A cluster operation and maintenance management system and method based on artificial intelligence

By building an AI-based cluster operation and maintenance management system, and utilizing the collaborative work of master nodes, worker nodes, and compute nodes, combined with deep learning models to analyze real-time operational data, the system solves the problems of server failures and load imbalances in the cluster, and achieves efficient task execution and operation and maintenance management.

CN118694672BActive Publication Date: 2024-11-01北京英沣特能源技术有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411170949.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-01
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

In existing cluster operation and maintenance systems, server failures or unbalanced loads can lead to a decline in cluster processing and analysis capabilities, making it impossible to accurately complete user tasks and effectively manage the operation and maintenance status of master nodes.

Method used

We build an AI-based cluster operation and maintenance management system. Through the collaborative work of master nodes, worker nodes, and computing nodes, we use TCP protocol communication and deep learning models to analyze real-time operation data, predict interference index and interference intervention value, optimize task distribution and control, and avoid load imbalance.

Benefits of technology

It improves the efficiency of cluster task execution, ensures that the cluster accurately completes user tasks, optimizes operation and maintenance management, and reduces operational data discrepancies caused by interference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118694672B_ABST
    Figure CN118694672B_ABST
Patent Text Reader

Abstract

The present invention discloses a cluster operation and maintenance management system and method based on artificial intelligence, belonging to the technical field of cluster operation and maintenance management. The present invention comprises: S10: analyzing the distribution of each task code; S20: predicting the real-time interference index of each other target working node to the selected target working node, and analyzing the interference intervention value of each other target working node to the selected target working node; S30: analyzing the real-time interference of each other target working node to the selected target working node; S40 regulating the operation data of the selected target working node. The present invention predicts the interference index of each other target working node to the target working node by constructing a matrix equation, avoiding the large difference between the operation data of the corresponding target working node and the target operation data due to the interference between each cold source group control subtask, which is conducive to ensuring that the cluster can accurately complete the cold source group control task issued by the user end.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cluster operation and maintenance management, and specifically to a cluster operation and maintenance management system and method based on artificial intelligence. Background Art

[0002] A cluster is a distributed system consisting of multiple computer nodes. These nodes are usually connected through a high-speed network and work together to achieve large-scale data processing and analysis. Clusters can provide high availability, high scalability and high-efficiency computing capabilities, so they have wide application value in the field of big data collection.

[0003] In the existing cluster operation and maintenance system, a server is usually configured to control the operating status of each working node in the cluster. When the configured server fails or the server has an unbalanced load, the cluster's ability to process and analyze data will be reduced. In addition, in the existing system, it is impossible to analyze the interference between the working nodes based on the operating data of each working node in the cluster, resulting in the cluster operation and maintenance system being unable to accurately complete the tasks issued by the user end, and being unable to effectively manage the operation and maintenance of each master node in the cluster. Summary of the invention

[0004] The purpose of the present invention is to provide a cluster operation and maintenance management system and method based on artificial intelligence to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a cluster operation and maintenance management method based on artificial intelligence, the method comprising:

[0006] S10: Based on the cold source group control system, a cluster for executing cold source group control tasks is constructed. The cluster includes a master node, a working node and a computing node. The working nodes of the cluster are the system components in the cold source group control system. The working nodes are used to execute cold source group control subtasks. The master node of the cluster is the control scheduling server in the cold source group control system. The master node is used to schedule and control the system components according to the cold source group control subtasks distributed by the computing nodes. The computing nodes are used to split the cold source group control tasks issued by the user end to obtain several cold source group control subtasks, and distribute each cold source group control subtask to the matching master node. The cluster realizes mutual communication between different nodes through the TCP protocol. The computing node determines the target working node according to the cold source group control tasks issued by the user end, and analyzes the distribution of each task code in combination with the computing power of the master node for each task code converted by the cold source group control subtask.

[0007] S20: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, the collected real-time operation data is compared with the real-time target operation data, the real-time abnormal feature matrix of the selected target working node is analyzed, and the real-time interference index of the other target working nodes to the selected target working node is predicted in combination with the feature matrix analyzed during the abnormal operation time period of the selected target working node, and the interference intervention value of the other target working nodes to the selected target working node is analyzed;

[0008] S30: training the deep learning model using the predicted interference index and interference intervention value to obtain an optimized deep learning model, inputting the real-time operation data of the selected target working node into the optimized deep learning model, and analyzing the real-time interference of other target working nodes to the selected target working node;

[0009] S40: According to the analyzed real-time interference of other target working nodes on the selected target working node, the operation and maintenance decision of the cluster is determined, and the target master node adjusts the operation data of the selected target working node according to the determination result.

