware performance indicators, network traffic indicators, and application performance indicators. Step S2.2: The perceptual agent sends the analysis result and the extraction result to the graph database; Step S2.3: The graph database extracts correlation features between the parsing results and the extraction results and converts them into a first semantic vector; Step S2.4: The graph database matches the first semantic vector with the second semantic vector of the operation and maintenance causal knowledge graph of the intelligent computing cloud platform to determine whether there is a causal association graph with the computing power operation data. Step S2.5: In the absence of a matching causal relationship graph, calculate the information entropy based on the correlation characteristics of the computing power operation data to determine whether there is abnormal computing power in the intelligent computing cloud platform.

[0016] In one embodiment, step S3 includes: Step S3.1: When the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs abnormal detection on the correlation features corresponding to the abnormal computing power to determine the abnormal features of the abnormal computing power. Step S3.2: The diagnostic agent performs root cause analysis based on the abnormal features to determine the abnormal cause of the abnormal computing power.

[0017] In one embodiment, step S6 includes: Step S6.1: The evaluation agent obtains the operation and maintenance results of the abnormal computing power, the operation and maintenance results include the abnormal handling results and the actual operating status of the abnormal computing power after operation and maintenance management; Step S6.2: The evaluation agent obtains the normal operating status of the abnormal computing power in the intelligent computing cloud platform and determines the deviation between the actual operating status and the normal operating status; Step S6.3: When the anomaly handling result and the deviation meet different preset conditions, the evaluation agent performs subsequent operation and maintenance operations in a differentiated manner.

[0018] In one embodiment, the subsequent operation and maintenance processing includes at least one of updating the graph database, sending the intelligent operation and maintenance results to the sensing agent, and executing a review mechanism.

[0019] In one embodiment, the method further includes: Step S7: The sensing agent monitors the operation process of this intelligent operation and maintenance in real time. Step S8: The sensing agent encrypts and writes at least one of the following into the audit log of the intelligent computing cloud platform: the operation and maintenance operation, the operation timestamp, the operation result, and the identifier of the computing power involved in the operation process. Step S9: In the event of any abnormal operation during the operation process, the scheduling agent cancels any operation, returns to the step of the sensing agent obtaining the computing power operation data, and re-plans the process to generate a new intelligent operation process to perform operation and maintenance on the abnormal computing power of the intelligent computing cloud platform.

[0020] Secondly, the present invention also provides an intelligent computing cloud platform that provides computing power based on a multi-agent intelligent operation and maintenance device, comprising: The building module is used to construct interactive intelligent agents, diagnostic intelligent agents, planning intelligent agents, scheduling intelligent agents, evaluation intelligent agents, graph databases, and decision engines for intelligent computing cloud platforms; The first processing module is used for the sensing agent to obtain computing power operation data within the current time window, and based on the computing power operation data, to search the operation and maintenance causal knowledge graph in the graph database for graph matching. If no matching causal association graph is found, the module calculates information entropy based on the computing power operation data. The information entropy is used to characterize the degree of computing power operation abnormality. The determination module is used to determine that there is abnormal computing power in the intelligent computing cloud platform when the information entropy is greater than a preset information entropy threshold. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power. The generation module is used by the planning agent to generate an intelligent operation and maintenance process for the abnormal computing power based on the cause of the abnormality. The operation and maintenance module is used by the scheduling agent to call the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process; The evaluation module is used by the evaluation agent to obtain and evaluate the intelligent operation and maintenance results of the abnormal computing power, and to perform corresponding subsequent operation and maintenance operations in a differentiated manner when the evaluation results meet different preset conditions.

[0021] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the intelligent computing cloud platform based on the multi-agent intelligent operation and maintenance method for providing computing power as described in the first aspect above.

[0022] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method described in the first aspect above.

[0023] Fifthly, the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method described in the first aspect above.

[0024] In this invention, step S1 involves the intelligent computing cloud platform constructing a perceptual agent, a diagnostic agent, a planning agent, a scheduling agent, an evaluation agent, a graph database, and a decision engine with interactive relationships. Step S2 involves the perceptual agent acquiring computing power operation data within the current time window and, based on this data, searching the graph database for causal knowledge graphs of operation and maintenance for graph matching. If no matching causal graph is found, information entropy is calculated based on the computing power operation data. Step S3 involves determining that abnormal computing power exists in the intelligent computing cloud platform if the information entropy exceeds a preset threshold. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the cause of the abnormality. Step S4 involves the planning agent generating an intelligent operation and maintenance process for the abnormal computing power based on the cause. Step S5 involves the scheduling agent calling the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process. Step S6 involves the evaluation agent acquiring and evaluating the intelligent operation and maintenance results of the abnormal computing power. If the evaluation results meet different preset conditions, corresponding subsequent operation and maintenance processing operations are executed in a differentiated manner. Therefore, by constructing a multi-agent system with interactive relationships, a graph database, and a decision engine on the intelligent computing cloud platform, intelligent operation and maintenance can be carried out on the intelligent computing cloud platform based on the autonomous perception and intelligent identification of various anomalies, including unknown anomalies, autonomous planning of intelligent operation and maintenance processes, and autonomous invocation of operation and maintenance tools to execute intelligent operation and maintenance functions. In particular, for the identification and operation and maintenance of unknown anomalies on the intelligent computing cloud platform, this not only solves the problem that existing operation and maintenance technologies cannot cope with unknown anomalies and the problems of missed and false alarms, and improves the comprehensiveness of operation and maintenance technologies, but also fundamentally solves the shortcomings of existing operation and maintenance technologies in terms of autonomy and comprehensiveness. Attached Figure Description

[0025] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of an intelligent computing cloud platform that provides computing power based on a multi-agent intelligent operation and maintenance method, provided by the present invention. Figure 2 This is a system architecture framework diagram of the intelligent computing cloud platform that provides computing power based on the multi-agent intelligent operation and maintenance method provided by the present invention; Figure 3 This is a flowchart illustrating the intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance method provided by the present invention. Figure 4 This is a data interaction diagram of the intelligent computing cloud platform providing computing power provided by the present invention, based on a multi-agent intelligent operation and maintenance method; Figure 5 This is a schematic diagram of the system interface of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method provided by the present invention; Figure 6 This is a structural diagram of an intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance device provided by the present invention; Figure 7 This is a structural diagram of an electronic device provided by the present invention. Detailed Implementation

[0027] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] The “computing power” mentioned in this invention refers to: the ability of computer equipment or computing / data center to process information; the ability of computer hardware and software to work together to perform a certain computing requirement; the computing power to achieve the target result output by processing information data; and a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0029] The "computational power" (CP) described in this invention refers to the ability of a data center server to process data and output results. It is a comprehensive indicator of a data center's computing power, encompassing general computing power, supercomputing power, and intelligent computing power. The commonly used unit of measurement is floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), with higher values ​​indicating stronger overall computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A supercomputers, 500,000 mainstream server CPUs, or 2 million mainstream laptops. The calculation formula is: CP = CP通用 +CP 智能 +CP 超级 .

[0030] The "Network Power" (NP) mentioned in this invention refers to the performance of data transmission capability of computing facilities, which includes comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, and involves network transmission within and between data centers. It is a comprehensive indicator for measuring network transmission scheduling capability.

[0031] The "Storage Power" (SP) described in this invention refers to the comprehensive capabilities of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon operation. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and internal storage devices within servers. The commonly used unit of measurement for storage capacity is exabytes (EB, 1EB = 2^60 bytes), the commonly used unit of measurement for performance is the number of read / write operations per second (IOPS / TB), and the disaster recovery ratio is an important indicator of security and reliability.

[0032] The "computing infrastructure" mentioned in this invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, enabling centralized computing, storage, transmission, and application of information.

[0033] The "new information infrastructure" mentioned in this invention refers to network infrastructure such as 5G networks, fiber optic broadband networks, backbone networks, international communication networks, and satellite internet; computing infrastructure such as data centers, general computing centers, intelligent computing centers, and supercomputing centers; and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0034] The “computing power” mentioned in this invention includes: general computing power, intelligent computing power, and supercomputing power.

[0035] The "general computing power" mentioned in this invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0036] The "intelligent computing power" mentioned in this invention refers to: a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for various artificial intelligence innovative applications, such as natural language processing and machine vision.

[0037] The “supercomputing power” mentioned in this invention refers to: a centralized computing resource that mainly uses the computing power provided by high-performance computing clusters such as supercomputers, and utilizes multiple computer systems working in parallel to handle extremely complex or data-intensive problems through a dedicated operating system. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0038] The "intelligent computing center" described in this invention refers to a facility that, through the use of large-scale heterogeneous computing resources, including general-purpose computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), primarily provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models). The intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0039] The "intelligent computing cloud platform" mentioned in this invention, abbreviated as "intelligent computing cloud", refers to a cloud computing platform that integrates hardware and software resources based on an intelligent computing center.

[0040] The "intelligent computing center" mentioned in this invention includes, but is not limited to, "smart computing center".

[0041] The "intelligent computing center" mentioned in this invention, also known as an artificial intelligence computing center, is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0042] The "computing center" mentioned in this invention refers to a facility that is mainly composed of infrastructure such as wind, thermal, hydro, and electricity, and IT hardware and software equipment, and has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0043] The "supercomputing center" mentioned in this invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters. It can provide large-scale computing, storage and network services and is widely used in aerospace, defense, oil exploration, climate modeling and genome sequencing and other application scenarios.

[0044] The “computing resources” mentioned in this invention refer to the technologies and facilities required for the development of the digital society that have the ability to compute, transmit, store and apply information, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guaranteeing resources such as wind, fire, water and electricity.

[0045] The "model" mentioned in this invention includes, but is not limited to, "large language model" and "multimodal large model".

[0046] The "large language model" mentioned in this invention refers to a large-scale language model (LLM), which is a language model with a large number of parameters. It is designed to understand and generate human language, and is trained with a large amount of text data. It can perform a wide range of tasks, including text summarization, translation, and sentiment analysis.

[0047] The “Multimodal Large Models” mentioned in this invention refer to models that combine multimodal information such as text, images, videos, and audio for training, including but not limited to multimodal large language models.

[0048] The “computing power running task” mentioned in this invention refers to a specific workload or job executed on computing power resources that requires a certain amount of computing power support, usually involving complex data processing, numerical calculation, model training or simulation scenarios.

[0049] The “computing power resource pool” mentioned in this invention refers to the unified aggregation and centralized management of scattered computing power resources within an intelligent computing cloud platform through virtualization, resource pooling, and abstraction integration technologies, forming a shared, schedulable, elastically scalable, and on-demand allocated intensive computing power resource set.

[0050] The "intelligent agent" described in this invention refers to an intelligent system with a large language model (LLM) as its core "brain," integrating modules such as planning, memory, tool calling, and reflection. It is capable of autonomously understanding user goals, decomposing complex tasks, dynamically calling external resources, executing in a closed loop, and iteratively optimizing, with the core objective of proactively achieving goals.

[0051] The "perceptual intelligent agent" mentioned in this invention refers to a front-end intelligent unit in an intelligent computing cloud platform that is responsible for collecting operational indicators, status data, alarm information, and environmental parameters of devices, computing power, and business systems across the entire domain, perceiving the underlying operational status in real time, and providing raw data sources for upper-level analysis and processing.

[0052] The "diagnostic intelligent agent" mentioned in this invention refers to a dedicated functional unit in an intelligent computing cloud platform that relies on collected data, event logs, and topological relationships, combined with intelligent analysis algorithms, to automatically identify anomalies, locate the root cause of faults, and uncover potential hidden dangers, thereby realizing intelligent fault assessment and problem tracing.

