A method for generating repair tasks for server cluster abnormal scenarios based on big data AI

Abstract

The present invention discloses a method for generating server cluster abnormal scenario repair tasks based on big data AI, which relates to the field of machine learning technology, including: obtaining historical operation logs and real-time operation data of the server cluster, analyzing the multimodal abnormal fusion of the interaction relationship between the hardware layer and software layer of the server cluster, and generating target server cluster component operation abnormal scenario behavior data; performing causal correlation analysis based on the target server cluster component operation abnormal scenario behavior data and known server cluster abnormal scenario behavior performance big data, and evaluating the current abnormal deviation state of the target server cluster component; based on the current abnormal deviation state of the target server cluster component, marking the abnormal component node and the corresponding associated task sequence, establishing a non-interruption task migration resource allocation model, and generating the server cluster abnormal scenario repair task. The advantages of the present invention are: improving the accuracy of anomaly detection, fault recovery time and the repair capability of the server cluster.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a method for generating repair tasks for abnormal scenarios in server clusters based on big data AI. Background Art

[0002] Existing big data AI-based server cluster anomaly repair solutions use algorithms like Isolation Forest and K-Means for anomaly detection, combined with simple threshold alerts and predefined scripts to execute repair tasks. This leads to problems such as hardware-software data fragmentation, a lack of causal reasoning, and rigid repair strategies during the repair process of actual server cluster anomaly scenarios. This results in high false alarm rates, long mean time to repair, and difficulty handling cross-component cascading failures and dynamic resource contention scenarios, which can easily lead to secondary anomalies or service interruptions during the repair process. Summary of the Invention

[0003] To address the above technical issues, a method for generating repair tasks for server cluster anomaly scenarios based on big data AI is provided. This technical solution addresses the aforementioned existing big data AI-based server cluster anomaly scenario repair solutions, which implement anomaly detection through algorithms such as isolation forests and K-Means, combined with simple threshold alarms and predefined scripts to execute repair tasks. This leads to problems such as hardware-software data fragmentation, lack of causal reasoning, and rigid repair strategies during the repair of actual server cluster anomaly scenarios. These problems result in a high false alarm rate, a long average repair time, and difficulty handling cross-component cascading failures and dynamic resource competition scenarios, easily causing secondary anomalies or service interruptions during the repair process.

[0004] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0005] The method for generating repair tasks for server cluster abnormal scenarios based on big data AI includes:

[0006] S1. Obtain the historical operation logs and real-time operation data of the server cluster, analyze the multi-modal anomaly fusion of the interaction between the hardware layer and the software layer of the server cluster, and generate the target server cluster component operation abnormal scenario behavior data;

[0007] S2. Perform causal correlation analysis based on the target server cluster component's abnormal scenario behavior data and known server cluster abnormal scenario behavior performance big data to assess the current abnormal deviation status of the target server cluster component;

[0008] S3. Based on the current abnormal deviation status of the target server cluster component, mark the abnormal component nodes and the corresponding associated task sequences, establish a non-interruption task migration resource allocation model, and generate server cluster abnormal scenario repair tasks.

[0009] Preferably, original performance parameters of each component of the server cluster are collected according to unit timestamps, and the original performance parameters of each component are pre-processed using sliding window normalization to form a hardware performance normalization vector of the server cluster;

[0010] Collect historical operation logs of each component of the server cluster according to unit timestamps, train the BERT model, extract the semantic set of the server cluster logs, substitute it into the Softmax function, and assign semantic weights to the server cluster logs;

[0011] Using the attention weighted method, the known server cluster component operation abnormality scenarios are used as the query matrix, the log semantic set is used as the query key, and the hardware performance normalized vector of the server cluster is used as the query value to calculate the log feature vector of the server cluster.

[0012] Preferably, the hardware performance normalized vector of the server cluster is used as the hardware node, and the log feature vector of the server cluster is used as the software node. The Pearson correlation coefficient is used to calculate the correlation coefficient between the hardware node and the software node to generate the hardware node-software node connection edge. Sigmoid is used to assign weights to the hardware node-software node fully connected edge, and a hardware node-software node fully connected edge network of the server cluster is established.

