Cloud computing platform intelligent operation and maintenance method and system based on knowledge graph

Abstract

The invention discloses a cloud computing platform intelligent operation and maintenance method and system based on a knowledge graph, and relates to the technical field of cloud computing platform intelligent operation and maintenance, and the method comprises the following steps: a data collection and preprocessing step: collecting platform hardware operation, software logs, service calling and network transmission data; a knowledge graph construction and dynamic updating step: defining the entity type of the operation and maintenance field; a multi-dimensional real-time monitoring step: building a four-dimensional monitoring model based on the knowledge graph; an abnormity intelligent detection and positioning step of fusing association reasoning and a machine learning algorithm to detect abnormity; an operation and maintenance decision intelligent generation step: matching historical cases and combining resource loads; an operation and maintenance execution and effect feedback step: executing the operation and maintenance plan and monitoring the effect; and an operation and maintenance knowledge precipitation and optimization step: extracting operation and maintenance knowledge rules and updating the operation and maintenance knowledge rules to the knowledge graph. According to the method, full-process intelligent operation and maintenance of the cloud computing platform are realized based on the knowledge graph, resource utilization is balanced and optimized, and the platform stability and the operation and maintenance efficiency are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for cloud computing platforms, and in particular to an intelligent operation and maintenance method and system for cloud computing platforms based on knowledge graphs. Background Technology

[0002] With the rapid development of cloud computing technology and the widespread adoption of various service models such as IaaS, SaaS, and PaaS, cloud computing platforms have gradually become the core infrastructure supporting industry applications and social services. The scale of platforms continues to expand, with the number of hardware resources such as servers, virtual machines, and containers, as well as software components such as application services and network devices, growing exponentially, forming a complex and tightly interconnected operations and maintenance (O&M) system. The surge in multi-source O&M data, including hardware operation data, software log data, service call data, and network transmission data, has significantly increased the complexity and difficulty of O&M work. Traditional manual O&M models rely on the experience and judgment of O&M personnel, which is not only inefficient but also struggles to handle challenges such as real-time processing of massive amounts of data and locating complex interconnected problems. Human error can easily lead to delays in fault handling, affecting the stability and reliability of platform services.

[0003] While existing cloud computing platform operation and maintenance technologies have incorporated automation tools and machine learning algorithms, significant shortcomings remain, failing to form a complete intelligent operation and maintenance closed loop. At the data processing level, various types of operation and maintenance data are stored in a scattered manner, lacking a unified standardized processing mechanism. Missing values, outliers, and duplicate data affect the accuracy of subsequent analysis, and data of different dimensions are difficult to compare effectively, failing to comprehensively reflect the platform's operational status. At the anomaly detection and localization level, most technologies only monitor single indicators, lacking in-depth mining of data correlations, making it difficult to locate the root cause and propagation path of anomalies, often resulting in "treating the symptoms but not the root cause." At the operation and maintenance decision-making level, existing systems rely heavily on preset rules, failing to fully integrate historical operation and maintenance cases and real-time resource load conditions. The generated operation and maintenance solutions have poor adaptability, with prominent issues of uneven resource allocation, and a frequent mismatch between core business operation and maintenance needs and resource supply.

[0004] While the initial application of knowledge graph technology in the operations and maintenance (O&M) field has solved some association analysis problems, many shortcomings still limit its effectiveness. Existing O&M solutions based on knowledge graphs often rely on static knowledge graph construction, resulting in lagging updates to entity attributes and relationships. This makes it difficult to respond promptly to platform configuration changes and new O&M data, leading to a disconnect between inference results and actual operational status. Furthermore, the lack of a quantitative evaluation mechanism for entity importance results in low knowledge graph inference efficiency, hindering rapid support for O&M decisions in complex scenarios. In addition, the O&M knowledge accumulation mechanism is inadequate; valuable data such as historical cases and anomaly handling solutions have not been effectively transformed into reusable knowledge rules, making it difficult for platform O&M capabilities to achieve iterative improvement through continuous learning. As cloud computing platforms increasingly demand stability, security, and efficiency, the shortcomings of existing O&M technologies in areas such as data processing accuracy, anomaly detection speed, decision-making intelligence, and knowledge reusability are becoming increasingly apparent. There is an urgent need for a comprehensive intelligent O&M method and system that integrates knowledge graph technology with cloud computing O&M needs to address the industry pain points of "disorganized data, difficult location, slow decision-making, and scattered knowledge" in current O&M work. Summary of the Invention

[0005] This invention proposes an intelligent operation and maintenance method and system for a cloud computing platform based on knowledge graphs, in order to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs, comprising the following steps: The data collection and preprocessing steps involve collecting four types of data from the platform: hardware, software, services, and network. Invalid data is removed using a cleaning algorithm, and then the data is standardized to convert the data dimensions and generate a structured operation and maintenance dataset. The knowledge graph construction and dynamic update steps are as follows: define entity types and relationships in the operation and maintenance domain based on standardized datasets, construct an initial knowledge graph using a graph database, and automatically update graph attributes, relationships, and weights according to a preset mechanism when data meets the standards or configuration changes. The multi-dimensional real-time monitoring process relies on a knowledge graph to build a four-dimensional monitoring model, sets indicator thresholds, collects data every second through real-time components, and analyzes fluctuation trends by comparing with historical data. The anomaly intelligent detection and localization steps integrate correlation reasoning and machine learning algorithms to detect anomalies. When the indicators exceed the threshold, an alarm is triggered. The graph is traversed to locate core nodes, trace the propagation path, and form an anomaly event archive. The intelligent generation steps for operation and maintenance decisions are as follows: matching historical cases to obtain candidate solutions, combining resource load assessment for feasibility, selecting the optimal solution and generating a detailed operation and maintenance plan; Operation and maintenance execution and effect feedback steps: distribute the plan to the execution components, monitor the execution progress and effect, collect post-operation and maintenance data to evaluate the effectiveness, and store relevant data in the knowledge base; The steps for accumulating and optimizing operational knowledge include: organizing the knowledge base content, refining operational knowledge rules, updating them to the graph, and optimizing monitoring thresholds and algorithm parameters based on data trends.

[0007] Furthermore, it also includes an abnormal risk level assessment step, which calculates the risk level of abnormal events by constructing an abnormal risk assessment formula, as follows: in The risk level is abnormal. The weighting coefficient for abnormal effects; This is the propagation speed weighting coefficient; The duration weighting coefficient; The number of entities affected by the anomaly; For the first Importance coefficient of each affected entity; For the first The correlation weight of each affected entity; This represents the number of time segments in which the anomaly propagated. For the first The abnormal propagation rate of each time segment; For the first The duration of each time segment; The duration of the abnormality.

[0008] Furthermore, it also includes a dynamic allocation step for operation and maintenance resources. Based on the real-time resource load of the cloud computing platform and the operation and maintenance task requirements, operation and maintenance resources are allocated reasonably. During the resource allocation process, priority is given to the operation and maintenance tasks corresponding to core businesses. After the resource allocation is completed, the resource usage is monitored in real time.