[0053] The "planning intelligent agent" mentioned in this invention refers to an intelligent processing unit in an intelligent computing cloud platform that combines business needs, resource load, operational constraints, and long-term operational requirements to comprehensively formulate rational planning schemes for resource allocation, capacity expansion, task arrangement, and operation and maintenance deployment.

[0054] The "scheduling intelligent agent" described in this invention refers to an execution unit in an intelligent computing cloud platform that, based on a planning scheme and real-time operating load, invokes corresponding operation and maintenance tools to dynamically allocate computing power, cluster resources, business tasks, etc., perform load balancing, task orchestration, and resource migration, thereby achieving on-demand scheduling and efficient reuse of resources.

[0055] The "evaluation agent" mentioned in this invention refers to an analysis unit in an intelligent computing cloud platform that performs quantitative analysis and comprehensive evaluation of resource operation quality, resource utilization, scheduling execution effect, fault handling results, operation and maintenance management capabilities, and outputs evaluation conclusions and optimization suggestions.

[0056] The "graph database" mentioned in this invention refers to a dedicated database that stores the operation and maintenance knowledge of an intelligent computing cloud platform in a unified manner in the form of a graph structure, supporting relational queries, source tracing analysis, and topology visualization.

[0057] The "decision engine" mentioned in this invention refers to the core control center in an intelligent computing cloud platform that integrates diagnostic results and preset control decisions, and automatically generates standardized operation and maintenance process generation and verification decisions through algorithm and rule operations.

[0058] The "computing power operation data" mentioned in this invention refers to the full amount of underlying operating status and working condition monitoring data that is collected in real time and periodically generated during the continuous operation of various computing power nodes (servers, computing hosts, virtual machines, containers, edge computing units, etc.) and heterogeneous computing power resources (CPUs, GPUs, memory, computing power chips, etc.) in the intelligent computing cloud platform. It can be used to characterize the real-time operating status of computing power hardware, computing units, and cluster resources.

[0059] The "computing node" mentioned in this invention refers to an independent physical or logical computing unit in an intelligent computing cloud platform, mainly composed of computing servers, artificial intelligence computing hosts, and accelerated computing devices.

[0060] The "Kubernetes Control Plane" mentioned in this invention refers to a graphical and visual management interface provided to the cluster in an intelligent computing cloud platform. It replaces traditional command-line operations, centrally displays all resources such as cluster nodes, container groups (Pods), containers, controllers, storage, networks, and control plane components, and supports integrated management of resource viewing, status monitoring, log querying, event browsing, resource editing, scaling up and down, and operation and maintenance, thereby realizing visualized management and control of cluster resources and daily operation and maintenance.

[0061] The "Operations and Maintenance Console" mentioned in this invention refers to: a unified operations and maintenance entry point for the intelligent computing cloud platform, which centrally manages computing facilities, cloud clusters, business services and hardware devices; and integrates status monitoring, command issuance, fault handling, access control, log auditing and policy configuration to achieve integrated operations and maintenance management across the entire domain.

[0062] The “acquisition probe” mentioned in this invention refers to an acquisition component in an intelligent computing cloud platform, which can be used to acquire computing power operation data in the intelligent computing cloud platform.

[0063] Please see Figure 1 , Figure 1 This is a flowchart of an intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance method, as provided by the present invention. Figure 1 As shown, the method includes: Step S1: The intelligent computing cloud platform constructs a perceptual intelligent agent, a diagnostic intelligent agent, a planning intelligent agent, a scheduling intelligent agent, an evaluation intelligent agent, a graph database, and a decision engine with interactive relationships.

[0064] The aforementioned graph database can be used to store the operation and maintenance causal knowledge graph of the intelligent computing cloud platform. This operation and maintenance causal knowledge graph can be constructed by organizing multiple operation and maintenance question-and-answer pairs of the intelligent computing cloud platform through association relationships.

[0065] The aforementioned decision engine can be used to assist planning agents in planning and generating intelligent operation and maintenance processes for intelligent computing cloud platforms. It can assist planning agents in solving complex problems through deep thinking.

[0066] In one embodiment, when constructing interactive intelligent agents, diagnostic intelligent agents, planning intelligent agents, scheduling intelligent agents, and evaluation intelligent agents on an intelligent computing cloud platform, the present invention can deploy each intelligent agent on the management and control server of the intelligent computing cloud platform and perform security initialization configuration for each intelligent agent to integrate an integrated security management and control system.

[0067] For example, specific implementations of security initialization configuration for each agent may include: Hierarchical permission system and principle of least privilege: A hierarchical permission system is established, including administrators, operational agents (including perception agents, diagnostic agents, planning agents, scheduling agents, and evaluation agents), and read-only users. Each agent possesses only the preset, minimal operational permissions, strictly limiting its access and operation capabilities beyond its scope.

[0068] Secure Communication and Data Encryption: Each intelligent agent establishes connections with all servers (including computing and storage nodes), network devices, and various monitoring tools (such as CPU / GPU utilization, memory, disk I / O, and network traffic monitoring) within the intelligent computing cloud platform through encrypted interfaces conforming to security protocols (such as transport layer security protocols). All real-time collected server operation data, alarm logs, configuration information, and other sensitive operational data are encrypted end-to-end using strong encryption algorithms such as Advanced Encryption Standard (AES).

[0069] Toolset Integration and Permission Thresholds: Integrate commonly used operation and maintenance toolsets (such as service restart, resource allocation, container management, log cleanup, etc.) for each intelligent agent, and set fine-grained operation permission thresholds, operation frequency limits, and mandatory operation log recording functions for each tool.

[0070] It should be noted that, in the embodiments of the present invention, each intelligent agent built in the intelligent computing cloud platform can continuously improve its own capabilities by learning and optimizing autonomously through the historical operation and maintenance data of the intelligent computing cloud platform.

[0071] Step S1 achieves beneficial technical effects: By establishing a secure and reliable communication and operation channel between the intelligent agent and the intelligent computing cloud platform, it ensures that the intelligent agent can comprehensively, compliantly, and autonomously perceive the status of the intelligent computing cloud platform and execute operations. Through access control and data encryption, unauthorized access and sensitive data leakage are prevented at the source, laying the foundation for subsequent autonomous and secure operation and maintenance. In other words, by introducing integrated security designs such as hierarchical permissions and data encryption, data security and operational compliance during the operation and maintenance process are ensured, thereby fundamentally solving the shortcomings of existing operation and maintenance technologies in terms of autonomy and security.

[0072] Step S2: The sensing agent acquires the computing power operation data within the current time window, and based on the computing power operation data, searches the operation and maintenance causal knowledge graph in the graph database for graph matching. If no matching causal relationship graph is found, the information entropy is calculated based on the computing power operation data.

[0073] The aforementioned information entropy is used to characterize the degree of anomaly in the computing power operation of an intelligent computing cloud platform. In other words, by calculating information entropy, the degree of anomaly in the computing power operation state of the intelligent computing cloud platform can be quantified. For example, a higher information entropy indicates greater fluctuations and more chaotic distribution of computing power operation data, representing a higher degree of anomaly in the computing power operation state. Conversely, a lower information entropy indicates more stable computing power operation data, representing a more normal computing power operation state and a lower degree of anomaly.

[0074] The size of the aforementioned time window can be set according to actual operation and maintenance needs. For example, the size of the time window can be set to 10 seconds, or it can be set to 10 minutes, etc., and this invention does not limit it in this way.

[0075] The aforementioned computing power operation data may include, but is not limited to, the computing power resource load data of the intelligent computing cloud platform (such as CPU utilization, GPU utilization, memory or disk utilization, etc.), the operation data of computing power operation tasks (such as computing power operation task status, logs, computing power consumption, etc.), storage and input data (such as disk read and write speed, data read and write latency, etc.), network communication data (such as network traffic, latency, packet loss rate, etc.), environmental data (such as alarm logs, etc.), and resource scheduling data (such as the utilization rate of computing power nodes in the intelligent computing cloud platform, resource allocation rate, etc.). The specific data can be determined according to the actual situation, and this invention does not limit it.

[0076] The aforementioned causal relationship graph can be used to determine whether there is computing power with known anomalies in the intelligent computing cloud platform. That is, if a matching causal relationship graph exists, it can be determined that there is computing power with known anomalies in the intelligent computing cloud platform; if no matching causal relationship graph exists, it can be determined that there is no computing power with known anomalies in the intelligent computing cloud platform.

[0077] It should be noted that, in the embodiments of the present invention, the perceptual agent can read the computing power operation data of the intelligent computing cloud platform in real time through an encrypted and legally verified channel. Simultaneously, during the data reading phase, the perceptual agent strictly performs data source legality verification and integrity checks to prevent malicious injection of false data.

[0078] It should be noted that, in the embodiments of the present invention, the specific implementation method for calculating information entropy based on computing power operation data can be preset as needed. For example, joint entropy or weighted entropy can be calculated by performing multi-dimensional operation and maintenance index calculations on the computing power operation data, or the information entropy calculation of computing power operation data can be implemented using approximate entropy, sample entropy, fuzzy entropy, etc. Other implementation methods can also be used to calculate information entropy, and the present invention does not limit this.

[0079] It should be noted that, in the embodiments of the present invention, when calculating information entropy based on computing power operation data, the calculation can be performed from the dimension of computing power resources, that is, the information entropy is calculated independently based on the computing power operation data of each computing power resource, so as to determine the specific abnormal computing power according to the information entropy of each computing power resource.

[0080] In an embodiment of the present invention, after the sensing agent obtains the computing power operation data within the current time window, it can first search the operation and maintenance causal knowledge graph in the graph database based on the computing power operation data to perform graph matching, in order to determine whether there is computing power with known anomalies in the intelligent computing cloud platform. If no matching causal association graph is found, it can be determined that there is no computing power with known anomalies in the intelligent computing cloud platform. At this time, information entropy can be calculated based on the computing power operation data in order to further determine whether there is computing power with unknown anomalies in the intelligent computing cloud platform.

[0081] Step S2 achieves beneficial technical effects: when no causal relationship graph matching the computing power operation data is found in the graph database, information entropy is calculated based on the computing power operation data, which can effectively solve the problem of the inability to identify unknown anomalies and missed reports in the existing operation and maintenance technology of intelligent computing cloud platforms, and significantly improve the accuracy and coverage of anomaly identification.

[0082] Step S3: If the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power.

[0083] The aforementioned information entropy threshold can be used as a critical value for determining whether there is any unknown anomaly in the computing power of the intelligent computing cloud platform. It can be set according to the needs of operation and maintenance, and this invention does not limit it.

[0084] The aforementioned abnormal computing power can refer to the deviation of computing power such as CPU, GPU, computing power cluster, and virtual computing power unit within the intelligent computing cloud platform from the normal steady-state baseline during operation, resulting in problems such as abnormal load, performance degradation, resource failure, and scheduling disorder, making it impossible to provide computing services stably, efficiently, and in compliance with regulations.

[0085] The aforementioned causes of anomalies can be the reasons for computing power anomalies, and can be a set of root cause hypotheses, which can be determined based on the specific computing power. For example, anomalies may include, but are not limited to, hardware anomalies (such as CPU / GPU core damage, computing node crashes, insufficient memory, etc.), software anomalies (such as GPU computing driver anomalies, driver version incompatibility, operating system freezes, virtual machine anomalies, computing power scheduling service anomalies, etc.), computing power task anomalies (such as sudden large-scale computing tasks, task infinite loops, excessively large parameters in model training or inference tasks, etc.), and resource scheduling and allocation anomalies (such as uneven computing power allocation, resource preemption, load balancing failure, etc.), etc. This invention does not limit these.