[0013] Extract each adjacent node in the fully connected edge network of the server cluster's hardware nodes and software nodes, establish the server cluster's adjacent node array, and form the server cluster's hardware node-software adjacent node matrix;

[0014] The hardware node-software adjacent node matrix of the server cluster is used to train the GCN graph convolutional network. The fully connected layer is used to calculate the abnormal probability of each hardware node-software adjacent node, and the abnormal scenario behavior data of the target server cluster components are compiled.

[0015] Preferably, the DTW dynamic event warping algorithm is used to recursively calculate the maximum similarity between the big data of the known server cluster abnormal scenario behavior performance and the target server cluster component operation abnormal scenario behavior data, screen out the characteristic parameters of the known server cluster abnormal scenario behavior performance, and establish the known target server cluster component abnormal scenario operation performance characteristic time series data;

[0016] Based on the time series data of the performance characteristics of the known target server cluster components in abnormal scenarios, the data is substituted into the fully connected edge network of the hardware nodes and software nodes of the server cluster. According to the computing power call path of the hardware nodes and software nodes, a causal graph of the operation of the known target server cluster components in abnormal scenarios is established;

[0017] Using the independent conditional testing method, the correlation strength of each edge of the causal graph of the known target server cluster component abnormal scenario operation is verified, the causal chain between each node is determined, and the causal structure graph of the known target server cluster component abnormal scenario operation is obtained;

[0018] Based on the MLP multi-layer perceptron, the expected impact of the corresponding abnormal scenario operation performance characteristics under the numerical values ​​of the hardware nodes and software nodes in the causal structure diagram of the abnormal scenario operation of the known target server cluster component is verified, and the hardware-software causal intervention representation associated with the abnormal scenario of the known target server cluster component is obtained.

[0019] Preferably, determining standardized operating parameters of hardware nodes and software nodes of known target server cluster components, and forming a standardized parameter array of known target server cluster components;

[0020] Normalize the hardware-software causal intervention representation associated with the known target server cluster component abnormal scenario, and assign weights to the time series data of the operational performance characteristics of the known target server cluster component abnormal scenario;

[0021] Using the Euclidean distance formula, with the weight of the known target server cluster component's abnormal scenario performance characteristics as the external intervention factor, the spatial distance between the known target server cluster component's abnormal scenario performance characteristics time series data and the known target server cluster group component's standardized parameter array is calculated to determine the target server cluster component's current abnormal deviation status indicator;

[0022]

[0023] in, is the current abnormal deviation status indicator of the i-th component of the target server cluster, The performance weight of the jth abnormal scenario operation of the i-th component of the known target server cluster, The performance characteristic value of the jth abnormal scenario of the i-th component of the known target server cluster, is the normalized parameter of the i-th component of the target server cluster, and n is the total number of components in the target server cluster.

[0024] Preferably, based on the target server cluster component's currently to-be-executed associated task sequence, the target server cluster component is substituted into the fully connected edge network of the hardware nodes and software nodes of the server cluster to determine the component load resource demand characteristic parameters of the to-be-executed associated task sequence, and to construct the target server cluster component's currently to-be-executed associated task demand characteristic vector;

[0025] The fully connected edge network of hardware nodes and software nodes of the server cluster is used to verify the resource conflict intensity between the characteristic vectors of the current pending associated task requirements of the target server cluster components per unit time, and to determine the priority of the current pending associated task requirements of the target server cluster components.

[0026] Preferably, based on Logistic regression, the current abnormal deviation state indicator of the target server cluster component is used as a given condition, the component load demand characteristic data per unit time of the associated task sequence to be executed is used as the output, and the shutdown risk probability of the target server cluster component executing the associated task is used as the label data output;

[0027] According to the construction parameters of the target server cluster components, the initialization performance parameters of the target server cluster components are determined, and a digital twin model of the target server cluster components is established.