[0010] Further, the S10 includes:

[0011] S101: The computing node obtains the cold source group control task issued by the user end, splits the obtained cold source group control task according to the content of the cold source group control task executed by each working node, and obtains a number of cold source group control subtasks. Each working node executes one cold source group control subtask, and determines the target working node according to the execution status of the cold source group control subtask by each working node;

[0012] S102: The data volume of the task code converted from the cold source group control subtask executed by each target working node is obtained, and the target working nodes are selected one by one in the order of the data volume from large to small. The computing node obtains the data volume E of the task code converted from the cold source group control subtask executed by the selected target working node, and the computing node calculates the data volume E of the task code according to γ i =U i / Y i The computing power of the master node numbered i for the task code converted by the selected target working node for the cold source group control subtask is predicted, where i=1,2,…,n, represents the number corresponding to each master node, n represents the total number of master nodes, and Y i Indicates the maximum amount of data that the master node numbered i can store, X i Indicates the amount of task code data stored by the master node numbered i. If Y i -X i -E≤0, then U i=0, if Y i -X i -E>0, then U i =Y i -X i -E;

[0013] S103: with maxγ i The corresponding number is determined, and the determined number is set to k, k=1,2,…,n, then the computing node distributes the task code converted from the cold source group control subtask executed by the selected target working node to the master node numbered k, where max represents the maximum value symbol;

[0014] S104: Repeat the operations of S102-S103 until all target working nodes are selected, and determine the distribution of each task code to avoid load imbalance between the master nodes, thereby affecting the cluster's execution efficiency of the cold source group control task and causing the cold source group control system to fail to operate normally.

[0015] Further, the S20 includes:

[0016] S201: The master node receiving the task code is recorded as the target master node, and a target master node is randomly selected. If the task code received by the selected target master node is obtained by converting the cold source group control subtask executed by the selected target working node, a communication channel is established between the selected master node and the selected target working node through the TCP protocol, and the selected master node transmits the received task code to the selected target working node through the established communication channel;

[0017] S202: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, and the real-time target operation data of the selected target working node is obtained, and the collected real-time operation data is compared with the obtained real-time target operation data, and a plane rectangular coordinate system is constructed with time as the horizontal coordinate and the operation data as the vertical coordinate, and the real-time operation data that are inconsistent with the comparison are represented in the plane rectangular coordinate system, and the real-time operation data points represented in the plane rectangular coordinate system are numbered in chronological order, and the numbering result is: j=1,2,…,m; m represents the total number, according to T=[g 12 ,g 23 ,…,g (m-1)m ] determines the real-time abnormal feature matrix of the selected target working node, where g (m-1)mIt represents the slope value calculated by the real-time operation data points numbered m-1 and m. Based on the determination method of the real-time abnormal feature matrix, the feature matrix of each other target working node in the time period [t1, t1+d*(m-1)] is determined, where t1 represents the collection time corresponding to the real-time operation data numbered 1, and d represents the collection interval time of the operation data;

[0018] S203: According to the matrix equation The real-time interference index of other target working nodes to the selected target working node is analyzed, where q=1,2,…,p, represents the corresponding number of other target working nodes, p represents the total number of other target working nodes, and a q represents the interference index of the target working node numbered q on the selected target working node, W q represents the characteristic matrix of the target working node numbered q in the time period [t1, t1+d*(m-1)], T´ represents the real-time characteristic matrix obtained according to the real-time target operation data of the selected target working node, and W q ´ represents the feature matrix obtained according to the real-time target operation data of the target working node numbered q;

[0019] S204: According to W q -W q 'Analyze the interference intervention value of the target working node numbered q on the selected target working node, record W q -W q =[s q12 ,s q23 ,…,s q(m-1)m ], if s q12 +…+s qc(c+1) = 0 and s q12 +…+s q(c+1)(c+2) ≠0, c=1,2,…,m-2, then the target operation data corresponding to the target working node numbered q at the time point t1+d*(c+1) is the interference intervention value of the target working node numbered q on the selected target working node.

[0020] Further, the S30 includes:

[0021] S301: Determine the interference index of the target working node numbered q under different interference intervention values ​​to the selected target working node, obtain several sets of training data, each set of training data includes an interference index and an interference intervention value, and put each set of training data obtained into a set A q In the example, 7:3 is used as set A. qThe training set and the test set are obtained respectively according to the division ratio. The deep learning model f(v) = w × v + b is trained by the training set, and the trained deep learning model is tested by the test set. The weight w and the bias b in the deep model f(v) = w × v + b are adjusted to obtain the optimized deep learning model. v represents the interference intervention value, and f(v) represents the interference index.