[0086] Step S3 achieves beneficial technical effects: By performing root cause analysis based on computing power operation data, the diagnostic agent can break free from the constraints of existing operation and maintenance technologies of intelligent computing cloud platforms, which can only identify known anomalies of preset rule types, and achieve autonomous identification of unknown anomalies, thus greatly improving the coverage of anomaly identification.

[0087] Step S4: Plan the intelligent operation and maintenance process of generating abnormal computing power based on the cause of the abnormality.

[0088] The above intelligent operation and maintenance process can be determined based on the specific abnormal computing power and the cause of the abnormality.

[0089] For example, the abnormal computing power is a server, and the cause of the abnormality is a GPU failure in the server. The planning agent can generate an intelligent operation and maintenance process for the server that failed based on the cause of the abnormality as follows: ① Immediately isolate the abnormal server (to prevent the abnormality from spreading) → ② Intelligently migrate the tasks on the failed server to healthy nodes → ③ Attempt to diagnose and execute GPU driver reload or server restart → ④ Continuously monitor the server status to confirm recovery → ⑤ Resume task execution. This invention does not limit the scope of this process.

[0090] In one embodiment, when the sensing agent obtains computing power operation data within the current time window and retrieves the operation and maintenance causal knowledge graph in the graph database based on the computing power operation data for graph matching, if a matching causal relationship graph is found, it can be determined that there is computing power with known anomalies in the intelligent computing cloud platform. At this time, the planning agent can directly reuse the historical intelligent operation and maintenance process corresponding to the known anomaly, without having to plan the intelligent operation and maintenance process from scratch.

[0091] Step S4 achieves beneficial technical effects: By leveraging the powerful planning and reasoning capabilities of the planning agent, it autonomously plans and generates intelligent operation and maintenance processes for abnormal computing power, overcoming the limitation of existing technologies that "can only handle known anomalies and cannot cope with unknown anomalies or dynamic scenarios," and reducing reliance on manual operation and maintenance, thereby improving the adaptability and efficiency of intelligent operation and maintenance process generation.

[0092] Step S5: The scheduling agent calls the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on abnormal computing power according to the intelligent operation and maintenance process.

[0093] The aforementioned intelligent operation and maintenance tools may include, but are not limited to, abnormal computing power isolation tools, task migration tools, abnormal computing power restart tools, and computing power resource scaling tools, etc., which can be determined according to the specific intelligent operation and maintenance process.

[0094] For example, taking the intelligent operation and maintenance process "① immediately isolate the abnormal server (to prevent the abnormal from spreading) → ② intelligently migrate the tasks on the faulty server to healthy nodes → ③ attempt to diagnose and execute GPU driver reload or server restart → ④ continuously monitor the server status to confirm recovery → ⑤ resume task execution" as an example, the corresponding intelligent operation and maintenance tools may include abnormal computing power isolation tools, task migration tools, driver repair tools, computing power node restart tools, status monitoring tools, and task recovery tools. This invention does not limit these.

[0095] In embodiments of the present invention, after receiving the intelligent operation and maintenance process, the scheduling agent will strictly follow the steps and permission scope of the intelligent operation and maintenance process, and call the integrated operation and maintenance tools through the encrypted interface to perform operation and maintenance operations (such as automatically restarting services, migrating tasks, clearing cache, etc.), which greatly improves the security and compliance of operation and maintenance operations.

[0096] Step S5 achieves beneficial technical effects: By scheduling intelligent agents to call the corresponding intelligent operation and maintenance tools according to the intelligent operation and maintenance process to perform intelligent operation and maintenance operations on abnormal computing power, the entire process of intelligent computing cloud platform operation and maintenance is automated, significantly improving operation and maintenance efficiency, minimizing the need for manual operation, and solving the problems of "slow abnormal handling and high labor costs".

[0097] Step S6: The evaluation agent obtains and evaluates the intelligent operation and maintenance results of abnormal computing power. If the evaluation results meet different preset conditions, the agent performs the corresponding subsequent operation and maintenance operations in a differentiated manner.

[0098] The aforementioned intelligent operation and maintenance results may include, but are not limited to, the results of abnormal computing power handling and the actual operating status of abnormal computing power after intelligent operation and maintenance management.

[0099] The aforementioned preset conditions can be used to differentiate the subsequent operation and maintenance processing operations. They can include conditions set for unknown anomalies and conditions set for known anomalies, and can be set according to operation and maintenance needs.

[0100] In one embodiment, subsequent operation and maintenance processing may include, but is not limited to, updating the graph database, sending intelligent operation and maintenance results to the sensing agent, and executing a review mechanism, at least one of these.

[0101] The aforementioned review mechanism is a control mechanism that performs secondary verification on key aspects of the intelligent operation and maintenance process, such as anomaly identification and diagnosis, process planning and execution.

[0102] In other words, in the embodiments of the present invention, when the evaluation results meet different preset conditions, the graph database can be selectively updated or the intelligent operation and maintenance results can be sent to the sensing agent or a review mechanism can be executed.

[0103] Specifically, in one embodiment, the correspondence between different preset conditions and subsequent operation and maintenance operations can be as follows. The following content is merely an example; specific settings can be made according to operation and maintenance needs, and this invention does not limit the scope of the invention: For unknown anomalies, if the evaluation results meet the criteria of successful anomaly handling and the actual operating state of the abnormal computing power deviates little from its normal steady-state baseline on the intelligent computing cloud platform, the abnormal computing power is determined to have recovered to normal operating status. The evaluation agent can update the graph database based on the intelligent operation and maintenance results so that the graph database can supplement the operation and maintenance knowledge related to the unknown anomaly.

[0104] Alternatively, if the evaluation result meets the condition that the anomaly handling result is successful and the actual operating state of the abnormal computing power deviates significantly from its normal steady-state baseline, it can be determined that the abnormal computing power is still unstable after operation and maintenance management, and there is a risk of recurrence or other potential anomalies. The evaluation agent can send the intelligent operation and maintenance result to the perception agent so that the perception agent can continuously perceive and monitor.

[0105] Alternatively, if the evaluation result meets the requirements for handling anomalies, the evaluation agent can execute a review mechanism.

[0106] For known anomalies, if the evaluation results meet the criteria of successful anomaly handling and a significant deviation between the actual operating state of the abnormal computing power and its normal steady-state baseline on the intelligent computing cloud platform, it can be determined that the operation and maintenance knowledge related to the known anomaly in the graph database is defective or outdated. The evaluation agent can update the graph database based on the intelligent operation and maintenance results to correct and update the relevant historical operation and maintenance knowledge in the graph database.

[0107] Alternatively, if the evaluation result meets the criteria of successful anomaly handling and the actual operating state of the abnormal computing power deviates little from its normal steady-state baseline on the intelligent computing cloud platform, the evaluation agent can send the intelligent operation and maintenance result to the perception agent so that the perception agent can continuously perceive and monitor.

[0108] Alternatively, if the evaluation result meets the requirements for handling anomalies, the evaluation agent can execute a review mechanism.

[0109] In one embodiment, when updating the graph database, all information, including the type of each anomaly, identification features, planned processing steps, actual execution results, operation logs, and reasons for success or failure, can be incrementally updated to the graph database in an encrypted and structured manner. This graph database can only be accessed by the agent itself and authorized objects through an encrypted channel. Based on this, when encountering similar anomalies subsequently, the agent can directly retrieve and reuse historical intelligent operation and maintenance processes from the graph database without having to plan from scratch. Simultaneously, the agent can periodically analyze the experience data in the graph database, identify common anomaly patterns, optimize operation and maintenance processes, and perform deep learning on failed operation and maintenance processes to continuously improve its ability to handle complex faults.

[0110] In one embodiment, the intelligent agent can also dynamically adjust or suggest updates to the security policies in the integrated security management system based on the learned anomaly types, operational risks, and historical security events. For example, it can adjust the permission thresholds for certain types of operation and maintenance, increase the encryption strength for access to specific types of data, and strengthen the alarm rules for abnormal operation behaviors, thereby continuously improving the overall security protection capabilities of the intelligent computing cloud platform.

[0111] This enables each intelligent agent to continuously learn and autonomously optimize. As the operating time increases, intelligent operation and maintenance will become more efficient, accurate, and robust, fundamentally solving the core problem of "inability to cope with unknown anomalies and dynamic scenarios." By encrypting and storing experience and dynamically updating security policies, the level of integrated security management and control is continuously strengthened, ensuring the long-term safe and stable operation of the intelligent computing cloud platform.

[0112] Step S6 achieves beneficial technical effects: by evaluating the intelligent operation and maintenance results through an evaluation agent, and under different preset conditions, performing corresponding subsequent operation and maintenance operations in a differentiated manner, the operation and maintenance operations are precisely controlled, and the flexibility and reliability of intelligent operation and maintenance are improved.

[0113] In this invention, step S1 involves the intelligent computing cloud platform constructing a perceptual agent, a diagnostic agent, a planning agent, a scheduling agent, an evaluation agent, a graph database, and a decision engine with interactive relationships. Step S2 involves the perceptual agent acquiring computing power operation data within the current time window and, based on this data, searching the graph database for causal knowledge graphs of operation and maintenance for graph matching. If no matching causal graph is found, information entropy is calculated based on the computing power operation data. Step S3 involves determining that abnormal computing power exists in the intelligent computing cloud platform if the information entropy exceeds a preset threshold. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the cause of the abnormality. Step S4 involves the planning agent generating an intelligent operation and maintenance process for the abnormal computing power based on the cause. Step S5 involves the scheduling agent calling the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process. Step S6 involves the evaluation agent acquiring and evaluating the intelligent operation and maintenance results of the abnormal computing power. If the evaluation results meet different preset conditions, corresponding subsequent operation and maintenance processing operations are executed in a differentiated manner. Therefore, by constructing a multi-agent system with interactive relationships, a graph database, and a decision engine on the intelligent computing cloud platform, intelligent operation and maintenance can be carried out on the intelligent computing cloud platform based on the autonomous perception and intelligent identification of various anomalies, including unknown anomalies, autonomous planning of intelligent operation and maintenance processes, and autonomous invocation of operation and maintenance tools to execute intelligent operation and maintenance functions. In particular, for the identification and operation and maintenance of unknown anomalies on the intelligent computing cloud platform, this not only solves the problem that existing operation and maintenance technologies cannot cope with unknown anomalies and the problems of missed and false alarms, and improves the comprehensiveness of operation and maintenance technologies, but also fundamentally solves the shortcomings of existing operation and maintenance technologies in terms of autonomy and comprehensiveness.

[0114] It should be noted that those skilled in the art are generally accustomed to the traditional operation and maintenance model of "fixed rules + manual assistance" in the existing technology, and generally believe that problems can be solved by continuously expanding the rule base. They have failed to deeply understand that "autonomous learning and autonomous response to unknown anomalies by operation and maintenance tools" is the key to improving efficiency, and have further neglected the deep synergy between operation and maintenance autonomy and integrated security management. Deeply integrating "intelligent agents" with autonomous thinking, learning, and planning capabilities with "refined and integrated security management mechanisms" and applying them to the operation and maintenance of intelligent computing cloud platforms can simultaneously achieve autonomous anomaly handling and full-process security assurance. This involves an innovative combination of cross-domain knowledge, which goes beyond simple improvements to existing technologies. It requires creative labor to achieve and cannot be easily conceived or implemented by those skilled in the art through conventional means.