[0028] The target server cluster component digital twin model is updated using the shutdown risk probability of the server cluster component executing the associated tasks, and the load performance support vector of the target server cluster component is determined;

[0029] The load performance support vector of the target server cluster component is used to train the SVR support vector machine. The priority of the current associated tasks to be executed by the target server cluster component is used as the constraint condition, and the maximization of the completion of the associated tasks to be executed is used as the objective function. The hyperplane boundary of the target server cluster component's load performance execution sequence of tasks to be executed is generated, and the feature vector of the current associated tasks to be executed by the target server cluster component is used as the output, and the server cluster abnormal scenario repair task is used as the output.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] This paper proposes a big data AI-based solution for generating repair tasks for server cluster anomaly scenarios. This solution achieves precise anomaly detection through multimodal fusion analysis of hardware-software interaction relationships, quantifies abnormal deviation states through dynamic causal reasoning and intervention effect analysis, and generates lossless migration solutions using digital twin simulation and SVR optimization, forming a complete closed loop of "anomaly perception-root cause analysis-intelligent repair". Compared with traditional methods, this solution improves anomaly detection accuracy, fault recovery time, and server cluster repair capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Generate a flowchart for the method of repairing server cluster abnormal scenarios based on big data AI; DETAILED DESCRIPTION

[0033] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0034] Reference Figure 1 As shown in the figure, the method for generating repair tasks for server cluster abnormal scenarios based on big data AI includes:

[0035] Step 1: Obtain the historical operation logs and real-time operation data of the server cluster, analyze the multi-modal anomaly fusion of the interaction between the hardware layer and the software layer of the server cluster, and generate the target server cluster component operation abnormal scenario behavior data;

[0036] The step 1 includes the following:

[0037] Collect the raw performance parameters of each component of the server cluster according to the unit timestamp, pre-process them using sliding window normalization, and construct a standardized vector of the server cluster's hardware performance.

[0038] Collect historical operation logs of each component of the server cluster according to unit timestamps, train the BERT model, extract the semantic set of the server cluster logs, substitute it into the Softmax function, and assign semantic weights to the server cluster logs;

[0039] Using the attention weighted method, we use the known abnormal operation scenarios of server cluster components as the query matrix, the log semantic set as the query key, and the standardized vector of the server cluster's hardware performance as the query value to calculate the log feature vector of the server cluster.

[0040] The standardized hardware performance vector of the server cluster is used as the hardware node, and the log feature vector of the server cluster is used as the software node. The Pearson correlation coefficient is used to calculate the correlation coefficient between the hardware node and the software node to generate the hardware node-software node connection edge. Sigmoid is used to assign weights to the hardware node-software node fully connected edge, and a fully connected edge network of the server cluster's hardware node-software node is established.

[0041] Extract each adjacent node in the fully connected edge network of the server cluster's hardware nodes and software nodes, establish the server cluster's adjacent node array, and form the server cluster's hardware node-software adjacent node matrix;

[0042] Using the server cluster's hardware node-software adjacent node matrix, we train a GCN graph convolutional network. Using the fully connected layer, we calculate the anomaly probability of each hardware node-software adjacent node, and construct behavioral data for abnormal operation scenarios of the target server cluster components.

[0043] When using, combine the above steps.

[0044] As a further development, current server cluster monitoring mainly relies on threshold alerts and single-dimensional anomaly detection (such as Prometheus and Isolation Forest), while log analysis mostly uses keyword matching (such as the ELK stack). This leads to a semantic separation between hardware performance data and software logs, and a lack of dynamic interactive relationships. Traditional graph neural networks only model static hardware topology and are unable to capture cross-level correlated anomalies and root cause propagation paths, resulting in high false alarm rates and difficulty in locating root causes.

[0045] This solution uses the Pearson coefficient and Sigmoid weighting to achieve cross-modal dynamic association between hardware performance indicators and log semantics. It combines BERT-Softmax and the attention mechanism to enhance key anomaly features, and uses the GCN graph convolutional network to output interpretable anomaly probability distribution and propagation path. It also adopts sliding window normalization and lightweight adjacency matrix design to achieve accurate root cause location.