[0022] S302: Input the operating data of the target working node numbered q at time z into the optimized deep learning model, and calculate the interference index a of the target working node numbered q on the selected target working node at time z. qz to confirm.

[0023] Further, the S40 includes: selecting the target operation data r of the target working node at time z z , and the running data f of the selected target working node at time z z To obtain, if f z =r z , the task code of the target master node selected by the cluster scheduling control is not adjusted. If f z ≠r z , the task code of the target master node of the target work node selected by the scheduling control in the cluster is adjusted. The task code adjustment result is: z Adjust to ;

[0024] The target master node of the target working node selected by the scheduling control in the cluster is adjusted according to the value , and adjust the operating parameters of the selected target working node.

[0025] A cluster operation and maintenance management system based on artificial intelligence, the system comprising a task distribution analysis module, an operation data processing module, an interference analysis module and a cluster operation and maintenance management module;

[0026] The task distribution analysis module is used to analyze the distribution of each task code;

[0027] The operation data processing module is used to predict the real-time interference index of each other target working node to the selected target working node, and to analyze the interference intervention value of each other target working node to the selected target working node;

[0028] The interference analysis module analyzes the real-time interference of other target working nodes to the selected target working node;

[0029] The cluster operation and maintenance management module is used to perform operation and maintenance management on the scheduling control of each target master node on the target working node in the cluster.

[0030] Furthermore, the task distribution analysis module includes a target working node determination unit, a computing power capability prediction unit and a task code distribution unit;

[0031] The target working node determination unit determines the target working node according to the execution status of each working node on the cold source group control subtask;

[0032] The computing power prediction unit predicts the computing power of each master node for the task code converted from the cold source group control subtask executed by the selected target working node according to the data volume of the task code converted from the cold source group control subtask executed by the selected target working node, the storage capacity of each master node, and the storage of the task code by each master node;

[0033] The task code distribution unit receives the prediction result transmitted by the computing power prediction unit, and the computing node distributes the task code to the matching master node based on the received information.

[0034] Further, the operation data processing module includes a task code delivery unit, a characteristic matrix determination unit, an interference index prediction unit and an interference intervention value analysis unit;

[0035] The task code transmission unit transmits the task code received in the selected master node to the selected target working node through the communication channel constructed between the selected master node and the selected target working node;

[0036] The characteristic matrix determination unit collects the real-time operation data of the selected target working node and acquires the real-time target operation data of the selected target working node when the selected target working node starts to execute the task code, compares the collected real-time operation data with the acquired real-time target operation data, and determines the real-time abnormal characteristic matrix of the selected target working node and the characteristic matrices analyzed from other target working nodes during the abnormal operation time period of the selected target working node based on the comparison result;

[0037] The interference index prediction unit constructs a matrix equation based on the characteristic matrix transmitted by the characteristic matrix determination unit and the real-time abnormal characteristic matrix, and predicts the interference value of each other target working node to the selected target working node;

[0038] The interference intervention value analysis unit analyzes the interference intervention values ​​of other target working nodes to the selected target working node according to the characteristic matrix transmitted by the characteristic matrix determination unit.

[0039] Further, the interference analysis module includes an optimized deep learning model training unit and an interference situation analysis unit;

[0040] The optimized deep learning model training unit obtains a training set and a test set of the deep learning model according to the interference index of each other target working node on the selected target working node under different interference intervention values, and uses the training set and the test set to train and test the deep training model respectively to obtain an optimized deep learning model;

[0041] The interference situation analysis unit inputs the real-time operation data of other target working nodes into the optimized deep learning model, and determines the real-time interference index of other target working nodes to the selected target working node.

[0042] Furthermore, the cluster operation and maintenance management module compares the real-time target operation data of the selected target node with the real-time operation data of the selected target working node, and manages the operation and maintenance status of the target master node that schedules and controls the selected target working node in the cluster.

[0043] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0044] 1. The present invention determines the distribution of each task code by the computing power of each master node for each task code. During the distribution process, it can avoid load imbalance among the master nodes, thereby ensuring the execution efficiency of the cluster for the cold source group control task.