[0009] Furthermore, in the knowledge graph construction and dynamic update steps, an entity weight calculation mechanism is adopted. By constructing an entity importance assessment formula, the priority of entities in operation and maintenance decisions is quantified. The formula is as follows: in This is the entity importance assessment value; Business dependence of the entity; This refers to the number of historical failures of the entity. This represents the resource occupancy coefficient of the entity. The deployment level for entities.

[0010] Furthermore, in the anomaly intelligent detection and localization step, the isolated forest algorithm and knowledge graph association reasoning are integrated for anomaly detection. The isolated forest algorithm clusters the operation and maintenance data by constructing multiple isolated trees to identify abnormal data points that deviate from the normal data cluster. The knowledge graph association reasoning locates the root cause of the anomaly by traversing the entity association paths corresponding to the abnormal data points.

[0011] Furthermore, in the intelligent generation step of operation and maintenance decision-making, an operation and maintenance solution adaptability assessment mechanism is established, which evaluates the solution from three dimensions: technology adaptability, resource adaptability, and time adaptability. Technology adaptability assesses the degree of matching between the operation and maintenance solution and the cloud computing platform architecture and version; resource adaptability assesses the degree of compatibility between the resources required for the solution execution and the existing available resources on the platform; and time adaptability assesses the matching between the execution time of the solution and the allowable duration of anomalies.

[0012] Furthermore, in the step of knowledge accumulation and optimization for operations and maintenance, knowledge distillation technology is used to extract core operations and maintenance knowledge, distilling complex operations and maintenance cases and decision data into concise knowledge rules. During the knowledge distillation process, key operations and maintenance parameters and operation steps are retained, while redundant information is removed. Operations and maintenance experts are invited to review the extracted knowledge rules. Knowledge rules that pass the review are updated to the knowledge graph, while knowledge rules that fail the review are returned to the knowledge base for re-extraction.

[0013] Furthermore, it includes the following modules: The data acquisition and preprocessing module is configured with multiple types of acquisition sensors and log components to collect platform hardware, software, service, and network data. It cleans invalid data and standardizes the format through built-in algorithms to generate structured datasets. The knowledge graph construction and update module uses entity definition tools and relationship sorting components to define entity types and relationships in the operation and maintenance domain, uses a graph database to build an initial knowledge graph, and uses an automatic update trigger component to dynamically update the graph information based on data volume and configuration changes. The real-time monitoring module builds a four-dimensional monitoring model and presets indicator thresholds. It is configured to collect data every second using real-time acquisition components and compares real-time and historical data to capture fluctuation trends. The anomaly detection and localization module integrates the isolated forest algorithm component and the knowledge graph reasoning engine to detect abnormal data and trigger alarms. It uses graph traversal to traverse related information, locates abnormal nodes, traces the propagation path, and forms an event archive. The operation and maintenance decision generation module connects the knowledge base and resource monitoring components, matches similar processing solutions, evaluates feasibility from four dimensions, selects the optimal solution, and generates a detailed operation and maintenance plan. The operation and maintenance execution feedback module is equipped with execution components to support automatic and manual execution, monitor execution progress, collect post-operation and maintenance data to evaluate the effect, and store relevant data in the knowledge base; The knowledge accumulation and optimization module extracts operation and maintenance knowledge rules through built-in tools and updates them to the graph. It optimizes thresholds and algorithm parameters based on data trends and integrates an expert review interface. The data storage module uses a distributed storage architecture to store various operational and maintenance data, is configured with a backup component for regular backups, and supports fast data retrieval and retrieval.

[0014] Furthermore, it also includes a resource load balancing module, which monitors the CPU utilization, memory usage, disk read / write speed and network bandwidth utilization of the cloud computing platform in real time. When the utilization of a single resource exceeds 80%, the load balancing mechanism is activated to migrate the overloaded tasks to nodes with resource utilization below 40% and records the load migration log.

[0015] Furthermore, it also includes a security protection module, with a built-in data encryption component to encrypt the collected operation and maintenance data and the stored knowledge base data. It adopts an access control mechanism to divide users into three levels: administrators, operation and maintenance personnel, and viewers. Different levels of users have different data access and operation permissions.

[0016] Compared with existing technologies, the beneficial effects of this invention are: At the data processing and knowledge support level, this invention, through standardized data collection and preprocessing procedures, systematically eliminates invalid data and unifies data formats, significantly improving the quality and availability of operational data and providing a solid data foundation for subsequent operational analysis. The knowledge graph construction and dynamic update mechanism enables real-time iteration of entity attributes, relationships, and weight information, completely resolving the reasoning lag problem caused by the static nature of traditional knowledge graphs. Simultaneously, the entity importance assessment mechanism optimizes the reasoning efficiency of the knowledge graph, making operational analysis more targeted and accurately focusing on entities and relationships related to core business needs.

[0017] At the anomaly detection and operational decision-making level, the anomaly detection solution, which integrates knowledge graph-based reasoning and machine learning algorithms, overcomes the limitations of single-indicator monitoring. It can accurately locate core anomaly nodes, clearly trace anomaly propagation paths, and effectively define the scope of anomaly impact, changing the current situation where existing technologies prioritize detection over localization. The intelligent operational decision-making mechanism combines historical cases and real-time resource status to evaluate the feasibility of solutions from multiple dimensions. The selected optimal operational plan has stronger adaptability, avoiding resource waste and business impact caused by blind operations. Simultaneously, the closed-loop operation execution and effect feedback ensures that every operational operation is accurately implemented and its effectiveness verified.

[0018] At the resource scheduling and knowledge accumulation level, the dynamic allocation mechanism for operation and maintenance resources prioritizes core business needs. Simultaneously, through resource reclamation and redundancy reservation strategies, it achieves efficient resource utilization and dynamic balance, resolving the issues of uneven resource allocation and wasted idle resources in existing platforms. The knowledge accumulation and optimization mechanism extracts core operation and maintenance rules through knowledge distillation technology, combined with expert review and dynamic optimization, continuously enriching and improving the operation and maintenance knowledge base, driving the continuous iterative upgrade of the platform's operation and maintenance capabilities. Furthermore, the system's resource load balancing module and security protection module further enhance the stability and security of the cloud computing platform, comprehensively ensuring its continuous and efficient operation.