[0046] Step 2: Conduct causal correlation analysis based on the target server cluster component's abnormal scenario behavior data and known server cluster abnormal scenario behavior performance big data to evaluate the current abnormal deviation status of the target server cluster component;

[0047] The second step includes the following:

[0048] Using the DTW dynamic event regularization algorithm, we recursively calculate the maximum similarity between the big data of known server cluster abnormal scenario behavior performance and the target server cluster component operation abnormal scenario behavior data, filter out the characteristic parameters of known server cluster abnormal scenario behavior performance, and establish the time series data of known target server cluster component operation abnormal scenario performance characteristics;

[0049] As a further example, the abnormal scenario performance characteristics include but are not limited to: abnormal curves (CPU / memory and other indicators change over time), root cause location (Redis connection pool exhaustion)

[0050] Based on the time series data of the performance characteristics of the known target server cluster components in abnormal scenarios, the data is substituted into the fully connected edge network of the hardware nodes and software nodes of the server cluster. According to the computing power call path of the hardware nodes and software nodes, a causal graph of the operation of the known target server cluster components in abnormal scenarios is established;

[0051] Using the independent conditional testing method, the correlation strength of each edge of the causal graph of the known target server cluster component abnormal scenario operation is verified, the causal chain between each node is determined, and the causal structure graph of the known target server cluster component abnormal scenario operation is obtained;

[0052] Based on the MLP multi-layer perceptron, we verify the expected impact of the abnormal scenario operation characteristics under the change of the hardware and software node values ​​in the causal structure diagram of the abnormal scenario operation of the known target server cluster component, and obtain the hardware-software causal intervention representation associated with the abnormal scenario of the known target server cluster component;

[0053] Determine the hardware node-software node operation standardized parameters of the known target server cluster components, and form a standardized parameter array of the known target server cluster components;

[0054] Normalize the hardware-software causal intervention representation associated with the known target server cluster component abnormal scenario, and assign weights to the time series data of the operational performance characteristics of the known target server cluster component abnormal scenario;

[0055] Using the Euclidean distance formula, with the weight of the known target server cluster component's abnormal scenario performance characteristics as the external intervention factor, the spatial distance between the known target server cluster component's abnormal scenario performance characteristics time series data and the known target server cluster group component's standardized parameter array is calculated to determine the target server cluster component's current abnormal deviation status indicator;

[0056]

[0057] in, is the current abnormal deviation status indicator of the i-th component of the target server cluster, The performance weight of the jth abnormal scenario operation of the i-th component of the known target server cluster, The performance characteristic value of the jth abnormal scenario of the i-th component of the known target server cluster, is the normalized parameter of the i-th component of the target server cluster, and n is the total number of components in the target server cluster.

[0058] When using, combine the contents in the above steps.

[0059] As a further example, existing server cluster anomaly correlation analysis relies primarily on simple statistics or rule engines, making it difficult to model complex causal chains. Furthermore, DTW time series matching is only suitable for single-metric analysis and cannot handle multi-component interactions. Causal discovery techniques (such as Granger causality) are prone to misjudgment in heterogeneous hardware and software data. Root cause location methods (such as MicroRCA) rely on manual rules and are unable to adapt to dynamic scenarios. Consequently, existing solutions suffer from static causal assumptions, lack intervention effect analysis, distorted cross-modal distance calculations, and insufficient real-time performance.

[0060] This solution achieves accurate labeling of cross-modal causal chains through joint modeling of DTW time series alignment and causal graphs, combines MLP intervention simulation to quantify the impact of parameter adjustment, uses normalized weights to enhance key features, and uses standardized parameter arrays and Euclidean distance to verify the deviation status of server cluster components under the operating performance characteristics of known target server cluster components in abnormal scenarios, significantly improving the accuracy of causal inference in complex abnormal scenarios.

[0061] Step 3: Based on the current abnormal deviation status of the target server cluster component, mark the abnormal component node and the corresponding associated task sequence, establish a non-interruption task migration resource allocation model, and generate the server cluster abnormal scenario repair task;

[0062] The step three includes the following:

[0063] Based on the target server cluster component's current pending associated task sequence, the fully connected edge network of the server cluster's hardware nodes and software nodes is substituted to determine the component load resource demand characteristic parameters of the pending associated task sequence, and a characteristic vector of the target server cluster component's current pending associated task demand is constructed;

[0064] Using the fully connected edge network of hardware nodes and software nodes of the server cluster, the resource conflict intensity between the characteristic vectors of the current pending associated task requirements of the target server cluster components per unit time is verified, and the priority of the current pending associated task requirements of the target server cluster components is determined;

[0065] Based on logistic regression, the current abnormal deviation status indicator of the target server cluster component is used as a given condition, the component load demand characteristic data per unit time of the associated task sequence to be executed is used as the output, and the shutdown risk probability of the target server cluster component executing the associated task is used as the label data output;

[0066] According to the construction parameters of the target server cluster components, the initialization performance parameters of the target server cluster components are determined, and a digital twin model of the target server cluster components is established.