[0045] 2. The present invention compares the collected real-time operation data with the acquired real-time target operation data, determines the real-time abnormal feature matrix of the selected target working node and the feature matrix of other target working nodes, and constructs a matrix equation based on the determination result to predict the interference index of other target working nodes to the target working node, thereby avoiding the large difference between the operation data of the corresponding target working node and the target operation data due to the interference between the cold source group control subtasks, which is beneficial to ensure that the cluster can accurately complete the cold source group control tasks issued by the user end.

[0046] 3. The present invention finds the relationship between the interference intervention value and the interference index, which is conducive to the rapid operation and maintenance management of the target operation data of each target master node, further improving the use effect of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 It is a workflow diagram of an artificial intelligence-based cluster operation and maintenance management system and method of the present invention.

[0049] Figure 2It is a schematic diagram of the working principle structure of an artificial intelligence-based cluster operation and maintenance management system and method of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1 and Figure 2 The present invention provides a technical solution: a cluster operation and maintenance management method based on artificial intelligence, the method comprising:

[0052] S10: Based on the cold source group control system, a cluster for executing cold source group control tasks is constructed. The system components in the cold source group control system include refrigeration pumps, cooling pumps, cooling towers, etc. The cluster includes a master node, a working node, and a computing node. The working nodes of the cluster are the system components in the cold source group control system, and the master node of the cluster is the control scheduling server in the cold source group control system. The cluster realizes mutual communication between different nodes through the TCP protocol. The TCP protocol is a connection-oriented, reliable, byte stream-based transport layer communication protocol. The computing node determines the target working node according to the cold source group control task issued by the user end, and analyzes the distribution of each task code in combination with the computing power of the master node for each task code converted by the cold source group control subtask;

[0053] S10 includes:

[0054] S101: The computing node obtains the cold source group control task issued by the user end, splits the obtained cold source group control task according to the content of the cold source group control task executed by each working node, and obtains a number of cold source group control subtasks. Each working node executes one cold source group control subtask. According to the execution status of the cold source group control subtask by each working node, the target working node is determined. The target working node refers to the working node that needs to execute the cold source group control subtask;

[0055] S102: The data volume of the task code converted from the cold source group control subtask executed by each target working node is obtained, and the target working nodes are selected one by one in order of data volume from large to small. The computing node obtains the data volume E of the task code converted from the cold source group control subtask executed by the selected target working node. The specific method for converting the cold source group control subtask into the task code is as follows: obtaining the configuration file and the preset task template corresponding to the cold source group control subtask, generating the scaffolding code corresponding to the cold source group control subtask according to the configuration file and the preset task template, the configuration file includes a task definition file and a parameter type definition file, the task definition file includes the task name of the cold source group control subtask, the input parameters and output parameters of the cold source group control subtask, the parameter type definition file includes the type of the input parameter and the type of the output parameter, generating the task code of the cold source group control subtask according to the scaffolding code and the business logic code, the business logic code includes the scheduling order information and the scheduling mode information of each cold source group control subtask, the scheduling mode information includes synchronous call and asynchronous call, the method belongs to the prior art, and the computing node generates the task code according to the γ i =U i / Y i The computing power of the master node numbered i for the task code converted by the selected target working node for the cold source group control subtask is predicted, where i=1,2,…,n, represents the number corresponding to each master node, n represents the total number of master nodes, and Y i Indicates the maximum amount of data that the master node numbered i can store, X i Indicates the amount of task code data stored by the master node numbered i. If Y i -X i -E≤0, then U i =0, if Y i -X i -E>0, then U i =Y i -X i -E;

[0056] S103: with maxγ i The corresponding number is determined, and the determined number is set to k, k=1,2,…,n, then the computing node distributes the task code converted from the cold source group control subtask executed by the selected target working node to the master node numbered k, where max represents the maximum value symbol;

[0057] S104: Repeat the operations of S102-S103 until all target working nodes are selected, and determine the distribution of each task code;

[0058] S20: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, the collected real-time operation data is compared with the real-time target operation data, the real-time abnormal feature matrix of the selected target working node is analyzed, and the real-time interference index of the other target working nodes to the selected target working node is predicted in combination with the feature matrix analyzed during the abnormal operation time period of the selected target working node, and the interference intervention value of the other target working nodes to the selected target working node is analyzed;

[0059] The S20 includes:

[0060] S201: The master node receiving the task code is recorded as the target master node, and a target master node is randomly selected. If the task code received by the selected target master node is obtained by converting the cold source group control subtask executed by the selected target working node, a communication channel is established between the selected master node and the selected target working node through the TCP protocol, and the selected master node transmits the received task code to the selected target working node through the established communication channel;