[0019] This invention not only adapts to various cloud computing service models and meets the operation and maintenance needs of platforms of different sizes, but also provides a complete technical framework of "data-knowledge-decision-execution-accumulation" for the field of cloud computing operation and maintenance, promoting the transformation of cloud computing operation and maintenance from traditional manual operation and semi-automated management to intelligent operation and maintenance throughout the entire process, and has broad industry application value. Attached Figure Description

[0020] Figure 1 This is a schematic block diagram of an intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs proposed in this invention. Figure 2 This is a schematic block diagram of a system for an intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs, as proposed in this invention. Figure 3 A bar chart comparing the average fault location time under different operation and maintenance modes; Figure 4 This is a line graph showing the monthly trend of resource utilization after the implementation of this invention. Figure 5 A bar chart comparing tenant complaint rates for different solutions in a public cloud scenario. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0024] Reference Figures 1 to 5 A knowledge graph-based intelligent operation and maintenance method for cloud computing platforms includes the following steps: Data collection and preprocessing steps: Collect hardware operation data, software log data, service call data, and network transmission data from the cloud computing platform. Hardware operation data includes server CPU utilization, memory usage, disk read / write speed, and network bandwidth utilization. Software log data includes application error logs, process execution logs, and system event logs. Service call data involves microservice interface call frequency, response time, and call success rate. Network transmission data includes packet loss rate, latency, and number of connections. Data cleaning algorithms are used to remove missing values, outliers, and duplicate data. Data standardization processing is used to convert data of different dimensions into standardized data of a unified range, generating a structured operation and maintenance dataset. The knowledge graph construction and dynamic update steps are as follows: Based on a standardized operation and maintenance dataset, define entity types in the cloud computing platform operation and maintenance domain, including server entities, virtual machine entities, container entities, application service entities, and network device entities. Identify the relationships between entities, covering dependency relationships, call relationships, deployment relationships, and topology relationships. Use a graph database to store entities and their relationships to construct an initial operation and maintenance knowledge graph. Set up a knowledge update trigger mechanism so that when new operation and maintenance data reaches a preset threshold or platform configuration changes, automatically update entity attributes and relationships in the knowledge graph, and synchronously update entity weights and relationship strengths. Multi-dimensional real-time monitoring steps; a monitoring model is built based on the operation and maintenance knowledge graph, and monitoring indicators are set in four dimensions: hardware, software, service and network. Normal threshold range and warning threshold range are set for each indicator. The monitoring indicator data is collected once per second through the real-time data acquisition component. The collected real-time data is compared and analyzed with the historical normal data in the knowledge graph to capture the data fluctuation trend. Anomaly detection and localization steps: Integrating the associative reasoning capabilities of knowledge graphs and machine learning algorithms, anomaly detection is performed on the monitoring data. When the monitoring indicator data exceeds the warning threshold range, an anomaly alarm is triggered. By traversing all entities and relationships associated with the anomaly indicator through the knowledge graph, the core node of the anomaly is located, the anomaly propagation path is traced, the scope of the anomaly's impact is clarified, and anomaly feature information is extracted to form an anomaly event archive. The intelligent generation steps for operation and maintenance decisions are as follows: Based on the historical operation and maintenance cases in the abnormal event archive and knowledge graph, similar abnormal handling solutions are matched. Combined with the real-time operation and maintenance resource load, the feasibility of candidate solutions is evaluated. The evaluation dimensions include processing time, resource consumption, impact scope and success rate. The optimal operation and maintenance decision solution is selected from these and a detailed operation and maintenance plan including operation steps, execution subject and time nodes is generated. Operation and maintenance execution and effect feedback steps: The generated operation and maintenance plan is distributed to the execution component of the cloud computing platform to automatically execute operation and maintenance operations or prompt operation and maintenance personnel to execute them manually. The execution progress and effect of operation and maintenance operations are monitored in real time, platform operation data after operation and maintenance are collected, the changes in indicators before and after operation and maintenance are compared and analyzed, the operation and maintenance effect is evaluated to determine whether the operation and maintenance effect has achieved the expected goal, and the operation and maintenance process, results and effect evaluation data are stored in the knowledge base to provide reference for subsequent operation and maintenance decisions. Operations and maintenance knowledge accumulation and optimization steps: Regularly summarize and organize operations and maintenance cases, anomaly handling solutions and effect evaluation data in the knowledge base, extract common problems and optimal solutions to form operations and maintenance knowledge rules, update them to the operations and maintenance knowledge graph, optimize monitoring indicator thresholds and anomaly detection algorithm parameters based on the changing trends of operations and maintenance data, and continuously improve the adaptability and intelligence of operations and maintenance methods.

[0025] This invention also includes an anomaly risk level assessment step, which calculates the risk level of anomaly events by constructing an anomaly risk assessment formula, thereby determining the priority of operation and maintenance response. The formula is as follows: in This represents the abnormal risk level, with a value ranging from 0 to 10. The higher the value, the higher the risk. The weighting coefficient for abnormal impact is 0.4. This is the propagation speed weighting coefficient, with a value of 0.3; This is the duration weighting coefficient, with a value of 0.3; The number of entities affected by the anomaly; For the first The importance coefficient of each affected entity ranges from 1 to 5, with 5 for core business entities and 1 for non-core entities. For the first The relevance weight of each affected entity ranges from 0.1 to 1.0, with a higher weight for more related entities. This represents the number of time segments in which the anomaly propagated. For the first The rate of anomalous propagation in a time segment, expressed in units per minute, which is the number of newly affected entities per minute; For the first The duration of each time segment, in minutes; This represents the duration of the abnormality, in minutes, with a value range of 0-100. Values ​​exceeding 100 are counted as 100.

[0026] This invention also includes a dynamic allocation step for operation and maintenance resources. Based on the real-time resource load of the cloud computing platform and the requirements of operation and maintenance tasks, operation and maintenance resources such as CPU, memory, disk and network bandwidth are reasonably allocated to solve the problem of low operation and maintenance efficiency caused by uneven resource allocation. During the resource allocation process, priority is given to the operation and maintenance tasks corresponding to core businesses, and 20% of redundant resources are reserved to cope with sudden abnormal events. After the resource allocation is completed, the resource usage is monitored in real time. When the resource utilization rate is lower than 30%, the resource reclamation mechanism is activated to put the idle resources back into the resource pool for other operation and maintenance tasks to call.

[0027] In this invention, during the knowledge graph construction and dynamic update steps, an entity weight calculation mechanism is used to optimize the reasoning efficiency of the knowledge graph. An entity importance assessment formula is constructed to quantify the priority of entities in operational decisions, as follows: in This is the entity importance assessment value, ranging from 0 to 10; This represents the business dependency of an entity, i.e., the number of businesses that depend on this entity to run, with a value ranging from 1 to 50. This represents the number of historical failures of the entity, with a value ranging from 0 to 100. This is the resource occupancy coefficient of the entity, which is the ratio of the resources occupied by the entity to the total resources of the platform, and its value ranges from 0.01 to 1.0. This represents the deployment level of the entity, with a value ranging from 1 to 5. The platform's bottom-level basic entities take the value 1, and the top-level application entities take the value 5.

[0028] In this invention, the anomaly detection and localization step integrates the isolated forest algorithm and knowledge graph association reasoning for anomaly detection. The isolated forest algorithm clusters operation and maintenance data by constructing multiple isolated trees to identify abnormal data points that deviate from the normal data cluster. The knowledge graph association reasoning locates the root cause of the anomaly by traversing the entity association paths corresponding to the abnormal data points. When the isolated forest algorithm detects that the anomaly score of the abnormal data is higher than the preset threshold of 0.8, it triggers deep reasoning of the knowledge graph to extract the first-level and second-level associated entities of the abnormal node, generate an anomaly propagation path diagram, and intuitively present the range of anomaly spread.