[0067] The target server cluster component digital twin model is updated using the shutdown risk probability of the server cluster component executing the associated tasks, and the load performance support vector of the target server cluster component is determined;

[0068] The load performance support vector of the target server cluster component is used to train the SVR support vector machine. The priority of the current associated tasks to be executed by the target server cluster component is used as the constraint condition, and the maximization of the completion of the associated tasks to be executed is used as the objective function. The hyperplane boundary of the target server cluster component's load performance execution sequence of tasks to be executed is generated, and the feature vector of the current associated tasks to be executed by the target server cluster component is used as the output, and the server cluster abnormal scenario repair task is used as the output.

[0069] When using, combine the contents in the above steps.

[0070] Furthermore, current task migration technologies (such as the Kubernetes scheduler) primarily rely on static resource quotas and simple rules. Resource conflict detection relies solely on single-dimensional threshold alerts. Digital twin applications are limited to physical device simulation and lack software task integration. Risk prediction models are unable to address complex anomalies. This results in rigid migration strategies failing to detect hidden hardware anomalies, local optimization ignoring task cascading effects, risk quantification lacking real-time anomaly correlation, and lagging digital twin model updates.

[0071] This solution achieves unified representation of hardware-software multi-dimensional constraints through demand feature vector modeling, and combines dynamic conflict assessment to improve the accuracy of compound anomaly identification; uses logistic regression and real-time digital twins to achieve high-precision risk prediction and intervention simulation; improves task completion rate based on SVR hyperplane boundary optimization; safeguards critical task resources through priority-aware scheduling; and improves the accuracy of accurate task risk prediction and optimal resource allocation.

[0072] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating repair tasks for server cluster abnormal scenarios based on big data AI, characterized by: include: S1. Obtain the historical operation logs and real-time operation data of the server cluster, analyze the multi-modal anomaly fusion of the interaction between the hardware layer and the software layer of the server cluster, and generate the target server cluster component operation abnormal scenario behavior data; S2. Perform causal correlation analysis based on the target server cluster component's abnormal scenario behavior data and known server cluster abnormal scenario behavior performance big data to assess the current abnormal deviation status of the target server cluster component; S3. Based on the current abnormal deviation status of the target server cluster component, mark the abnormal component node and the corresponding associated task sequence, establish a non-interruption task migration resource allocation model, and generate the server cluster abnormal scenario repair task; Said S1 comprises: Collect the raw performance parameters of each component of the server cluster according to the unit timestamp, pre-process them using sliding window normalization, and construct a standardized vector of the server cluster's hardware performance. Collect historical operation logs of each component of the server cluster according to unit timestamps, train the BERT model, extract the semantic set of the server cluster logs, substitute it into the Softmax function, and assign semantic weights to the server cluster logs; Using the attention weighted method, we use the known abnormal operation scenarios of server cluster components as the query matrix, the log semantic set as the query key, and the standardized vector of the server cluster's hardware performance as the query value to calculate the log feature vector of the server cluster. Said S1 further comprises: The standardized hardware performance vector of the server cluster is used as the hardware node, and the log feature vector of the server cluster is used as the software node. The Pearson correlation coefficient is used to calculate the correlation coefficient between the hardware node and the software node to generate the hardware node-software node connection edge. Sigmoid is used to assign weights to the hardware node-software node fully connected edge, and a fully connected edge network of the server cluster's hardware node-software node is established. Extract each adjacent node in the fully connected edge network of the server cluster's hardware nodes and software nodes, establish the server cluster's adjacent node array, and form the server cluster's hardware node-software adjacent node matrix; The hardware node-software adjacent node matrix of the server cluster is used to train the GCN graph convolutional network. The fully connected layer is used to calculate the abnormal probability of each hardware node-software adjacent node, and the abnormal scenario behavior data of the target server cluster components are compiled.