[0061] S202: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, and the real-time target operation data of the selected target working node is obtained. When the target working node is a cooling pump, the operation data is the temperature of the cooling water in the cooling pump. When the target working node is a freezing pump, the operation data is the temperature of the freezing pump. The target operation data refers to the operation data of the target working node analyzed according to the task code. The collected real-time operation data is compared with the obtained real-time target operation data. A plane rectangular coordinate system is constructed with time as the horizontal coordinate and the operation data as the vertical coordinate. The real-time operation data that are inconsistent with the comparison are represented in the plane rectangular coordinate system. The real-time operation data points represented in the plane rectangular coordinate system are numbered in chronological order. The numbering result is: j=1,2,…,m; m represents the total number. According to T=[g 12 ,g 23 ,…,g (m-1)m ] determines the real-time abnormal feature matrix of the selected target working node, where g (m-1)m It represents the slope value calculated by the real-time operation data points numbered m-1 and m. Based on the determination method of the real-time abnormal feature matrix, the feature matrix of each other target working node in the time period [t1, t1+d*(m-1)] is determined, where t1 represents the collection time corresponding to the real-time operation data numbered 1, and d represents the collection interval time of the operation data;

[0062] S203: According to the matrix equation The real-time interference index of other target working nodes to the selected target working node is analyzed, where q=1,2,…,p, represents the number corresponding to other target working nodes, p represents the total number of other target working nodes, and other target working nodes refer to target working nodes other than the selected target working node, a q represents the interference index of the target working node numbered q on the selected target working node, W q represents the characteristic matrix of the target working node numbered q in the time period [t1, t1+d*(m-1)], T´ represents the real-time characteristic matrix obtained according to the real-time target operation data of the selected target working node, and W q ´ represents the feature matrix obtained according to the real-time target operation data of the target working node numbered q;

[0063] S204: According to W q -W q 'Analyze the interference intervention value of the target working node numbered q on the selected target working node, record W q -W q =[s q12 ,s q23 ,…,s q(m-1)m ], if s q12 +…+s qc(c+1) = 0 and s q12 +…+s q(c+1)(c+2) ≠0, c=1,2,…,m-2, then the target operation data corresponding to the target working node numbered q at the time point t1+d*(c+1) is the interference intervention value of the target working node numbered q on the selected target working node;

[0064] S30: training the deep learning model using the predicted interference index and interference intervention value to obtain an optimized deep learning model, inputting the real-time operation data of the selected target working node into the optimized deep learning model, and analyzing the real-time interference of other target working nodes to the selected target working node;

[0065] S30 includes:

[0066] S301: Determine the interference index of the target working node numbered q under different interference intervention values ​​to the selected target working node, obtain several sets of training data, each set of training data includes an interference index and an interference intervention value, and put each set of training data obtained into a set A q In the example, 7:3 is used as set A. qThe training set and the test set are obtained respectively according to the division ratio. The deep learning model f(v) = w × v + b is trained by the training set, and the trained deep learning model is tested by the test set. The weight w and the bias b in the deep model f(v) = w × v + b are adjusted to obtain the optimized deep learning model. v represents the interference intervention value, and f(v) represents the interference index.

[0067] S302: Input the operating data of the target working node numbered q at time z into the optimized deep learning model, and calculate the interference index a of the target working node numbered q on the selected target working node at time z. qz to determine;

[0068] S40: according to the analyzed real-time interference of other target working nodes on the selected target working node, the operation and maintenance decision of the cluster is determined, and the target master node adjusts and controls the operation data of the selected target working node according to the determination result;

[0069] S40 includes: performing target operation data r of the selected target working node at time z z , and the running data f of the selected target working node at time z z To obtain, if f z =r z , the task code of the target master node selected by the cluster scheduling control is not adjusted. If f z ≠r z , the task code of the target master node of the target work node selected by the scheduling control in the cluster is adjusted. The task code adjustment result is: z Adjust to ;

[0070] The target master node of the target working node selected by the scheduling control in the cluster is adjusted according to the value , and adjust the operating parameters of the selected target working node.