[0115] In one embodiment, step S4 includes: Step S4.1: The planning agent sends the cause of the anomaly to the decision engine, so that the decision engine can generate and return intelligent operation and maintenance process generation and verification decisions based on the cause of the anomaly.

[0116] The above-mentioned intelligent operation and maintenance process generation decision can be used to plan intelligent agents to generate intelligent operation and maintenance processes, which may include, but are not limited to, operation and maintenance operation types (such as isolating abnormal computing power, task migration, etc.), the order and execution logic of each operation and maintenance operation (parallel execution or serial execution), etc. The specific details can be determined according to actual operation and maintenance needs, and this invention does not limit them.

[0117] The aforementioned verification decisions can be used to plan intelligent agents to perform multi-dimensional security verifications on the generated intelligent operation and maintenance processes. These verifications may include, but are not limited to, permission verification, scope verification, impact assessment, and compliance verification. The specific settings can be configured according to actual operation and maintenance needs, and this invention does not impose any limitations on them.

[0118] For example, permission verification can check whether the agent possesses the minimum permissions required to perform specific operational operations. Scope verification can confirm whether the operational object (such as servers, services, data, etc.) is within the authorized operational scope of the agent. Risk level verification can estimate the potential impact of specific operational operations on business continuity, data integrity, and system stability, and request higher permissions or manual confirmation before high-risk operations. Compliance verification can ensure that specific operational operations comply with preset security compliance rules, such as permission thresholds for operational operations, encryption strength for access to specific types of data, and alarm rules for abnormal operational behavior.

[0119] In this embodiment of the invention, after determining the cause of the abnormal computing power, the planning agent sends the cause of the abnormality to the decision engine. The decision engine can perform in-depth analysis based on the cause of the abnormality, and combine it with preset operation and maintenance decisions and constraint rules to generate intelligent operation and maintenance process generation decisions and verification decisions to solve the abnormal problem, and return them to the planning agent to assist the planning agent in generating accurate and compliant intelligent operation and maintenance processes.

[0120] Step S4.1 achieves beneficial technical effects: By planning the interaction between the intelligent agent and the decision engine, the intelligent operation and maintenance process generated by the decision engine generates decisions and verifies decisions, thereby assisting the planning intelligent agent in generating intelligent operation and maintenance processes. This eliminates manual intervention, realizes safe and compliant autonomous customization of operation and maintenance processes, and solves the problem in existing operation and maintenance technologies that "can only handle known anomalies and cannot cope with unknown anomalies or complex scenarios".

[0121] Step S4.2: Based on the intelligent operation and maintenance process, the planning agent generates decisions and causes of anomalies, and plans multiple candidate intelligent operation and maintenance processes for abnormal computing power.

[0122] Step S4.2 Achieve beneficial technical effects: After receiving the intelligent operation and maintenance process generation decision and verification decision returned by the decision engine, the planning agent assists in planning candidate intelligent operation and maintenance processes for generating abnormal computing power by using the intelligent operation and maintenance process generation decision. This avoids the defects of poor adaptability and single operation and maintenance method of fixed operation and maintenance processes, and can select the optimal intelligent operation and maintenance process to improve the adaptability and flexibility of intelligent operation and maintenance.

[0123] Step S4.3: The planning agent simulates and verifies each candidate intelligent operation and maintenance process based on the verification decision to obtain the intelligent operation and maintenance process with abnormal computing power.

[0124] The aforementioned intelligent operation and maintenance process includes at least one of the following operations: isolating abnormal computing power, scheduling computing tasks on abnormal computing power, generating maintenance work orders, and rolling back the driver of abnormal computing power.

[0125] It should be noted that the intelligent operation and maintenance process may also include, but is not limited to, operation status monitoring operations, task recovery operations, etc., which can be determined according to actual operation and maintenance needs. This invention does not limit this.

[0126] In one embodiment, for a known anomaly, the planning agent can directly reuse the historical operation and maintenance process of that known anomaly. In this process, the planning agent can send the known anomaly to the decision engine to obtain the verification decision returned by the decision engine. Based on the verification decision, the historical operation and maintenance process can be verified to determine whether the historical operation and maintenance process complies with the current security and compliance rules, and to prevent unauthorized, insecure, or non-compliant operation and maintenance operations in the historical operation and maintenance process.

[0127] Step S4.3 achieves beneficial technical effects: The planning agent simulates and verifies each candidate intelligent operation and maintenance process based on the verification decision. This not only enables the selection of the optimal intelligent operation and maintenance process, but also eliminates unauthorized, insecure, and non-compliant operation and maintenance operations from the operation and maintenance process planning stage, thus ensuring the security of intelligent operation and maintenance.

[0128] In one embodiment, step S4.3 includes: Step S4.3.1: For each candidate intelligent operation and maintenance process with abnormal computing power, plan the intelligent agent to simulate execution, and score the simulation results of the candidate intelligent operation and maintenance process based on the preset scoring criteria.

[0129] The above scoring criteria may include, but are not limited to, scoring criteria for indicators such as the overall execution time of the operation and maintenance process, the duration of task interruption, the loss of task migration interruption, resource consumption, and the success rate of anomaly repair. For example, the shorter the overall execution time of the operation and maintenance process, the shorter the duration of task interruption, the lower the loss of task migration interruption, the less resource consumption, and the higher the success rate of anomaly repair, the higher the corresponding indicator score. The specific criteria can be set according to the actual operation and maintenance needs, and this invention does not limit them.

[0130] The simulation results mentioned above may include, but are not limited to, the overall execution time of the operation and maintenance process, the duration of task interruption, the loss of task migration interruption, resource overhead, and the success rate of anomaly repair, etc. The specific settings can be configured according to the operation and maintenance needs, and this invention does not limit them.

[0131] It should be noted that the specific implementation method of the planning intelligent agent simulating the execution of candidate intelligent operation and maintenance processes can be set according to the operation and maintenance needs, and this invention does not limit it.

[0132] For example, in one embodiment, for each candidate intelligent operation and maintenance process with abnormal computing power, the planning agent can simulate the execution of the candidate intelligent operation and maintenance process through a sandbox exercise. That is, the planning agent can simulate and extrapolate each candidate intelligent operation and maintenance process in a simulation sandbox environment without affecting the real business, simulate the entire process operation of the candidate intelligent operation and maintenance process, and collect the simulation results of the candidate intelligent operation and maintenance process in order to score the candidate intelligent operation and maintenance process.

[0133] Step S4.3.1 Achieve beneficial technical effects: The planning agent simulates the execution of each candidate intelligent operation and maintenance process and scores the simulation results of each candidate intelligent operation and maintenance process based on the scoring index, objectively quantifying each intelligent operation and maintenance process and effectively improving the objectivity and standardization of intelligent operation and maintenance process evaluation.

[0134] Step S4.3.2: Based on the verification decision, the planning intelligent agent verifies at least one of the following: permissions, scope of operation, risk level, and compliance of the candidate intelligent operation and maintenance process with the highest score, and obtains the verification result.

[0135] It should be noted that the specific explanations of the permissions, scope of operation, risk level and compliance of the candidate intelligent operation and maintenance process can be found in the relevant descriptions in step S4.1 of the embodiments of the present invention, and will not be repeated here.

[0136] In an embodiment of the present invention, when the planning agent verifies at least one of the permissions, scope of operation, risk level, and compliance of the candidate intelligent operation and maintenance process with the highest score based on verification decision, it can verify each step of the candidate intelligent operation and maintenance process with the highest score to determine whether each step is safe and compliant.

[0137] The above verification result can be considered successful if every step of the candidate intelligent operation and maintenance process with the highest score passes the verification, or if any step fails the verification, the verification fails. The specific verification result can be determined according to the actual situation, and this invention does not limit it.

[0138] Step S4.3.2 Achieve beneficial technical effects: The planning intelligent agent performs multi-dimensional security verification on the candidate intelligent operation and maintenance process with the highest score, i.e. the optimal candidate intelligent operation and maintenance process, to eliminate unauthorized, insecure, and non-compliant operation and maintenance operations from the process planning stage, thereby ensuring the security of intelligent operation and maintenance.

[0139] Step S4.3.3: If the verification result indicates that the candidate intelligent operation and maintenance process with the highest score has failed the verification, the planning agent determines the reason for the verification failure based on the verification result.

[0140] The reasons for the above verification failure may include, but are not limited to, situations such as exceeding the permissions of steps in the intelligent operation and maintenance process, high risk level, exceeding the scope of operation, and non-compliance. The specific reasons can be determined according to the actual situation, and this invention does not limit them.

[0141] In an embodiment of the present invention, when the planning agent performs multi-dimensional security verification on the candidate intelligent operation and maintenance process with the highest score based on the verification decision, if the verification result indicates that the candidate intelligent operation and maintenance process has failed the verification, it can be determined that the candidate intelligent operation and maintenance process may have defects and potential risks and cannot be used directly. At this time, the reason for the verification failure can be determined based on the verification result so as to correct the candidate intelligent operation and maintenance process.

[0142] Step S4.3.3 achieves beneficial technical effects: by determining the reason for the failure of the candidate intelligent operation and maintenance process with the highest score, a basis is provided for subsequent correction of the candidate intelligent operation and maintenance process, thereby ensuring the security of intelligent operation and maintenance.

[0143] Step S4.3.4: Based on the reasons for verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the intelligent operation and maintenance process for abnormal computing power.

[0144] Step S4.3.4 Achieve beneficial technical effects: Based on the reasons for verification failure, the planning agent corrects the candidate intelligent operation and maintenance process with the highest score, solves the defects and potential risks of the candidate intelligent operation and maintenance process, and thus obtains a safe and compliant intelligent operation and maintenance process, which greatly improves the security of intelligent operation and maintenance of the intelligent computing cloud platform.

[0145] In one embodiment, step S4.3.4 includes: Step S4.3.4.1: Based on the reasons for the verification failure, the planning intelligent agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the modified candidate intelligent operation and maintenance process.

[0146] In an embodiment of the present invention, when the planning agent corrects the candidate intelligent operation and maintenance process with the highest score based on the verification failure reason, it can first determine the operation and maintenance parameters that need to be corrected in the candidate intelligent operation and maintenance process based on the verification failure reason, and then correct the corresponding operation and maintenance parameters based on the preset parameter correction rules.

[0147] The aforementioned parameter correction rules may include, but are not limited to, resource configuration corrections (such as task migration bandwidth thresholds, computing power resource allocation ratios, and healthy node load limits), security and compliance parameter corrections (such as operation and maintenance permission thresholds, unauthorized operation interception parameters, and data access encryption levels), anomaly handling parameter corrections (such as computing power isolation ranges, anomaly propagation blocking strengths, and driver overload execution levels), and business adaptation parameter corrections (task interruption tolerance duration, task migration start conditions, etc.). These rules can be set according to actual operation and maintenance needs, and this invention does not impose any limitations on them.

[0148] Step S4.3.4.1 Achieve beneficial technical effects: The planning agent corrects the candidate intelligent operation and maintenance process with the highest score based on the verification failure reason, accurately eliminates the defects and loopholes of the operation and maintenance process, improves the feasibility and security of the intelligent operation and maintenance process, and ensures the stable and reliable execution of subsequent operation and maintenance.

[0149] Step S4.3.4.2: The planning agent iteratively executes the verification process of the corrected candidate intelligent operation and maintenance process, and accumulates the number of iterations until the verification passes and the number of iterations is less than the preset maximum number of iterations, and then obtains the intelligent operation and maintenance process of abnormal computing power.

[0150] The maximum number of iterations can be set as needed. For example, the maximum number of iterations can be set to 5, but this invention does not limit this.