[0029] In this invention, an operation and maintenance solution adaptability evaluation mechanism is established in the intelligent generation step of operation and maintenance decision-making. The evaluation is carried out from three dimensions: technology adaptability, resource adaptability, and time adaptability. Technology adaptability evaluates the degree of matching between the operation and maintenance solution and the cloud computing platform architecture and version. Resource adaptability evaluates the degree of compatibility between the resources required for the solution execution and the existing available resources on the platform. Time adaptability evaluates the matching between the execution time of the solution and the allowable duration of anomalies. The evaluation score for each dimension ranges from 0 to 10 points. Solutions with a comprehensive evaluation score higher than 8 points are listed as the optimal candidate solutions, and solutions with a comprehensive evaluation score lower than 5 points are directly eliminated.

[0030] In this invention, during the knowledge accumulation and optimization process of operations and maintenance (O&M), knowledge distillation technology is used to extract core O&M knowledge. Complex O&M cases and decision data are distilled into concise knowledge rules. During the knowledge distillation process, key O&M parameters and operation steps are retained, redundant information is removed, a knowledge accuracy verification mechanism is constructed, and O&M experts are invited to review the extracted knowledge rules. Knowledge rules that pass the review are updated to the knowledge graph, while knowledge rules that fail the review are returned to the knowledge base for re-extraction. The application frequency and success rate of knowledge rules are regularly statistically analyzed, and knowledge rules with an application frequency of less than 5 times or a success rate of less than 60% are optimized and adjusted.

[0031] This invention includes the following modules: Data acquisition and preprocessing module; configured with multiple types of data acquisition sensors and log acquisition components to collect hardware operation data, software log data, service call data and network transmission data of the cloud computing platform; built-in data cleaning and standardization processing algorithms to remove invalid data and convert it into a structured dataset in a unified format, providing data support for subsequent operation and maintenance analysis; Knowledge graph construction and update module; built-in entity definition tool and relationship sorting component, defines entity types and relationships in the field of cloud computing platform operation and maintenance, uses graph database to store and build initial operation and maintenance knowledge graph, sets up automatic update trigger component, and dynamically updates entity attributes, relationships and weight information according to the amount of new data and platform configuration changes; Real-time monitoring module: Build a monitoring model in four dimensions: hardware, software, service and network. Preset normal thresholds and warning thresholds for each monitoring indicator. Configure real-time data acquisition components to collect monitoring indicator data once per second. Compare real-time data with historical normal data, capture data fluctuation trends and provide real-time feedback on monitoring results. Anomaly detection and localization module; integrates isolated forest algorithm components and knowledge graph reasoning engine, detects abnormal data points through algorithms, triggers anomaly alarms when monitoring indicators exceed warning thresholds, uses knowledge graph to traverse related entities and relationships, locates core abnormal nodes, traces propagation paths and forms anomaly event archive; Operation and maintenance decision generation module; connects the operation and maintenance knowledge base and real-time resource monitoring components, matches similar anomaly handling solutions, evaluates the feasibility of solutions from four dimensions: processing time, resource consumption, impact scope and success rate, selects the optimal solution and generates a detailed operation and maintenance plan including operation steps, execution subject and time nodes; Operation and maintenance execution feedback module; equipped with operation and maintenance operation execution components, supporting both automatic and manual execution modes, real-time monitoring of the execution progress of the operation and maintenance plan, collection of platform operation data after operation and maintenance, comparative analysis of changes in indicators before and after operation and maintenance, evaluation of operation and maintenance effectiveness, and storage of relevant data in the knowledge base; Knowledge accumulation and optimization module; built-in knowledge summarization and organization tools and algorithm parameter optimization components, regularly extract operation and maintenance knowledge rules and update them to the knowledge graph, optimize monitoring indicator thresholds and anomaly detection algorithm parameters based on operation and maintenance data change trends, and integrate expert review interface to ensure the accuracy and applicability of knowledge rules; The data storage module adopts a distributed storage architecture to store structured operation and maintenance datasets, operation and maintenance knowledge graph data, abnormal event archives, operation and maintenance cases, and real-time monitoring data. It is configured with a data backup component to perform regular data backups, supports fast data query and retrieval, and ensures the security and reliability of data storage.

[0032] This invention also includes a resource load balancing module, which monitors the CPU utilization, memory usage, disk read / write speed, and network bandwidth utilization of the cloud computing platform in real time. When the utilization of a single resource exceeds 80%, the load balancing mechanism is activated to migrate the overloaded tasks to nodes with resource utilization below 40%. During the migration process, the business services are maintained to run continuously, load migration logs are recorded, and the reasons for load imbalance are analyzed to provide a basis for subsequent resource configuration optimization. This module supports automatically adjusting the resource allocation strategy according to changes in resource demand during peak and off-peak periods.

[0033] This invention also includes a security protection module with a built-in data encryption component to encrypt the collected operation and maintenance data and the stored knowledge base data. It adopts an access control mechanism to divide users into three levels: administrators, operation and maintenance personnel, and viewers. Different levels of users have different data access and operation permissions. Administrators have full permissions, operation and maintenance personnel can only perform operation and maintenance operations and view related data, and viewers can only browse operation and maintenance reports. This module monitors unauthorized access behavior in real time. When abnormal login or unauthorized operation is detected, it immediately triggers a security alarm and records detailed logs to ensure the security of the system and data.

[0034] The following two examples further illustrate the specific implementation of this system: Example 1: Intelligent Operation and Maintenance of Private Cloud Platform for Large Group Enterprises This embodiment applies to the private cloud platform of a large group enterprise. The platform adopts a hybrid IaaS and PaaS architecture, supporting six core business areas: finance, human resources, production, and sales. It deploys over 2000 entities, including servers, virtual machines, and containers, generating over 500GB of operational data daily. Traditional operations and maintenance rely on manual troubleshooting, with an average fault location time exceeding 60 minutes. Resource utilization is uneven, with resource shortages during peak core business periods coexisting with idle resources in non-core business areas. This invention utilizes a knowledge graph-driven intelligent operations and maintenance method and system to automate and intelligently manage the entire operations and maintenance process, improving platform stability and resource utilization efficiency.