[0071] A cluster operation and maintenance management system based on artificial intelligence, the system includes a task distribution analysis module, an operation data processing module, an interference analysis module and a cluster operation and maintenance management module;

[0072] The task distribution analysis module is used to analyze the distribution of each task code;

[0073] The task distribution analysis module includes a target working node determination unit, a computing power capacity prediction unit, and a task code distribution unit;

[0074] The target working node determination unit determines the target working node according to the execution status of each working node on the cold source group control subtask;

[0075] The computing power prediction unit predicts the computing power of each master node for the task code converted from the cold source group control subtask executed by the selected target working node according to the data volume of the task code converted from the cold source group control subtask executed by the selected target working node, the storage capacity of each master node, and the storage status of the task code of each master node;

[0076] The task code distribution unit receives the prediction results transmitted by the computing power prediction unit, and the computing node distributes the task code to the matching master node based on the received information;

[0077] The operation data processing module is used to predict the real-time interference index of each other target working node to the selected target working node, and to analyze the interference intervention value of each other target working node to the selected target working node;

[0078] The operation data processing module includes a task code delivery unit, a characteristic matrix determination unit, an interference index prediction unit and an interference intervention value analysis unit;

[0079] The task code transmission unit transmits the task code received in the selected master node to the selected target working node through the communication channel established between the selected master node and the selected target working node;

[0080] When the selected target working node starts to execute the task code, the characteristic matrix determination unit collects the real-time operation data of the selected target working node and obtains the real-time target operation data of the selected target working node, compares the collected real-time operation data with the obtained real-time target operation data, and determines the real-time abnormal characteristic matrix of the selected target working node and the characteristic matrices analyzed from other target working nodes during the abnormal operation time period of the selected target working node based on the comparison result;

[0081] The interference index prediction unit constructs a matrix equation based on the characteristic matrix and the real-time abnormal characteristic matrix transmitted by the characteristic matrix determination unit, and predicts the interference value of each other target working node to the selected target working node;

[0082] The interference intervention value analysis unit analyzes the interference intervention values ​​of other target working nodes to the selected target working node according to the characteristic matrix transmitted by the characteristic matrix determination unit;

[0083] The interference analysis module analyzes the real-time interference of other target working nodes to the selected target working node;

[0084] The interference analysis module includes an optimized deep learning model training unit and an interference situation analysis unit;

[0085] The deep learning model optimization training unit obtains a training set and a test set of the deep learning model according to the interference index of other target working nodes on the selected target working node under different interference intervention values, and uses the training set and the test set to train and test the deep training model respectively to obtain an optimized deep learning model;

[0086] The interference situation analysis unit inputs the real-time operation data of each other target working node into the optimized deep learning model, and determines the real-time interference index of each other target working node to the selected target working node;

[0087] The cluster operation and maintenance management module is used to perform operation and maintenance management on the scheduling control of each target master node on the target working node in the cluster;

[0088] The cluster operation and maintenance management module compares the real-time target operation data of the selected target node with the real-time operation data of the selected target working node, and manages the operation and maintenance status of the target master node that schedules and controls the selected target working node in the cluster.

[0089] Example 1: Assuming that the data volume of the task code converted by the cold source group control subtask executed by the selected target working node is E=40 bits, the maximum data volume Y1=200 bits that the master node numbered 1 can store, and the data volume X1=50 bits of the task code stored by the master node numbered 1, the computing power capability of the master node numbered 1 for the task code converted by the cold source group control subtask executed by the selected target working node is:

[0090] Since Y1-X1-E=200-50-40=110>0, U1=Y1-X1-E=110;

[0091] γ1=U1 / Y1=110 / 200=0.55;

[0092] The computing power of the master node numbered 1 for the task code converted by the selected target working node for the cold source group control subtask is 0.55.