[0151] In an embodiment of the present invention, after the planning agent corrects the candidate operation and maintenance process with the highest score based on the reason for the verification failure, it can re-verify the corrected candidate intelligent operation and maintenance process, that is, iteratively execute the process verification to determine whether the corrected candidate intelligent operation and maintenance process is safe and compliant, and accumulate the number of iterations until the verification passes and the number of iterations is less than the maximum number of iterations. Then, the candidate intelligent operation and maintenance process that has passed the verification can be identified as the intelligent operation and maintenance process with abnormal computing power.

[0152] Step S4.3.4.2 Achieve beneficial technical effects: The planning agent performs iterative process verification on the corrected candidate intelligent operation and maintenance process, and obtains a safe and compliant intelligent operation and maintenance process for abnormal computing power when the verification passes and the number of iterations is less than the maximum number of iterations. This not only corrects the original defects of the intelligent operation and maintenance process, but also avoids introducing new defects during the correction process, ensuring the security and stability of the intelligent operation and maintenance process for abnormal computing power.

[0153] Step S4.3.4.3: If the number of iterations is greater than or equal to the maximum number of iterations, terminate the current process correction process and execute the downgrade processing mechanism to obtain the intelligent operation and maintenance process for abnormal computing power.

[0154] The aforementioned degradation processing mechanism can serve as a fallback mechanism for process correction, and it can be configured according to operational needs. For example, the degradation processing mechanism can be a manual review mechanism, that is, the operational process is manually reviewed and corrected; this invention does not limit this approach.

[0155] In an embodiment of the present invention, when the planning agent iteratively performs process verification on the modified candidate intelligent operation and maintenance process, if no intelligent operation and maintenance process that passes verification is obtained when the number of iterations is greater than or equal to the maximum number of iterations, the current process correction process can be terminated and a downgrade processing mechanism can be executed, such as manual process correction, to obtain an intelligent operation and maintenance process with abnormal computing power.

[0156] Step S4.3.4.3 achieves beneficial technical effects: When the planning agent performs iterative process verification on the corrected candidate intelligent operation and maintenance process, it terminates the current process correction process if the number of iterations is greater than or equal to the maximum number of iterations, thus avoiding the iteration from falling into an infinite loop. The intelligent operation and maintenance process that obtains abnormal computing power by executing the degradation processing mechanism can ensure the orderly progress of intelligent operation and maintenance work and improve the stability and fault tolerance of intelligent operation and maintenance.

[0157] In one embodiment, step S2 includes: Step S2.1: The perceptual agent performs semantic parsing on the text information in the computing power operation data, extracts key performance indicators from the computing power operation data, and obtains the parsing results and extraction results.

[0158] The aforementioned text information includes at least one of the computation task logs and alarm logs of heterogeneous computing power, and the aforementioned performance indicators include at least one of the hardware performance indicators, network traffic indicators, and application performance indicators.

[0159] For example, the computation task logs of heterogeneous computing power may include, but are not limited to, artificial intelligence task logs, edge computing task logs, and supercomputing task logs.

[0160] For example, hardware performance metrics may include, but are not limited to, CPU / GPU utilization, memory / disk I / O, hardware load, and device power consumption.

[0161] For example, network process metrics may include, but are not limited to, network traffic, transmission latency, and packet loss rate.

[0162] For example, application performance metrics may include, but are not limited to, task response latency, task success rate, AI inference / training time, and task interruption rate.

[0163] The above parsing results may include key information in the text information, including but not limited to computing power identifier, anomaly description information, alarm type, task running status or anomaly details, etc.

[0164] In embodiments of the present invention, the perceptual agent can deeply understand and associate the semantic information of multi-source heterogeneous data by performing semantic parsing on the text information in the computing power operation data. For example, it can distinguish between "normal fluctuations of instantaneous high load" and "abnormal conditions of continuous high load accompanied by specific fault logs", reduce the interference of normal invalid data, and accurately parse out the key information in the text information that may have problems.

[0165] The extraction results mentioned above may include key performance indicators that have problems, including but not limited to excessive memory usage, excessive CPU utilization, network bandwidth congestion, abnormal network traffic, excessively long task execution time, and increased task failure rate. The specific indicators can be determined according to the actual situation, and this invention does not limit them.

[0166] Step S2.1 Achieve beneficial technical effects: The perceptual agent performs semantic analysis on the text information in the computing power operation data of the intelligent computing cloud platform, extracts key performance indicators, and efficiently and accurately discovers hidden anomalies and explicit indicator degradation problems in the intelligent computing cloud platform. This reduces the interference of invalid data in the computing power operation data and improves the efficiency and accuracy of subsequent anomaly identification.

[0167] Step S2.2: The perceptual agent sends the parsing results and extraction results to the graph database.

[0168] Step S2.2 achieves beneficial technical effects: the perceptual agent sends the analysis results and extraction results to the graph database, providing accurate and reliable data support for subsequent anomaly identification, thereby improving the accuracy of intelligent operation and maintenance of the intelligent computing cloud platform.

[0169] Step S2.3: The graph database extracts the correlation features of the parsing results and extraction results and transforms them into the first semantic vector.

[0170] The aforementioned correlation features may include, but are not limited to, structured features such as entities, abnormal relationships, indicator degradation, abnormal causality, and business associations, which can be determined based on actual operation and maintenance conditions. For example, correlation features may be "GPU malfunction - GPU utilization exceeds limits, high memory usage - centralized deployment of high-load artificial intelligence tasks - AI training / inference tasks become sluggish", "Server driver failure - hardware response delay, abnormal process exit - forced device restart - computing task interruption", etc. This invention does not limit these features.

[0171] The first semantic vector mentioned above can be used to characterize the semantic meaning, association, and anomaly type of the relevance features.

[0172] It should be noted that the specific implementation method by which the graph database transforms relevance features into first semantic vectors can be set according to actual needs. For example, the graph database can transform relevance features into first semantic vectors through vector space mapping, and this invention does not limit this.

[0173] Step S2.3 achieves beneficial technical effects: The graph database extracts correlation features from the analysis results and extraction results and transforms them into computable semantic vectors, realizing deep association and fusion of unstructured text information and quantitative indicator data, providing conditions for improving the efficiency and accuracy of subsequent anomaly identification.

[0174] Step S2.4: The graph database matches the first semantic vector with the second semantic vector of the operation and maintenance causal knowledge graph of the intelligent computing cloud platform to determine whether there is a causal relationship graph with the computing power operation data.

[0175] The aforementioned second semantic vector can be a semantic vector pre-transformed from the causal knowledge graph of the operation and maintenance of the intelligent computing cloud platform.

[0176] It should be noted that the specific implementation method for matching the first semantic vector with the second semantic vector of the operation and maintenance causal knowledge graph of the intelligent computing cloud platform in the graph database can be set according to actual operation and maintenance needs. For example, semantic vector matching can be performed through similarity calculation, or it can be performed through distance calculation, etc. This invention does not limit this.

[0177] Step S2.4 achieves beneficial technical effects: The graph database can quickly determine whether the intelligent computing cloud platform has known anomalies through semantic vector matching, thereby improving the efficiency and accuracy of anomaly identification.

[0178] Step S2.5: In the absence of a matching causal relationship graph, calculate the information entropy based on the correlation characteristics of the computing power operation data to determine whether there is abnormal computing power in the intelligent computing cloud platform.

[0179] Step S2.5 achieves beneficial technical effects. In the absence of a matching causal relationship graph, it can be determined that there is no computing power with known anomalies in the intelligent computing cloud platform. In order to further determine whether there is computing power with unknown anomalies in the intelligent computing cloud platform, information entropy is calculated based on the correlation characteristics of computing power operation data. The degree of dispersion, distribution disorder and uncertainty in the correlation characteristics are quantified to achieve the identification of unknown anomalies and completely solve the problem that existing operation and maintenance technologies cannot identify unknown anomalies.

[0180] In one embodiment, step S3 includes: Step S3.1: When the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs anomaly detection on the correlation features corresponding to the abnormal computing power to determine the abnormal features of the abnormal computing power.

[0181] Step S3.1 achieves beneficial technical effects: The diagnostic agent identifies abnormal features by performing anomaly detection on the correlation characteristics of abnormal computing power, breaking the limitations of fixed rules and templates in the existing operation and maintenance technology of intelligent computing cloud platforms, accurately identifying unknown abnormal features of abnormal computing power, and providing accurate basis for subsequent root cause analysis and intelligent operation and maintenance process generation.

[0182] Step S3.2: The diagnostic agent performs root cause analysis based on abnormal features to determine the abnormal cause of abnormal computing power.

[0183] It should be noted that the diagnostic agent can not only determine the cause of anomalies through root cause analysis, but also identify unknown anomalies (such as GPU driver anomalies or memory leak patterns caused by specific novel models) through pattern matching.

[0184] Step S3.2 achieves beneficial technical effects: The diagnostic agent performs root cause analysis on abnormal features to determine the abnormal causes of abnormal computing power, especially unknown anomalies, thus completely solving the problem that the existing operation and maintenance technology of the intelligent computing cloud platform cannot identify unknown anomalies, and significantly improving the accuracy and comprehensiveness of anomaly identification.

[0185] In one embodiment, step S6 includes: Step S6.1: Evaluate the intelligent operation and maintenance results of the intelligent agent for obtaining abnormal computing power. The intelligent operation and maintenance results include the abnormal handling results and the actual operating status of the abnormal computing power after operation and maintenance management.

[0186] The actual operating status of the aforementioned computing power may include, but is not limited to, the operating status of computing hardware resources, computing tasks, cluster scheduling, resource load, etc.

[0187] Step S6.1 Achieves beneficial technical effects: In the intelligent operation and maintenance process of the intelligent computing cloud platform of the present invention, the intelligent operation and maintenance results of abnormal computing power are obtained by evaluating the intelligent agent, which provides a basis for subsequent measurement of the intelligent operation and maintenance effect and realization of the complete closed loop of intelligent operation and maintenance.

[0188] Step S6.2: Evaluate the normal operating status of the abnormal computing power acquired by the intelligent agent in the intelligent computing cloud platform, and determine the deviation between the actual operating status and the normal operating status.

[0189] The above-mentioned normal operating state can be considered as the steady-state baseline for the normal operation of computing power in the intelligent computing cloud platform.

[0190] Step S6.2 Achieve beneficial technical effects: The evaluation agent determines the actual operating status of the abnormal computing power after operation and maintenance management, and the deviation between it and its normal operating status in the intelligent computing cloud platform, in order to determine whether the abnormal computing power has recovered to the normal operating status, thereby achieving an objective measurement of the adaptability and execution effect of the intelligent operation and maintenance process and the abnormal computing power.

[0191] Step S6.3: If the anomaly handling results and deviations meet different preset conditions, perform subsequent operation and maintenance operations in a differentiated manner.

[0192] In embodiments of the present invention, when determining whether the abnormal processing result and the deviation meet preset conditions, a deviation threshold can be preset for the deviation (which can be set as needed and is not limited here). The condition can then be determined based on the relationship between the deviation and the deviation threshold, as well as the abnormal processing result. For example, if the deviation is less than the deviation threshold, it can be determined that the actual operating state of the abnormal computing power deviates little from its normal operating state in the intelligent computing cloud platform; if the deviation is greater than or equal to the deviation threshold, it can be determined that the actual operating state of the abnormal computing power deviates significantly from its normal operating state.

[0193] It should be noted that the preset conditions for the abnormal handling results and deviations to be satisfied, as well as the specific implementation methods of the subsequent operation and maintenance processing operations corresponding to the preset conditions, can be referred to the relevant description in step S6 of the embodiments of the present invention, and will not be repeated here.