[0035] I. System Deployment The system deployment covers all nodes of the group's private cloud platform. The data acquisition and preprocessing module is configured with hardware data acquisition sensors and log acquisition components. The hardware sensors are deployed at the CPU, memory, disk, and network interfaces of each server. The log acquisition components are connected to the output ports of application logs, process logs, and system event logs. The knowledge graph construction and update module uses a distributed graph database to store entities and relationships, and is configured with entity definition tools and relationship analysis components, with pre-defined basic attribute fields for five types of entities: servers, virtual machines, containers, application services, and network devices. The real-time monitoring module builds a four-dimensional monitoring model of hardware, software, services, and network. The anomaly detection and localization module integrates an isolated forest algorithm component and a knowledge graph inference engine. The operation and maintenance decision generation module connects to the group's historical operation and maintenance knowledge base and has a built-in solution evaluation component. The operation and maintenance execution feedback module is equipped with an automatic execution plugin and a manual operation terminal. The knowledge accumulation and optimization module integrates a knowledge distillation tool and an expert review interface. The data storage module adopts a distributed storage architecture, divided into five major data storage partitions, and configured with a scheduled backup component. A resource load balancing module and a security protection module are also deployed. The resource load balancing module is linked to the platform resource scheduling center, and the security protection module is connected to the group's unified identity authentication system.

[0036] II. Implementation Steps Data acquisition and preprocessing steps: Collect platform hardware operation data such as CPU utilization, memory usage, disk read / write speed, and network bandwidth utilization; software log data such as application error logs, process execution logs, and system event logs; service call data such as microservice interface call frequency, response time, and call success rate; and network transmission data such as packet loss rate, latency, and number of connections. Missing values ​​are removed using the mean imputation method, outliers are identified and removed based on the 3σ principle, duplicate data is removed using hash deduplication, and data of different dimensions is converted to the range of 0 to 1 using a standardization formula to generate a structured operation and maintenance dataset.

[0037] Knowledge graph construction and dynamic update steps: Define server entities including IP address, hardware model, deployment location, etc.; virtual machine entities including image version, allocated resources, server, etc.; container entities including image name, running status, dependent services, etc.; application service entities including service name, interface address, call relationship, etc.; and network device entities including device type, IP network segment, connection port, etc. Identify and analyze the dependencies, call relationships, deployment relationships, and topology relationships between entities, and construct an initial operation and maintenance knowledge graph using a graph database. Set a knowledge update trigger threshold of 10GB of new data or a platform configuration change. Upon triggering, automatically update entity attributes and relationships, and synchronously update entity weights and relationship strengths. Quantify entity priority using an entity importance assessment formula, selecting the group's core financial system server for calculation. =45 represents the number of services that rely on this server. =3 represents the number of historical failures. =0.08 is the resource utilization coefficient. =2 represents the deployment level; substituting this into the formula... Calculated =45×(1+3) / (0.08×(1+2))=45×4 / 0.24=750, after normalization =10, indicating that the server is the highest priority entity.

[0038] Multi-dimensional real-time monitoring steps: A monitoring model is built based on an operations and maintenance knowledge graph. Hardware metrics include CPU utilization and memory usage; software metrics include log error rate and process survival rate; service metrics include call success rate and response time; and network metrics include packet loss rate and latency. Normal and warning threshold ranges are set for each metric. Real-time data acquisition components collect monitoring metric data every second. The real-time data is compared and analyzed with historical normal data in the knowledge graph to capture data fluctuation trends and provide real-time feedback.

[0039] Anomaly Detection and Localization Steps: Anomaly detection is performed by integrating the Isolation Forest algorithm and knowledge graph-based association reasoning. The Isolation Forest algorithm constructs 50 isolated trees to cluster operational data and identify anomalous data points. When an application server's CPU utilization score of 0.85 is detected, exceeding the preset threshold of 0.8, deep reasoning based on the knowledge graph is triggered. The system traverses the virtual machines, containers, and application service entities associated with this server, locating the core node of the anomaly as a financial accounting application process consuming excessive resources. The anomaly propagation path is traced back to this process consuming CPU resources, causing slow server response and consequently affecting the associated report generation service. The scope of the anomaly's impact is determined to cover three core business modules of the finance department, and anomaly feature information is extracted to form an anomaly event archive.

[0040] Anomaly Risk Level Assessment Steps: Assess the risk level of the anomaly event by substituting the values ​​into the anomaly risk assessment formula. .in =0.4, =0.3, =0.3, =8 represents the number of affected entities. The core business entity takes a value of 5, ranging from 3 to 5. Values ​​range from 0.6 to 1.0. =32; =3 represents the number of time segments. The units per minute are 2, 3, and 4 respectively. All are 5 units per minute. =(10+15+20) / 3=15; =12 means the anomaly has lasted for the duration in minutes. (Calculated...) =0.4×32+0.3×15+0.3×12=12.8+4.5+3.6=20.9, after normalization =10, determined as the highest priority, and the operation and maintenance response is initiated immediately.

[0041] The intelligent generation steps for operation and maintenance decisions are as follows: Based on the abnormal event archive, similar cases are matched with the knowledge base to obtain three candidate solutions. Combining real-time resource load conditions, the solutions are evaluated from four dimensions: processing time, resource consumption, impact scope, and success rate. Solution 1, with a comprehensive evaluation score of 9.2, is selected as the optimal solution. A detailed operation and maintenance plan is generated, specifying the operation steps as terminating the abnormal process, restarting the application service, and monitoring resource usage. The execution entity is the Linux group of the group's operation and maintenance department, and the timeframe is to complete the task within 10 minutes.

[0042] Operation and maintenance execution and effect feedback steps: The operation and maintenance plan is distributed to the execution component, which automatically terminates abnormal processes and restarts application services. Execution progress is monitored in real time, and data such as CPU utilization and response time are collected after the operation and maintenance. Comparative analysis shows that CPU utilization decreased from 95% to 35%, and response time decreased from 8 seconds to 1 second, achieving the expected operation and maintenance goals. The operation and maintenance process, results, and effect evaluation data are stored in a knowledge base.

[0043] Dynamic allocation steps for operation and maintenance resources: Based on the platform's real-time resource load, CPU and memory resources are allocated to prioritize financial accounting applications, with 20% of resources reserved for redundancy. If the resource utilization rate of a non-core business server is detected to be 25%, the resource reclamation mechanism is activated, adding the idle 2GB of memory and 1 CPU core to the resource pool for use by other operation and maintenance tasks.

[0044] Operations and maintenance knowledge accumulation and optimization steps: Knowledge distillation technology is used to extract the core knowledge rules for anomaly handling, retaining key parameters such as anomaly characteristics, handling steps, and resource configuration, while removing redundant log information. Three senior operations and maintenance experts are invited to review the knowledge rules, and they are updated to the knowledge graph after approval. The application frequency of these rules is regularly statistically analyzed, and they are automatically matched and executed when similar anomalies occur subsequently.