[0093] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0094] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cluster operation and maintenance management method based on artificial intelligence, characterized in that: The method comprises: S10: Based on the cold source group control system, a cluster for executing cold source group control tasks is constructed. The cluster includes a master node, a working node, and a computing node. The working nodes of the cluster are the system components in the cold source group control system, and the master node of the cluster is the control scheduling server in the cold source group control system. The cluster realizes mutual communication between different nodes through the TCP protocol. The computing node determines the target working node according to the cold source group control task issued by the user end, and analyzes the distribution of each task code in combination with the computing power of the master node for each task code converted by the cold source group control subtask; The S10 includes: S101: The computing node obtains the cold source group control task issued by the user end, splits the obtained cold source group control task according to the content of the cold source group control task executed by each working node, and obtains a number of cold source group control subtasks. Each working node executes one cold source group control subtask, and determines the target working node according to the execution status of the cold source group control subtask by each working node; S102: The data volume of the task code converted from the cold source group control subtask executed by each target working node is obtained, and the target working nodes are selected one by one in the order of the data volume from large to small. The computing node obtains the data volume E of the task code converted from the cold source group control subtask executed by the selected target working node, and the computing node calculates the data volume E of the task code according to γ i =U i / Y i The computing power of the master node numbered i for the task code converted by the selected target working node for the cold source group control subtask is predicted, where i=1,2,…,n, represents the number corresponding to each master node, n represents the total number of master nodes, and Y i Indicates the maximum amount of data that the master node numbered i can store, X i Indicates the amount of task code data stored by the master node numbered i. If Y i -X i -E≤0, then U i =0, if Y i -X i -E>0, then U i =Y i -X i -E; S103: with maxγ i The corresponding number is determined, and the determined number is set to k, k=1,2,…,n, then the computing node distributes the task code converted from the cold source group control subtask executed by the selected target working node to the master node numbered k, where max represents the maximum value symbol; S104: Repeat the operations of S102-S103 until all target working nodes are selected, and determine the distribution of each task code; S20: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, the collected real-time operation data is compared with the real-time target operation data, the real-time abnormal feature matrix of the selected target working node is analyzed, and the real-time interference index of the other target working nodes to the selected target working node is predicted in combination with the feature matrix analyzed during the abnormal operation time period of the selected target working node, and the interference intervention value of the other target working nodes to the selected target working node is analyzed; S30: training the deep learning model using the predicted interference index and interference intervention value to obtain an optimized deep learning model, inputting the real-time operation data of the selected target working node into the optimized deep learning model, and analyzing the real-time interference of other target working nodes to the selected target working node; S40: According to the analyzed real-time interference of other target working nodes on the selected target working node, the operation and maintenance decision of the cluster is determined, and the target master node adjusts the operation data of the selected target working node according to the determination result.

2. The cluster operation and maintenance management method based on artificial intelligence according to claim 1, characterized in that: The S20 includes: S201: The master node receiving the task code is recorded as the target master node, and a target master node is randomly selected. If the task code received by the selected target master node is obtained by converting the cold source group control subtask executed by the selected target working node, a communication channel is established between the selected master node and the selected target working node through the TCP protocol, and the selected master node transmits the received task code to the selected target working node through the established communication channel; S202: When the selected target working node starts to execute the task code, the real-time operation data of the selected target working node is collected, and the real-time target operation data of the selected target working node is obtained, and the collected real-time operation data is compared with the obtained real-time target operation data, and a plane rectangular coordinate system is constructed with time as the horizontal coordinate and the operation data as the vertical coordinate, and the real-time operation data that are inconsistent with the comparison are represented in the plane rectangular coordinate system, and the real-time operation data points represented in the plane rectangular coordinate system are numbered in chronological order, and the numbering result is: j=1,2,…,m; m represents the total number, according to T=[g 12 ,g 23 ,…,g (m-1)m ] determines the real-time abnormal feature matrix of the selected target working node, where g (m-1)m It represents the slope value calculated by the real-time operation data points numbered m-1 and m. Based on the determination method of the real-time abnormal feature matrix, the feature matrix of each other target working node in the time period [t1, t1+d*(m-1)] is determined, where t1 represents the collection time corresponding to the real-time operation data numbered 1, and d represents the collection interval time of the operation data; S203: According to the matrix equation The real-time interference index of other target working nodes to the selected target working node is analyzed, where q=1,2,…,p, represents the corresponding number of other target working nodes, p represents the total number of other target working nodes, and a q represents the interference index of the target working node numbered q on the selected target working node, W q represents the characteristic matrix of the target working node numbered q in the time period [t1, t1+d*(m-1)], T´ represents the real-time characteristic matrix obtained according to the real-time target operation data of the selected target working node, and W q ´ represents the feature matrix obtained according to the real-time target operation data of the target working node numbered q; S204: According to W q -W q 'Analyze the interference intervention value of the target working node numbered q on the selected target working node, record W q -W q =[s q12 ,s q23 ,…,s q(m-1)m ],s q(m-1)m Represents the feature matrix W q The value of the m-1th element in the feature matrix W q The difference between the values ​​of the m-1th element in ´, if s q12 +…+s qc(c+1) = 0 and s q12 +…+s q(c+1)(c+2) ≠0, c=1,2,…,m-2, then the target operation data corresponding to the target working node numbered q at the time point t1+d*(c+1) is the interference intervention value of the target working node numbered q on the selected target working node.