[0194] Step S6.3 achieves beneficial technical effects: For different anomaly handling results and deviations, different preset conditions are matched, and the graph database is selectively updated, the operation and maintenance results are sent to the sensing agent, and the review mechanism is executed. This not only realizes the closed-loop management of the entire intelligent operation and maintenance chain, but also realizes the execution of corresponding operation and maintenance processing operations on demand. This ensures the dynamic updating of the operation and maintenance knowledge graph of the intelligent computing cloud platform, the timely synchronization of intelligent operation and maintenance results, and the automatic review of anomaly handling failures, thereby improving the accuracy and flexibility of intelligent operation and maintenance.

[0195] In one embodiment, the method further includes: Step S7: The sensing agent monitors the operation process of this intelligent operation and maintenance in real time.

[0196] The above-mentioned operation process can refer to the entire operation process of this intelligent operation and maintenance.

[0197] Step S7 achieves beneficial technical effects: the sensing agent monitors the operation process of this intelligent operation and maintenance in real time, realizes full-link controllability of intelligent operation and maintenance, can detect abnormal situations in the operation and maintenance process in a timely manner, and ensures that the intelligent operation and maintenance process proceeds in an orderly and stable manner.

[0198] Step S8: The sensing agent encrypts and writes at least one of the following into the audit log of the intelligent computing cloud platform: the operation and maintenance operation, the operation timestamp, the operation result, and the identifier of the computing power involved in the operation process.

[0199] It should be noted that the specific implementation method of data encryption by the sensing agent can be set according to actual operation and maintenance needs. For example, a strong encryption algorithm can be used for encryption, and this invention does not limit this.

[0200] It should be noted that the perception agent can also encrypt and write all data read and write operations, anomaly identification decision-making processes, relevant data snapshots, and internal reasoning paths of each agent into the audit log in real time.

[0201] Step S8 achieves beneficial technical effects: the sensing agent ensures the traceability, transparency, and security of the entire intelligent operation and maintenance process by encrypting all data in the operation process and writing it into the audit log of the intelligent computing cloud platform in real time.

[0202] Step S9: In the event of any abnormal operation during the operation process, cancel any operation by scheduling the intelligent agent, return to the step of the sensing intelligent agent to obtain computing power operation data, re-plan the process, and generate a new intelligent operation and maintenance process to perform operation and maintenance on the abnormal computing power of the intelligent computing cloud platform.

[0203] In one embodiment, after the intelligent operation and maintenance process is completed, the sensing agent can also feed back the anomaly handling result (such as the anomaly has been resolved and the server status has returned to normal) in encrypted form to the designated authorized object with the corresponding permissions, ensuring that the information is only visible to the authorized party, preventing information leakage, and improving the data security of intelligent operation and maintenance.

[0204] Step S9 achieves beneficial technical effects: by promptly rolling back abnormal operation and maintenance operations and replanning the intelligent operation and maintenance process, it enables rapid loss mitigation of abnormal operation and maintenance operations, corrects defects and problems in the intelligent operation and maintenance process, ensures the stable and reliable operation and maintenance of the entire chain, and improves the fault tolerance of intelligent operation and maintenance.

[0205] In summary, the main technical advantages of the intelligent operation and maintenance method provided by this invention are as follows: 1. Ultimate data security: Each intelligent agent, graph database, and decision engine are deployed locally. Sensitive operation and maintenance data are strictly transferred and stored within the cloud platform and are not transmitted externally. This ensures the highest level of data security and fully meets strict data localization and compliance requirements.

[0206] 2. Low latency and high efficiency: Operation and maintenance decisions and operations are completed locally, with no network latency, fast response and processing speed for anomalies, ensuring business continuity.

[0207] 3. High degree of customization and autonomous control: The integrated security management strategy can be highly customized and updated autonomously according to local needs, which is highly flexible and easier to adapt to the security compliance standards of various industries and enterprises.

[0208] The intelligent operation and maintenance method provided by this invention deeply integrates the "autonomous perception, autonomous planning, autonomous execution, and autonomous learning capabilities of intelligent agents" with an "integrated, refined, and full-process security control mechanism," achieving an original breakthrough in the field of intelligent computing cloud platform operation and maintenance. It solves the fundamental shortcomings of traditional operation and maintenance models in "responding to unknown anomalies" and "adapting to dynamic scenarios," while simultaneously overcoming the industry challenge of intelligent operation and maintenance and full-process security and compliance collaboration.

[0209] The intelligent operation and maintenance method provided by this invention has the following verifiable technical effects: Exception handling speed: Significantly shortens the time for handling complex and unknown exceptions compared to traditional methods.

[0210] Anomaly detection accuracy: Significantly improves accuracy, and basically eliminates false alarms and missed alarms.

[0211] Operation and maintenance labor costs: Significantly reduces labor costs and greatly reduces the need for manual on-call personnel.

[0212] Operation and maintenance security: Enables full traceability of operation and maintenance processes, significantly reducing the risk of sensitive data leakage.

[0213] The intelligent operation and maintenance method provided by this invention can be directly applied to all intelligent computing cloud platforms (such as model training and inference platforms) and can be seamlessly extended to edge computing clusters, supercomputing centers, financial intelligent platforms, and medical intelligent auxiliary diagnostic platforms with extremely high security and stability requirements. It requires no large-scale modification of existing hardware, is easy to deploy, and is readily and rapidly promoted for application. Furthermore, it can flexibly adapt to the stringent security and compliance standards of different industries. It possesses significant economic and social benefits and strong market competitiveness. Economic benefits: It saves enterprises huge amounts of operation and maintenance manpower costs, significantly reduces business losses caused by interruptions in artificial intelligence tasks, and significantly improves the operational efficiency and value of cloud platform assets.

[0214] Social benefits: Through a comprehensive and integrated security management design, data security incidents can be effectively prevented, and user privacy and the security of core corporate data assets can be protected.

[0215] Strong market competitiveness: The unique technical advantages of "autonomous operation and maintenance + integrated security management" provided by this invention accurately solve the core pain point of "difficulty in balancing operation and maintenance efficiency and security" in the industry, which will enable the cloud platform using this invention to have a strong differentiated competitiveness among similar products.

[0216] Please see Figure 2 , Figure 2 This is a system architecture framework diagram of the intelligent computing cloud platform that provides computing power based on the multi-agent intelligent operation and maintenance method provided by the present invention.

[0217] like Figure 2 As shown in the figure, the system architecture framework diagram of the intelligent computing cloud platform for providing computing power proposed in this invention based on the multi-agent intelligent operation and maintenance method includes an intelligent agent decision layer, a graph database and a decision engine. The intelligent agent decision layer includes a perception intelligent agent, a diagnosis intelligent agent, a planning intelligent agent, a scheduling intelligent agent and an evaluation intelligent agent with interactive relationships.

[0218] in, Figure 2 The interaction and control layer and the infrastructure layer shown represent the architectural layers of the intelligent computing cloud platform.

[0219] based on Figure 2 The system framework diagram shown in this invention, when implementing the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method, may include the following steps: 1. The sensing agent autonomously senses the computing power operation data of the intelligent computing cloud platform within the current time window from the operation and maintenance console, and sends the computing power operation data to the graph database.

[0220] 2. The graph database performs graph retrieval and matching based on computing power operation data, and sends the computing power operation data and the matched causal relationship graphs (if any) to the diagnostic agent.

[0221] 3. The diagnostic agent receives data sent from the graph database and performs root cause analysis to determine the abnormal cause of abnormal computing power in the intelligent computing cloud platform and sends it to the planning agent.

[0222] 4. The planning agent sends the cause of the anomaly to the decision engine to obtain the decision generated and returned by the decision engine, autonomously plans the intelligent operation and maintenance process for generating abnormal computing power, and sends it to the scheduling agent.

[0223] 5. The scheduling agent calls the corresponding operation and maintenance tools based on the intelligent operation and maintenance process, and sends operation and maintenance instructions to the K8s control plane through the called operation and maintenance tools, so that the K8s control plane can perform operation and maintenance operations on abnormal computing power in the computing power resource pool based on the operation and maintenance instructions.

[0224] 6. After the intelligent operation and maintenance process of abnormal computing power is completed, the evaluation agent can obtain and evaluate the intelligent operation and maintenance results, and based on the different preset conditions met by the evaluation results, perform the corresponding subsequent operation and maintenance operations in a differentiated manner, namely, selectively execute branch 1 to update the graph database, branch 2 to send the intelligent operation and maintenance results to the perception agent, and branch 3 to perform the review mechanism. The conditions that each branch needs to meet can be referred to the relevant descriptions in the above embodiments of the present invention, which will not be elaborated here.

[0225] Furthermore, the sensing agent can monitor the operation process of this intelligent operation and maintenance in real time. It can encrypt and write at least one of the operation and maintenance operations, operation timestamps, operation results and computing power identifiers involved in the operation process into the audit log of the intelligent computing cloud platform. In the event of any operation and maintenance operation failure during the operation process, the agent can cancel any operation and return to the step of the sensing agent obtaining the computing power operation data, and re-plan the process to generate a new intelligent operation and maintenance process to perform operation and maintenance on the abnormal computing power of the intelligent computing cloud platform.

[0226] It should be noted that, based on Figure 2 The system framework diagram shown implements the specific implementation of each step of the intelligent computing cloud platform for providing computing power proposed in this invention based on the multi-agent intelligent operation and maintenance method. You can refer to the relevant descriptions in the above embodiments of this invention, which will not be repeated here.

[0227] Please see Figure 3 , Figure 3 This is a flowchart illustrating the intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance method provided by the present invention.

[0228] like Figure 3 As shown, the intelligent computing cloud platform for providing computing power proposed in this invention is based on a multi-agent intelligent operation and maintenance method, which includes four stages: perception, diagnosis and planning, scheduling, execution and evaluation.

[0229] 1. Perception and Dynamic Mapping Stage: The perceptual agent acquires the computing power operation data within the current time window and searches the graph database for causal relationship graphs that match the computing power operation data in order to determine whether there is any known anomaly in the intelligent computing cloud platform.

[0230] If no matching causal relationship graph is found, it can be determined that there is no computing power with known anomalies in the intelligent computing cloud platform. To further determine whether there is computing power with unknown anomalies in the intelligent computing cloud platform, information entropy is calculated based on computing power operation data. If the information entropy is less than or equal to a threshold, it is determined that all computing power in the intelligent computing cloud platform is in normal operation. At this point, computing power operation data for the next time window can be obtained through a sliding window, returning to the perception and dynamic mapping stage, thus realizing continuous perception and monitoring of computing power in the intelligent computing cloud platform.

[0231] When the information entropy is greater than the threshold, it can be determined that there is computing power with unknown anomalies in the intelligent computing cloud platform. At this time, the anomaly identification and intelligent operation and maintenance process planning can be carried out through the multi-agent diagnosis and planning stage.

[0232] 2. Multi-agent Diagnosis and Planning Phase: The diagnostic agent performs root cause analysis based on the computing power operation data of abnormal computing power, determines the cause of the abnormal computing power, and sends it to the planning agent. The planning agent interacts with the decision engine to obtain intelligent operation and maintenance process generation and verification decisions, and generates multiple candidate intelligent operation and maintenance processes based on the intelligent operation and maintenance process generation decisions and the cause of the abnormality. Each candidate intelligent operation and maintenance process is simulated and executed through sand table exercises to obtain the candidate intelligent operation and maintenance process with the best execution effect. Based on the verification decisions, the candidate intelligent operation and maintenance process is subjected to multi-dimensional security verification.