2. The method for generating server cluster abnormal scenario repair tasks based on big data AI according to claim 1 is characterized in that: The S2 includes: Using the DTW dynamic event regularization algorithm, we recursively calculate the maximum similarity between the big data of known server cluster abnormal scenario behavior performance and the target server cluster component operation abnormal scenario behavior data, filter out the characteristic parameters of known server cluster abnormal scenario behavior performance, and establish the time series data of known target server cluster component operation abnormal scenario performance characteristics; Based on the time series data of the performance characteristics of the known target server cluster components in abnormal scenarios, the data is substituted into the fully connected edge network of the hardware nodes and software nodes of the server cluster. According to the computing power call path of the hardware nodes and software nodes, a causal graph of the operation of the known target server cluster components in abnormal scenarios is established; Using the independent conditional testing method, the correlation strength of each edge of the causal graph of the known target server cluster component abnormal scenario operation is verified, the causal chain between each node is determined, and the causal structure graph of the known target server cluster component abnormal scenario operation is obtained; Based on the MLP multi-layer perceptron, the expected impact of the corresponding abnormal scenario operation performance characteristics under the numerical values ​​of the hardware nodes and software nodes in the causal structure diagram of the abnormal scenario operation of the known target server cluster component is verified, and the hardware-software causal intervention representation associated with the abnormal scenario of the known target server cluster component is obtained.

3. The method for generating server cluster abnormal scenario repair tasks based on big data AI according to claim 2 is characterized in that: Said S2 further comprises: Determine the hardware node-software node operation standardized parameters of the known target server cluster components, and form a standardized parameter array of the known target server cluster components; Normalize the hardware-software causal intervention representation associated with the known target server cluster component abnormal scenario, and assign weights to the time series data of the operational performance characteristics of the known target server cluster component abnormal scenario; Using the Euclidean distance formula, with the weight of the known target server cluster component's abnormal scenario performance characteristics as the external intervention factor, the spatial distance between the known target server cluster component's abnormal scenario performance characteristics time series data and the known target server cluster group component's standardized parameter array is calculated to determine the target server cluster component's current abnormal deviation status indicator; ； in, is the current abnormal deviation status indicator of the i-th component of the target server cluster, The performance weight of the jth abnormal scenario operation of the i-th component of the known target server cluster, The performance characteristic value of the jth abnormal scenario of the i-th component of the known target server cluster, is the normalized parameter of the i-th component of the target server cluster, and n is the total number of components in the target server cluster.

4. The method for generating server cluster abnormal scenario repair tasks based on big data AI according to claim 3 is characterized in that: The S3 includes: Based on the target server cluster component's current pending associated task sequence, the fully connected edge network of the server cluster's hardware nodes and software nodes is substituted to determine the component load resource demand characteristic parameters of the pending associated task sequence, and a characteristic vector of the target server cluster component's current pending associated task demand is constructed; The fully connected edge network of hardware nodes and software nodes of the server cluster is used to verify the resource conflict intensity between the characteristic vectors of the current pending associated task requirements of the target server cluster components per unit time, and to determine the priority of the current pending associated task requirements of the target server cluster components.

5. The method for generating server cluster abnormal scenario repair tasks based on big data AI according to claim 4 is characterized in that: Said S3 further comprises: Based on logistic regression, the current abnormal deviation status indicator of the target server cluster component is used as a given condition, the component load demand characteristic data per unit time of the associated task sequence to be executed is used as the output, and the shutdown risk probability of the target server cluster component executing the associated task is used as the label data output; According to the construction parameters of the target server cluster components, the initialization performance parameters of the target server cluster components are determined, and a digital twin model of the target server cluster components is established. The target server cluster component digital twin model is updated using the shutdown risk probability of the server cluster component executing the associated tasks, and the load performance support vector of the target server cluster component is determined; The load performance support vector of the target server cluster component is used to train the SVR support vector machine. The priority of the current associated tasks to be executed by the target server cluster component is used as the constraint condition, and the maximization of the completion of the associated tasks to be executed is used as the objective function. The hyperplane boundary of the target server cluster component's load performance execution sequence of tasks to be executed is generated, and the feature vector of the current associated tasks to be executed by the target server cluster component is used as the output, and the server cluster abnormal scenario repair task is used as the output.

Images