[0045] III. Effect Verification Table 1 Comparison of core platform indicators before and after maintenance in Example 1 Evaluation indicators Before maintenance After maintenance Improvement status Mean time to locate fault 62 minutes 8 minutes Significantly shorten Average resource utilization rate 45% 78% Significant improvement Exception handling success rate 72% 96% Significantly improved Maintenance labor costs 8 people / day 2 people / day Significantly reduced Platform service availability 98.2% 99.9% Significant improvement Table 1 clearly demonstrates the significant effects of this invention on the operation and maintenance of large enterprise private cloud platforms. Before operation and maintenance, the platform relied on manual troubleshooting, with an average fault location time of up to 62 minutes, resource utilization of only 45%, an imbalance in resource allocation between core and non-core businesses, an anomaly handling success rate of 72%, and a labor cost of 8 person-days required, making it difficult to guarantee platform service availability. Through the intelligent operation and maintenance method and system of this invention, the associative reasoning capability of the knowledge graph significantly shortens the fault location time to 8 minutes, the dynamic resource allocation mechanism increases the average resource utilization rate to 78%, and the anomaly detection scheme based on fusion algorithms achieves an anomaly handling success rate of 96%. The automation of the entire operation and maintenance process reduces manual input, lowering labor costs to 2 person-days, and increasing platform service availability to 99.9%, fully meeting the group's core business requirements for platform stability and efficiency, significantly reducing operation and maintenance costs, and improving business operation efficiency.

[0046] Example 2: Intelligent Operation and Maintenance of Public Cloud Platforms of Internet Companies This embodiment applies to the public cloud platform of an internet company. This platform primarily operates on a SaaS model, providing various services including e-commerce, social networking, and office applications, serving over ten million users. Platform nodes are distributed across three data centers, handling over one billion service requests daily. Traditional operations and maintenance (O&M) face challenges such as frequent unexpected anomalies, slow response to user complaints, and low knowledge reuse rates. Service interruptions are frequent during peak hours due to resource congestion, and historical O&M cases have not been effectively transformed into reusable knowledge. This invention constructs a knowledge graph-driven intelligent O&M system to achieve rapid anomaly response, dynamic resource allocation, and continuous knowledge accumulation, thereby improving platform service quality and O&M efficiency.

[0047] I. System Deployment The system deployment covers all public cloud nodes across three regional data centers. The data acquisition and preprocessing module is configured with distributed data acquisition sensors and log acquisition components. The sensors collect hardware operation data from servers across regions in real time, while the log acquisition components synchronously capture log data from various application types. The knowledge graph construction and update module uses a cross-regional deployed graph database to ensure real-time synchronization of entity relationship data across different data center nodes. It is equipped with entity definition tools and relationship sorting components to adapt to entity attribute definitions for different service types such as e-commerce, social networking, and office applications. The real-time monitoring module's four-dimensional monitoring model is tailored to the characteristics of public clouds, enhancing the monitoring of user-perceived indicators such as service call success rate and network latency. The anomaly detection and localization module integrates an isolated forest algorithm component that optimizes algorithm parameters for massive request data. The operation and maintenance decision generation module connects to the company's historical operation and maintenance case knowledge base, and the operation and maintenance execution feedback module supports cross-regional remote operation and automatic execution. The knowledge accumulation and optimization module accesses the company's operation and maintenance expert database, and the data storage module adopts a multi-site active-active architecture to ensure data security and availability. The resource load balancing module works in conjunction with the platform's elastic computing components, and the security protection module divides user permissions into three levels to meet the security requirements of multi-tenant public clouds.

[0048] II. Implementation Steps Data collection and preprocessing steps: Collect platform hardware operation data, software log data, service call data, and network transmission data. For service call data, focus on collecting cross-regional call data between nodes in different regions. For network transmission data, add cross-regional bandwidth usage indicators. Data cleaning algorithms are used to remove missing values, outliers, and duplicate data caused by network fluctuations during the collection process. Standardization processing transforms data from different regions and of different types into standardized data within a unified scope, generating a structured operation and maintenance dataset that supports unified cross-regional data analysis.

[0049] Knowledge graph construction and dynamic update steps: Define entity types adapted to public cloud services, add tenant entities and region node entities. Tenant entities include attributes such as tenant ID, service type, and resource quota, while region node entities include attributes such as data center location, network bandwidth, and number of nodes. Analyze the relationships between entities, adding tenant-service relationships and region-node relationships. Build an initial operation and maintenance knowledge graph using a cross-region graph database, setting up a knowledge update trigger mechanism. When the amount of new data added to a region node reaches 8GB or the service configuration changes, the knowledge graph is automatically updated and synchronized to database nodes in other regions. Calculate the priority of application service entities for core e-commerce services using an entity importance assessment formula. =50 represents the number of tenants that depend on this entity. =2 represents the number of historical failures. =0.1 is the resource utilization coefficient. =3 represents the deployment level; substituting this into the formula... Calculated =50×(1+2) / (0.1×(1+3))=50×3 / 0.4=375, after normalization =10, mark this entity as a key monitoring target.

[0050] Multi-dimensional real-time monitoring steps: A monitoring model is built based on an operations and maintenance knowledge graph. Monitoring indicators across four dimensions—hardware, software, service, and network—are optimized for the multi-tenant characteristics of public clouds. A new tenant complaint rate indicator is added to the service dimension, and the cross-regional transmission latency indicator is strengthened in the network dimension. Normal and warning threshold ranges are set for each indicator. The real-time data acquisition component collects monitoring data every second, comparing real-time data with historical normal data in the knowledge graph to capture data fluctuation trends across regional nodes, with a focus on monitoring the operational status of core e-commerce services.

[0051] Anomaly Detection and Localization Steps: Anomaly detection is performed by integrating the Isolation Forest algorithm and knowledge graph-based association reasoning. The Isolation Forest algorithm constructs 60 isolated trees to cluster massive service call data. When an anomaly score of 0.83 is detected for the call success rate of an e-commerce service in a certain region, exceeding the preset threshold of 0.8, deep reasoning based on the knowledge graph is triggered. The system traverses the tenant entities, server entities, and regional node entities associated with this service, locating the core anomaly node as a switch failure in the regional data center. The anomaly propagation path is traced back to the switch failure causing timeouts in e-commerce service calls in that region, which in turn affected cross-regional service calls, clearly indicating that the anomaly impacted over 200 tenants. Anomaly feature information is extracted to form an anomaly event archive.

[0052] Anomaly Risk Level Assessment Steps: Calculate the risk level of the anomaly event using the anomaly risk assessment formula. =0.4, =0.3, =0.3, =200 represents the number of affected entities. Values ​​range from 3 to 5, with e-commerce tenants corresponding to a value of 5. Values ​​range from 0.7 to 1.0. =850; =5 represents the number of time segments. The units are 30, 45, 50, 40, and 35 units per minute, respectively. All are in units of 4 minutes. =(120+180+200+160+140) / 5=160; =15 indicates the duration of the anomaly, measured in minutes. The calculation yields... =0.4×850+0.3×160+0.3×15=340+48+4.5=392.5, after normalization =10, initiate the highest priority operation and maintenance response.