[0233] If the verification fails, the candidate intelligent operation and maintenance process is modified and the process verification is iterated. The number of iterations is accumulated until the verification passes and the number of iterations is less than 5 (this maximum number of iterations is only an example and can be set as needed; no limit is imposed here). Then, the operation and maintenance operation is performed by calling the operation and maintenance tool through the hybrid scheduling phase.

[0234] Furthermore, if the number of iterations is greater than or equal to 5, the current process correction process is terminated, and a review mechanism is executed to obtain an intelligent operation and maintenance process for abnormal computing power.

[0235] 3. Hybrid Scheduling Phase: The scheduling agent calls the corresponding operation and maintenance tools based on the verified intelligent operation and maintenance process, and sends scheduling instructions to the K8s control plane through the operation and maintenance tools.

[0236] 4. Execution Evaluation Phase: The K8s control plane performs operation and maintenance operations on abnormal computing power in the computing power resource pool through the received scheduling instructions until the intelligent operation and maintenance process is completed. The evaluation agent obtains and evaluates the intelligent operation and maintenance results. If the evaluation results meet different preset conditions, it selectively executes branch 1 to update the graph database, branch 2 to send the intelligent operation and maintenance results to the perception agent, and branch 3 to perform the review mechanism.

[0237] Please see Figure 4 , Figure 4 This is a data interaction diagram of the intelligent computing cloud platform that provides computing power, based on a multi-agent intelligent operation and maintenance method, provided by the present invention.

[0238] like Figure 4 As shown, this interaction diagram illustrates the data interaction process between the acquisition probe, the sensing agent, the diagnostic agent, the planning agent, the decision engine, the scheduling agent, the K8s control plane, the computing nodes, and the evaluation agent.

[0239] Figure 4 The data interaction process shown may include the following steps: 1. The acquisition probe sends the computing power operation data of the intelligent computing cloud platform within the current time window to the sensing agent.

[0240] 2. The perceptual agent sends a dynamic causal graph to the diagnostic agent.

[0241] Specifically, the sensing agent receives computing power operation data sent by the acquisition probe and can send the computing power operation data to the graph database. The graph database can then extract relevant features based on the computing power operation data and match the extracted features with the operation and maintenance causal knowledge graph of the intelligent computing cloud platform to determine whether a matching causal association graph exists.

[0242] When a matching causal relationship graph exists, the correlation features and the causal relationship graph can be sent to the diagnostic agent through the graph database.

[0243] In the absence of a matching causal relationship graph, a causal graph corresponding to the relevant features can be generated from a graph database and sent to the diagnostic agent.

[0244] 3. The diagnostic agent sends a set of root cause hypotheses to the planning agent.

[0245] Specifically, the diagnostic agent receives the dynamic causal graph sent by the perceptual agent, performs root cause analysis on the causal graph, determines the abnormal causes of abnormal computing power in the intelligent computing cloud platform, i.e., the root cause hypothesis set, and sends it to the planning agent.

[0246] 4. The planning agent sends a decision acquisition request to the decision engine.

[0247] Specifically, the planning agent receives the set of root cause hypotheses, i.e., abnormal causes, sent by the diagnostic agent, and can generate a decision acquisition request based on the abnormal causes and send it to the decision engine.

[0248] 5. The decision engine returns the decision to the planning agent.

[0249] Specifically, in response to unknown anomalies, the decision engine receives a decision acquisition request sent by the planning agent, generates intelligent operation and maintenance process generation decisions and verification decisions based on the anomaly reasons in the decision acquisition request, and returns them to the planning agent.

[0250] For known anomalies, the decision engine can generate verification decisions based on the cause of the anomaly and return them to the planning agent.

[0251] 6. Plan the intelligent operation and maintenance process for the intelligent agent to send abnormal computing power to the scheduling intelligent agent.

[0252] Specifically, for unknown anomalies, the planning agent receives the intelligent operation and maintenance process generation and verification decisions returned by the decision engine. It can generate multiple candidate intelligent operation and maintenance processes based on the cause of the anomaly and the intelligent operation and maintenance process generation decisions. Then, it simulates the execution of each candidate intelligent operation and maintenance process to obtain the candidate intelligent operation and maintenance process with the best execution effect. After that, it performs multi-dimensional security verification on the candidate intelligent operation and maintenance process based on the verification strategy to obtain a safe and compliant intelligent operation and maintenance process and send it to the scheduling agent.

[0253] For known anomalies, the planning agent can perform multi-dimensional security verification on the historical intelligent operation and maintenance processes of the known anomalies based on the verification decisions returned by the decision engine, so as to obtain the safe and compliant historical intelligent operation and maintenance processes and send them to the scheduling agent.

[0254] 7. The scheduling agent sends scheduling instructions to the K8s control plane.

[0255] Specifically, the scheduling agent receives the intelligent operation and maintenance process sent by the planning agent, calls the corresponding operation and maintenance tools based on the intelligent operation and maintenance process, and sends scheduling instructions to the K8s control plane through the operation and maintenance tools.

[0256] 8. The K8s controller performs operation and maintenance operations on the computing nodes.

[0257] Specifically, the Kubernetes control plane receives scheduling instructions from the scheduling agent and can perform maintenance operations on abnormal computing power nodes, such as resource reallocation, based on the scheduling instructions.

[0258] 9. The computing power node sends the intelligent operation and maintenance results to the evaluation agent.

[0259] Specifically, after the intelligent operation and maintenance process is completed, the computing power node can send the intelligent operation and maintenance results to the evaluation agent. The intelligent operation and maintenance results can include the anomaly handling results and its actual operating status after operation and maintenance management.

[0260] 10. The evaluation agent assesses the results of intelligent operation and maintenance, and performs subsequent operation and maintenance operations in a differentiated manner based on the different conditions met by the evaluation results.

[0261] It should be noted that the specific implementation of each step in the data interaction process of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method in the embodiments of the present invention can be referred to the relevant descriptions in the above embodiments of the present invention, and will not be repeated here.

[0262] Please see Figure 5 , Figure 5 This is a schematic diagram of the system interface of the intelligent computing cloud platform that provides computing power based on the multi-agent intelligent operation and maintenance method provided by the present invention.

[0263] Figure 5 The system interface diagram includes: top status bar, central decision area, right-side tool area, left-side topology area, thought chain deduction window, decision verification matrix, and bottom audit area.

[0264] The top status bar above can visually display the current number of active alarms in the intelligent computing cloud platform.

[0265] The aforementioned central decision-making area can visually display the current progress bar of intelligent operation and maintenance.

[0266] The registry mentioned above can centrally manage various operation and maintenance tools.

[0267] The above-mentioned graph view can visualize causal relationship graphs, including but not limited to matching computing power operation data, or causal graphs generated from the correlation features corresponding to computing power operation data.

[0268] The aforementioned thought chain deduction window can visualize the reasoning and thinking process of each agent.

[0269] The decision verification matrix described above can be used to visually display the verification results of the intelligent operation and maintenance process.

[0270] The aforementioned audit logs can be used to store all data across the entire intelligent operation and maintenance chain.

[0271] Please see Figure 6 , Figure 6 This is a structural diagram of an intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance device, as provided by the present invention. Figure 6 As shown, the intelligent computing cloud platform providing computing power, based on a multi-agent intelligent operation and maintenance device 600, includes: Module 601 is used to build interactive intelligent agents, diagnostic intelligent agents, planning intelligent agents, scheduling intelligent agents, evaluation intelligent agents, graph databases, and decision engines for intelligent computing cloud platforms. The first processing module 602 is used to obtain computing power operation data within the current time window by the sensing agent, and to search the operation and maintenance causal knowledge graph in the graph database based on the computing power operation data for graph matching. If no matching causal association graph is found, the information entropy is calculated based on the computing power operation data. The information entropy is used to characterize the degree of abnormality in computing power operation. The determination module 603 is used to determine that there is abnormal computing power in the intelligent computing cloud platform when the information entropy is greater than the preset information entropy threshold. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power. The generation module 604 is used to plan the intelligent operation and maintenance process of the intelligent agent generating abnormal computing power based on the cause of the abnormality. The operation and maintenance module 605 is used to schedule intelligent agents to call the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on abnormal computing power according to the intelligent operation and maintenance process; The evaluation module 606 is used to evaluate the intelligent operation and maintenance results of the intelligent agent that acquires and evaluates abnormal computing power. When the evaluation results meet different preset conditions, the corresponding subsequent operation and maintenance processing operations are executed in a differentiated manner.

[0272] In one embodiment, the generation module 604 includes: The first sending unit is used to plan the intelligent agent to send the cause of the anomaly to the decision engine, so that the decision engine can generate and return the intelligent operation and maintenance process generation and verification decision based on the cause of the anomaly. The first generation unit is used to plan the intelligent agent to generate decisions and causes of anomalies based on the intelligent operation and maintenance process, and to plan and generate multiple candidate intelligent operation and maintenance processes for abnormal computing power. The simulation unit is used to plan the intelligent agent to simulate and verify each candidate intelligent operation and maintenance process based on the verification decision, and to obtain the intelligent operation and maintenance process of abnormal computing power. The intelligent operation and maintenance process includes at least one of the following operations: isolating abnormal computing power, scheduling computing tasks on abnormal computing power, generating maintenance work orders, and rolling back the driver of abnormal computing power.

[0273] In one embodiment, the above simulation unit is specifically used for: For each candidate intelligent operation and maintenance process with abnormal computing power, an intelligent agent is planned to simulate its execution, and the simulation results of the candidate intelligent operation and maintenance process are scored based on a preset scoring standard. Based on the verification decision, the planning intelligent agent verifies at least one of the following: the permissions, scope of operation, risk level, and compliance of the candidate intelligent operation and maintenance process with the highest score, and obtains the verification result. If the verification result indicates that the candidate intelligent operation and maintenance process with the highest score has failed the verification, the planning agent determines the reason for the verification failure based on the verification result. Based on the reasons for verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the intelligent operation and maintenance process with abnormal computing power.

[0274] In one embodiment, the above-mentioned simulation unit is further used for: Based on the reasons for verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the modified candidate intelligent operation and maintenance process. The planning agent iteratively executes the verification process of the corrected candidate intelligent operation and maintenance process, and accumulates the number of iterations until the verification passes and the number of iterations is less than the preset maximum number of iterations, then obtains the intelligent operation and maintenance process of abnormal computing power. If the number of iterations is greater than or equal to the maximum number of iterations, terminate the current process correction process and execute the degradation handling mechanism to obtain the intelligent operation and maintenance process for abnormal computing power.

[0275] In one embodiment, the first processing module 602 includes: The first processing unit is used to perform semantic parsing of text information in computing power operation data by the sensing agent, extract key performance indicators from the computing power operation data, and obtain parsing results and extraction results. The text information includes at least one of the computing task logs and alarm logs of heterogeneous computing power, and the performance indicators include at least one of hardware performance indicators, network traffic indicators and application performance indicators. The second sending unit is used for the sensing agent to send the parsing results and extraction results to the graph database; The second processing unit is used to extract relevant features from the parsing results and extraction results in the graph database and convert them into the first semantic vector. The matching unit is used to match the first semantic vector of the graph database with the second semantic vector of the operation and maintenance causal knowledge graph of the intelligent computing cloud platform to determine whether there is a causal relationship graph with the computing power operation data. The computing unit is used to calculate information entropy based on the correlation characteristics of computing power operation data in the absence of a matching causal relationship graph, in order to determine whether there is abnormal computing power in the intelligent computing cloud platform.