3. The cluster operation and maintenance management method based on artificial intelligence according to claim 2, characterized in that: The S30 includes: S301: Determine the interference index of the target working node numbered q under different interference intervention values ​​to the selected target working node, obtain several sets of training data, each set of training data includes an interference index and an interference intervention value, and put each set of training data obtained into a set A q In the example, 7:3 is used as set A. q The training set and the test set are obtained respectively according to the division ratio. The deep learning model f(v) = w × v + b is trained by the training set, and the trained deep learning model is tested by the test set. The weight w and the bias b in the deep model f(v) = w × v + b are adjusted to obtain the optimized deep learning model. v represents the interference intervention value, and f(v) represents the interference index. S302: Input the operating data of the target working node numbered q at time z into the optimized deep learning model, and calculate the interference index a of the target working node numbered q on the selected target working node at time z. qz to confirm.

4. The cluster operation and maintenance management method based on artificial intelligence according to claim 3 is characterized by: The step S40 includes: selecting the target operation data r of the target working node at time z z , and the running data f of the selected target working node at time z z To obtain, if f z =r z , the task code of the target master node selected by the cluster scheduling control is not adjusted. If f z ≠r z , the task code of the target master node of the target work node selected by the scheduling control in the cluster is adjusted. The task code adjustment result is: z Adjust to ; The target master node of the target working node selected by the scheduling control in the cluster is adjusted according to the value , and adjust the operating parameters of the selected target working node.

5. An artificial intelligence-based cluster operation and maintenance management system applied to the artificial intelligence-based cluster operation and maintenance management method according to any one of claims 1 to 4, characterized in that: The system includes a task distribution analysis module, an operation data processing module, an interference analysis module and a cluster operation and maintenance management module; The task distribution analysis module is used to analyze the distribution of each task code; The task distribution analysis module includes a target working node determination unit, a computing power capacity prediction unit and a task code distribution unit; The target working node determination unit determines the target working node according to the execution status of each working node on the cold source group control subtask; The computing power prediction unit predicts the computing power of each master node for the task code converted from the cold source group control subtask executed by the selected target working node according to the data volume of the task code converted from the cold source group control subtask executed by the selected target working node, the storage capacity of each master node, and the storage of the task code by each master node; The task code distribution unit receives the prediction result transmitted by the computing power prediction unit, and the computing node distributes the task code to the matching master node according to the received information; The operation data processing module is used to predict the real-time interference index of each other target working node to the selected target working node, and to analyze the interference intervention value of each other target working node to the selected target working node; The interference analysis module analyzes the real-time interference of other target working nodes to the selected target working node; The cluster operation and maintenance management module is used to perform operation and maintenance management on the scheduling control of each target master node on the target working node in the cluster.

6. The cluster operation and maintenance management system based on artificial intelligence according to claim 5, characterized in that: The operation data processing module includes a task code delivery unit, a feature matrix determination unit, an interference index prediction unit and an interference intervention value analysis unit; The task code transmission unit transmits the task code received in the selected master node to the selected target working node through the communication channel constructed between the selected master node and the selected target working node; The characteristic matrix determination unit collects the real-time operation data of the selected target working node and acquires the real-time target operation data of the selected target working node when the selected target working node starts to execute the task code, compares the collected real-time operation data with the acquired real-time target operation data, and determines the real-time abnormal characteristic matrix of the selected target working node and the characteristic matrices analyzed from other target working nodes during the abnormal operation time period of the selected target working node based on the comparison result; The interference index prediction unit constructs a matrix equation based on the characteristic matrix transmitted by the characteristic matrix determination unit and the real-time abnormal characteristic matrix, and predicts the interference value of each other target working node to the selected target working node; The interference intervention value analysis unit analyzes the interference intervention values ​​of other target working nodes to the selected target working node according to the characteristic matrix transmitted by the characteristic matrix determination unit.

7. The artificial intelligence-based cluster operation and maintenance management system according to claim 6, characterized in that: The interference analysis module includes an optimized deep learning model training unit and an interference situation analysis unit; The optimized deep learning model training unit obtains a training set and a test set of the deep learning model according to the interference index of each other target working node on the selected target working node under different interference intervention values, and uses the training set and the test set to train and test the deep training model respectively to obtain an optimized deep learning model; The interference situation analysis unit inputs the real-time operation data of other target working nodes into the optimized deep learning model, and determines the real-time interference index of other target working nodes to the selected target working node.

8. The artificial intelligence-based cluster operation and maintenance management system according to claim 7, characterized in that: The cluster operation and maintenance management module compares the real-time target operation data of the selected target node with the real-time operation data of the selected target working node, and manages the operation and maintenance status of the target master node that schedules and controls the selected target working node in the cluster.