[0053] The intelligent generation steps for operation and maintenance decisions are as follows: First, similar cases of switch fault handling are matched with the knowledge base to obtain four candidate solutions. These solutions are evaluated from three dimensions: technical adaptability, resource adaptability, and time adaptability. Technical adaptability assesses the degree of matching between the solution and the switch model and platform architecture; resource adaptability assesses the availability of spare parts and operation and maintenance personnel; and time adaptability assesses the matching of the solution's execution time with the tenant's tolerable downtime. Solution 3 received a comprehensive evaluation score of 8.6 and was selected as the optimal solution. A detailed operation and maintenance plan was generated, specifying the operation steps as switching to a backup switch, restarting network services, and verifying cross-regional connectivity. The execution entity is the local data center operation and maintenance team, and the timeframe is to complete the task within 15 minutes.

[0054] Operation and Maintenance Execution and Feedback Steps: The operation and maintenance plan is distributed to the local data center execution component. Operation and maintenance personnel cooperate with the automated execution component to complete the backup switch switchover and network service restart. Execution progress is monitored in real time, and data such as service call success rate and cross-regional latency are collected after the operation and maintenance. Comparisons show that the service call success rate has rebounded from 65% to 99.8%, and the cross-regional transmission latency has decreased from 200ms to 30ms, achieving the expected operation and maintenance results. Relevant data is stored in a knowledge base to provide a reference for handling similar faults in the future.

[0055] Dynamic allocation steps for operation and maintenance resources: Based on the platform's real-time resource load, prioritize allocating sufficient network bandwidth and computing resources to e-commerce services, reserving 20% ​​of redundant resources to handle sudden requests. If the resource utilization rate of the office service node is detected to be 28%, initiate the resource reclamation mechanism, reclaiming 1 CPU core and 1GB of memory resources and including them in the resource pool for peak-hour service requests from the e-commerce provider.

[0056] Operations and maintenance knowledge accumulation and optimization steps: Knowledge distillation technology is used to extract knowledge rules for handling switch faults, retaining key information such as fault characteristics, handling steps, and backup device switching procedures. Operations and maintenance experts are invited to review the knowledge rules, and they are updated to the knowledge graph after approval. The application frequency and success rate of knowledge rules are regularly statistically analyzed, and rules with low application frequency are optimized and adjusted to continuously enrich the knowledge base content.

[0057] III. Effect Verification Table 2 Comparison of the effects of different operation and maintenance schemes in Example 2 Evaluation indicators Traditional operation and maintenance solutions This invention's operation and maintenance solution Advantages Abnormal response time 45 minutes 10 minutes Faster response time Tenant complaint rate 3.2% 0.5% Better service quality Knowledge reuse rate 25% 88% Better knowledge transformation Number of service interruptions during peak hours 5 times a week Once a month Significantly improved stability Cross-regional operation and maintenance collaboration efficiency Low high Stronger collaborative capabilities Table 2 clearly demonstrates the significant advantages of this invention compared to traditional O&M solutions. Traditional O&M solutions, facing a massive number of public cloud users and geographically dispersed nodes, suffer from anomaly response times as long as 45 minutes, a tenant complaint rate of 3.2%, a historical O&M knowledge reuse rate of only 25%, frequent service interruptions during peak hours, and low cross-regional O&M collaboration efficiency, severely impacting user experience. This invention's intelligent O&M solution, leveraging the associative reasoning and dynamic updating capabilities of knowledge graphs, reduces anomaly response time to 10 minutes and lowers the tenant complaint rate to 0.5%. Knowledge accumulation and optimization mechanisms increase knowledge reuse rate to 88%, rapidly transforming historical cases into reusable O&M rules. Resource load balancing and cross-regional collaborative O&M reduce the number of service interruptions during peak hours to once per month, significantly improving cross-regional O&M collaboration efficiency. This solution is fully adapted to the characteristics of public cloud platforms—large user base, diverse service types, and wide geographical distribution—effectively improving the intelligence level and service quality of platform O&M and enhancing the platform's core competitiveness.

[0058] Reference Figure 3 This diagram visually quantifies the differences in fault location efficiency across different operation and maintenance (O&M) modes, serving as a core visualization dimension for assessing O&M response capabilities. Traditional manual O&M relies on the experience of O&M personnel for troubleshooting, lacking data correlation analysis support, resulting in a location time of up to 62 minutes. Semi-automated O&M only introduces basic algorithms without integrating business relationships, reducing the time to 35 minutes. The industry average of 40 minutes reflects the overall technical level in the cloud computing O&M field. While competing solutions optimize detection algorithms, they do not incorporate knowledge graphs, resulting in a location time of 28 minutes. This invention, relying on the entity association reasoning capabilities of knowledge graphs, can directly traverse the upstream and downstream relationships of abnormal nodes, accurately pinpointing the core of the fault and compressing the location time to 8 minutes, a reduction of over 80%. This directly verifies the core enabling value of knowledge graphs in the anomaly tracing process, laying a crucial foundation for accelerating O&M response.

[0059] Reference Figure 4 This graph clearly shows the monthly increase in resource utilization as the invention is implemented, serving as a core dimension for verifying the effectiveness of the resource scheduling strategy. The initial utilization rate of 45% in January reflects the imbalanced resource allocation under the traditional operation and maintenance model, with insufficient core business resources and idle non-core business resources coexisting. The utilization rate rapidly increased to 69% in February and March, stemming from the gradual implementation of the dynamic resource allocation mechanism and load balancing module. The strategies of reclaiming idle resources and prioritizing core business allocation effectively revitalized existing resources. The growth rate slowed in April and May but steadily increased to 78%, approaching the optimal resource utilization range for cloud computing platforms (75%-85%). The trend shows no significant fluctuations, indicating that the resource scheduling system of this invention has good adaptability. The redundant resource reservation mechanism not only ensures resource demand in case of sudden anomalies but also avoids resource waste. The continuously optimized allocation strategy gradually brings resource utilization efficiency closer to the optimal state, significantly reducing platform resource costs.

[0060] Reference Figure 5 This chart focuses on the core public cloud experience metric—tenant complaint rate—and visually demonstrates the optimization effect of this invention on service quality. Traditional operation and maintenance solutions have a complaint rate of 3.2%, mainly due to slow response times to anomalies and cross-regional service interruptions. The industry average of 2.5% reflects common pain points in public cloud operation and maintenance, with most solutions only alleviating problems through capacity expansion without addressing the root cause. Competitor optimization solutions, at 1.8%, while improving service invocation algorithms, lack cross-regional collaborative operation and maintenance capabilities. During the pilot phase of this invention, the complaint rate dropped to 1.2%, and further decreased to 0.5% after full implementation. The core reason is cross-regional knowledge graph synchronization, rapid anomaly response (within 10 minutes), and dynamic resource allocation, which significantly reduces service interruption time and response latency, resulting in a significant improvement in tenants' perception of service stability. The low complaint rate not only reflects the practicality of the technical solution but is also a core competitive advantage for the commercialization of public cloud services, validating the technical advantages of this invention in adapting to multi-tenant and cross-regional scenarios.