[0276] In one embodiment, the determining module 603 includes: The detection unit is used to determine that there is abnormal computing power in the intelligent computing cloud platform when the information entropy is greater than the preset information entropy threshold. The diagnostic agent performs abnormal detection on the correlation features corresponding to the abnormal computing power and determines the abnormal features of the abnormal computing power. The analysis unit is used to diagnose the intelligent agent by performing root cause analysis based on abnormal features to determine the abnormal cause of abnormal computing power.

[0277] In one embodiment, the evaluation module 606 includes: The acquisition unit is used to evaluate the intelligent operation and maintenance results of the intelligent agent acquiring abnormal computing power. The intelligent operation and maintenance results include the abnormal handling results and the actual operating status of the abnormal computing power after operation and maintenance management. The determination unit is used to evaluate the normal operating status of the intelligent agent acquiring abnormal computing power on the intelligent computing cloud platform and to determine the deviation between the actual operating status and the normal operating status. The execution unit is used to evaluate the agent's differentiated execution of subsequent operation and maintenance operations when the abnormal handling results and deviations meet different preset conditions.

[0278] In one embodiment, the aforementioned subsequent operation and maintenance processing includes at least one of updating the graph database, sending intelligent operation and maintenance results to the sensing agent, and executing a review mechanism.

[0279] In one embodiment, the device 600 further includes: The monitoring module (not shown in the figure) is used by the sensing agent to monitor the operation process of this intelligent operation and maintenance in real time. The writing module (not shown in the figure) is used to encrypt and write at least one of the following into the audit log of the intelligent computing cloud platform: the operation and maintenance operation, the operation timestamp, the operation result and the identifier of the computing power involved in the operation process. The second processing module (not shown in the figure) is used to cancel any operation and maintenance operation by scheduling the intelligent agent in the event of any abnormal operation and maintenance operation during the operation process, return to the step of obtaining computing power operation data by the sensing intelligent agent, re-plan the process, and generate a new intelligent operation and maintenance process to perform operation and maintenance on the abnormal computing power of the intelligent computing cloud platform.

[0280] The intelligent computing cloud platform providing computing power provided by this invention is based on a multi-agent intelligent operation and maintenance device. It can realize the various processes of the various embodiments of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method described above. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0281] It should be noted that the intelligent computing cloud platform that provides computing power in this invention is based on a multi-agent intelligent operation and maintenance device, which can be a device, or a component, integrated circuit, or chip in an electronic device.

[0282] The present invention also provides an electronic device, see below. Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 701, a processor 702, and a program or instructions stored in the memory 701 that run on the memory. When the program or instructions are executed by the processor 702, they can achieve the following: Figure 1The corresponding intelligent computing cloud platform that provides computing power is based on any step in the multi-agent intelligent operation and maintenance method embodiment and achieves the same beneficial effect, which will not be elaborated here.

[0283] The processor 702 can be a CPU, ASIC, FPGA, or GPU.

[0284] Those skilled in the art will understand that all or part of the steps of the above-described intelligent computing cloud platform based on the multi-agent intelligent operation and maintenance method embodiment can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0285] The present invention also provides a readable storage medium on which a computer program is stored, and which, when executed by a processor, can perform the above-described functions. Figure 1 The corresponding intelligent computing cloud platform that provides computing power is based on any step in the multi-agent intelligent operation and maintenance method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here. The storage medium mentioned is such as read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0286] The present invention also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described... Figure 1 The corresponding intelligent computing cloud platform that provides computing power is based on the various processes of the multi-agent intelligent operation and maintenance method implementation, and can achieve the same technical effect. To avoid repetition, it will not be described in detail here.

[0287] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: A alone, B alone, C alone, both A and B present, both B and C present, both A and C present, and A, B, and C present.

[0288] It should be noted that, in this document, 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. 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.

[0289] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or second terminal device, etc.) to execute the methods of the various embodiments of this application.

[0290] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance method, comprising: Step S1: The intelligent computing cloud platform constructs a perceptual intelligent agent, a diagnostic intelligent agent, a planning intelligent agent, a scheduling intelligent agent, an evaluation intelligent agent, a graph database, and a decision engine with interactive relationships; Step S2: The sensing agent acquires computing power operation data within the current time window, and based on the computing power operation data, searches the operation and maintenance causal knowledge graph in the graph database for graph matching. If no matching causal association graph is found, the information entropy is calculated based on the computing power operation data. The information entropy is used to characterize the degree of computing power operation abnormality. Step S3: If the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power. Step S4: The planning agent generates an intelligent operation and maintenance process for the abnormal computing power based on the cause of the abnormality. Step S5: The scheduling agent calls the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process; Step S6: The evaluation agent acquires and evaluates the intelligent operation and maintenance results of the abnormal computing power. When the evaluation results meet different preset conditions, the agent performs the corresponding subsequent operation and maintenance operations in a differentiated manner.

2. The method as described in claim 1, characterized in that, Step S4 includes: Step S4.1: The planning agent sends the cause of the anomaly to the decision engine, so that the decision engine generates and returns intelligent operation and maintenance process generation and verification decisions based on the cause of the anomaly. Step S4.2: The planning agent generates multiple candidate intelligent operation and maintenance processes for the abnormal computing power based on the intelligent operation and maintenance process and the cause of the anomaly. Step S4.3: The planning agent performs simulation verification on each candidate intelligent operation and maintenance process based on the verification decision to obtain the intelligent operation and maintenance process of the abnormal computing power. The intelligent operation and maintenance process includes at least one of the following operations: isolating the abnormal computing power, scheduling computing tasks on the abnormal computing power, generating maintenance work orders, and rolling back the driver of the abnormal computing power.

3. The method as described in claim 2, characterized in that, Step S4.3 includes: Step S4.3.1: For each candidate intelligent operation and maintenance process of the abnormal computing power, the planning agent simulates the execution and scores the simulation results of the candidate intelligent operation and maintenance process based on a preset scoring standard. Step S4.3.2: Based on the verification decision, the planning intelligent agent verifies at least one of the following: permissions, operation scope, risk level, and compliance of the candidate intelligent operation and maintenance process with the highest score, and obtains the verification result. Step S4.3.3: If the verification result indicates that the candidate intelligent operation and maintenance process with the highest score has failed the verification, the planning intelligent agent determines the reason for the verification failure based on the verification result. Step S4.3.4: Based on the reason for the verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the intelligent operation and maintenance process for the abnormal computing power.

4. The method as described in claim 3, characterized in that, Step S4.3.4 includes: Step S4.3.4.1: Based on the reasons for the verification failure, the planning agent modifies the candidate intelligent operation and maintenance process with the highest score to obtain the modified candidate intelligent operation and maintenance process. Step S4.3.4.2: The planning agent iteratively executes process verification on the corrected candidate intelligent operation and maintenance process, and accumulates the number of iterations until the verification is passed and the number of iterations is less than the preset maximum number of iterations, and then obtains the intelligent operation and maintenance process of the abnormal computing power. Step S4.3.4.3: If the number of iterations is greater than or equal to the maximum number of iterations, terminate the current process correction process and execute the downgrade processing mechanism to obtain the intelligent operation and maintenance process of the abnormal computing power.

5. The method as described in claim 1, characterized in that, Step S2 includes: Step S2.1: The perceptual agent performs semantic parsing on the text information in the computing power operation data, extracts key performance indicators from the performance indicators in the computing power operation data, and obtains parsing results and extraction results. The text information includes at least one of the computing task logs and alarm logs of heterogeneous computing power, and the performance indicators include at least one of hardware performance indicators, network traffic indicators, and application performance indicators. Step S2.2: The perceptual agent sends the analysis result and the extraction result to the graph database; Step S2.3: The graph database extracts correlation features between the parsing results and the extraction results and converts them into a first semantic vector; Step S2.4: The graph database matches the first semantic vector with the second semantic vector of the operation and maintenance causal knowledge graph of the intelligent computing cloud platform to determine whether there is a causal association graph with the computing power operation data. Step S2.5: In the absence of a matching causal relationship graph, calculate the information entropy based on the correlation characteristics of the computing power operation data to determine whether there is abnormal computing power in the intelligent computing cloud platform.

6. The method as described in claim 5, characterized in that, Step S3 includes: Step S3.1: When the information entropy is greater than the preset information entropy threshold, it is determined that there is abnormal computing power in the intelligent computing cloud platform. The diagnostic agent performs abnormal detection on the correlation features corresponding to the abnormal computing power to determine the abnormal features of the abnormal computing power. Step S3.2: The diagnostic agent performs root cause analysis based on the abnormal features to determine the abnormal cause of the abnormal computing power.

7. The method as described in claim 1, characterized in that, Step S6 includes: Step S6.1: The evaluation agent obtains the intelligent operation and maintenance results of the abnormal computing power, the intelligent operation and maintenance results include the abnormal handling results and the actual operating status of the abnormal computing power after operation and maintenance management; Step S6.2: The evaluation agent obtains the normal operating status of the abnormal computing power in the intelligent computing cloud platform and determines the deviation between the actual operating status and the normal operating status; Step S6.3: When the anomaly handling result and the deviation meet different preset conditions, the evaluation agent performs subsequent operation and maintenance operations in a differentiated manner.

8. The method as described in claim 1 or 7, characterized in that, The subsequent operation and maintenance processing includes at least one of updating the graph database, sending the intelligent operation and maintenance results to the sensing agent, and executing a review mechanism.

9. The method as described in claim 1, characterized in that, The method further includes: Step S7: The sensing agent monitors the operation process of this intelligent operation and maintenance in real time. Step S8: The sensing agent encrypts and writes at least one of the following into the audit log of the intelligent computing cloud platform: the operation and maintenance operation, the operation timestamp, the operation result, and the identifier of the computing power involved in the operation process. Step S9: In the event of any abnormal operation during the operation process, the scheduling agent cancels any operation, returns to the step of the sensing agent obtaining the computing power operation data, and re-plans the process to generate a new intelligent operation process to perform operation and maintenance on the abnormal computing power of the intelligent computing cloud platform.

10. An intelligent computing cloud platform providing computing power based on a multi-agent intelligent operation and maintenance device, characterized in that, include: The building module is used to construct interactive intelligent agents, diagnostic intelligent agents, planning intelligent agents, scheduling intelligent agents, evaluation intelligent agents, graph databases, and decision engines for intelligent computing cloud platforms; The first processing module is used for the sensing agent to obtain computing power operation data within the current time window, and based on the computing power operation data, to search the operation and maintenance causal knowledge graph in the graph database for graph matching. If no matching causal association graph is found, the module calculates information entropy based on the computing power operation data. The information entropy is used to characterize the degree of computing power operation abnormality. The determination module is used to determine that there is abnormal computing power in the intelligent computing cloud platform when the information entropy is greater than a preset information entropy threshold. The diagnostic agent performs root cause analysis based on the computing power operation data to determine the abnormal cause of the abnormal computing power. The generation module is used by the planning agent to generate an intelligent operation and maintenance process for the abnormal computing power based on the cause of the abnormality. The operation and maintenance module is used by the scheduling agent to call the corresponding intelligent operation and maintenance tools to perform intelligent operation and maintenance operations on the abnormal computing power according to the intelligent operation and maintenance process; The evaluation module is used by the evaluation agent to obtain and evaluate the intelligent operation and maintenance results of the abnormal computing power, and to perform corresponding subsequent operation and maintenance operations in a differentiated manner when the evaluation results meet different preset conditions.

11. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, it implements the steps of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method as described in any one of claims 1 to 9.

13. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the steps of the intelligent computing cloud platform providing computing power based on the multi-agent intelligent operation and maintenance method as described in any one of claims 1 to 9.

Images