[0061] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based intelligent operation and maintenance method for cloud computing platforms, characterized in that, Includes the following steps: The data collection and preprocessing steps involve collecting four types of data from the platform: hardware, software, services, and network. Invalid data is removed using a cleaning algorithm, and then the data is standardized to convert the data dimensions and generate a structured operation and maintenance dataset. The knowledge graph construction and dynamic update steps are as follows: define entity types and relationships in the operation and maintenance domain based on standardized datasets, construct an initial knowledge graph using a graph database, and automatically update graph attributes, relationships, and weights according to a preset mechanism when data meets the standards or configuration changes. The multi-dimensional real-time monitoring process relies on a knowledge graph to build a four-dimensional monitoring model, sets indicator thresholds, collects data every second through real-time components, and analyzes fluctuation trends by comparing with historical data. The anomaly intelligent detection and localization steps integrate correlation reasoning and machine learning algorithms to detect anomalies. When the indicators exceed the threshold, an alarm is triggered. The graph is traversed to locate core nodes, trace the propagation path, and form an anomaly event archive. The intelligent generation steps for operation and maintenance decisions are as follows: matching historical cases to obtain candidate solutions, combining resource load assessment for feasibility, selecting the optimal solution and generating a detailed operation and maintenance plan; Operation and maintenance execution and effect feedback steps: distribute the plan to the execution components, monitor the execution progress and effect, collect post-operation and maintenance data to evaluate the effectiveness, and store relevant data in the knowledge base; The steps for accumulating and optimizing operational knowledge include: organizing the knowledge base content, refining operational knowledge rules, updating them to the graph, and optimizing monitoring thresholds and algorithm parameters based on data trends.

2. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, It also includes an anomaly risk level assessment step, which calculates the risk level of an anomaly event by constructing an anomaly risk assessment formula, as follows: in The risk level is abnormal. The weighting coefficient for abnormal effects; This is the propagation speed weighting coefficient; The duration weighting coefficient; The number of entities affected by the anomaly; For the first Importance coefficient of each affected entity; For the first The correlation weight of each affected entity; This represents the number of time segments in which the anomaly propagated. For the first The abnormal propagation rate of each time segment; For the first The duration of each time segment; The duration of the abnormality.

3. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, It also includes a dynamic allocation step for operation and maintenance resources. Based on the real-time resource load of the cloud computing platform and the operation and maintenance task requirements, operation and maintenance resources are allocated reasonably. During the resource allocation process, priority is given to the operation and maintenance tasks corresponding to core businesses. After the resource allocation is completed, the resource usage is monitored in real time.

4. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, In the knowledge graph construction and dynamic update steps, an entity weight calculation mechanism is adopted. By constructing an entity importance assessment formula, the priority of entities in operation and maintenance decisions is quantified. The formula is as follows: in This is the entity importance assessment value; Business dependence of the entity; This refers to the number of historical failures of the entity. This represents the resource occupancy coefficient of the entity. The deployment level for entities.

5. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, In the anomaly detection and localization step, the isolated forest algorithm and knowledge graph association reasoning are integrated for anomaly detection. The isolated forest algorithm clusters the operation and maintenance data by constructing multiple isolated trees to identify abnormal data points that deviate from the normal data cluster. The knowledge graph association reasoning locates the root cause of the anomaly by traversing the entity association paths corresponding to the abnormal data points.

6. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, In the intelligent generation step of operation and maintenance decision-making, an operation and maintenance solution adaptability assessment mechanism is established, which evaluates the solution from three dimensions: technology adaptability, resource adaptability, and time adaptability. Technology adaptability assesses the degree of matching between the operation and maintenance solution and the cloud computing platform architecture and version. Resource adaptability assesses the degree of compatibility between the resources required for the solution execution and the existing available resources on the platform. Time adaptability assesses the matching between the execution time of the solution and the allowable duration of anomalies.

7. The intelligent operation and maintenance method for a cloud computing platform based on knowledge graphs according to claim 1, characterized in that, In the process of knowledge accumulation and optimization in operations and maintenance, knowledge distillation technology is used to extract core operations and maintenance knowledge. Complex operations and maintenance cases and decision data are distilled into concise knowledge rules. During the knowledge distillation process, key operations and maintenance parameters and operation steps are retained, while redundant information is removed. Operations and maintenance experts are invited to review the extracted knowledge rules. Knowledge rules that pass the review are updated to the knowledge graph, while knowledge rules that fail the review are returned to the knowledge base for re-extraction.

8. A cloud computing platform intelligent operation and maintenance system based on knowledge graphs as described in any one of claims 1-7, characterized in that, Includes the following modules: The data acquisition and preprocessing module is configured with multiple types of acquisition sensors and log components to collect platform hardware, software, service, and network data. It cleans invalid data and standardizes the format through built-in algorithms to generate structured datasets. The knowledge graph construction and update module uses entity definition tools and relationship sorting components to define entity types and relationships in the operation and maintenance domain, uses a graph database to build an initial knowledge graph, and uses an automatic update trigger component to dynamically update the graph information based on data volume and configuration changes. The real-time monitoring module builds a four-dimensional monitoring model and presets indicator thresholds. It is configured to collect data every second using real-time acquisition components and compares real-time and historical data to capture fluctuation trends. The anomaly detection and localization module integrates the isolated forest algorithm component and the knowledge graph reasoning engine to detect abnormal data and trigger alarms. It uses graph traversal to traverse related information, locates abnormal nodes, traces the propagation path, and forms an event archive. The operation and maintenance decision generation module connects the knowledge base and resource monitoring components, matches similar processing solutions, evaluates feasibility from four dimensions, selects the optimal solution, and generates a detailed operation and maintenance plan. The operation and maintenance execution feedback module is equipped with execution components to support automatic and manual execution, monitor execution progress, collect post-operation and maintenance data to evaluate the effect, and store relevant data in the knowledge base; The knowledge accumulation and optimization module extracts operation and maintenance knowledge rules through built-in tools and updates them to the graph. It optimizes thresholds and algorithm parameters based on data trends and integrates an expert review interface. The data storage module uses a distributed storage architecture to store various operational and maintenance data, is configured with a backup component for regular backups, and supports fast data retrieval and retrieval.

9. The intelligent operation and maintenance system for a cloud computing platform based on knowledge graphs according to claim 8, characterized in that, It also includes a resource load balancing module, which monitors the CPU utilization, memory usage, disk read / write speed and network bandwidth utilization of the cloud computing platform in real time. When the utilization of a single resource exceeds 80%, the load balancing mechanism is activated to migrate the overloaded tasks to nodes with resource utilization below 40% and records the load migration log.

10. The intelligent operation and maintenance system for a cloud computing platform based on knowledge graphs according to claim 8, characterized in that, It also includes a security protection module, with a built-in data encryption component to encrypt the collected operation and maintenance data and the stored knowledge base data. It adopts an access control mechanism to divide users into three levels: administrators, operation and maintenance personnel, and viewers. Different levels of users have different data access and operation permissions.

Images