A method, storage medium, device and computer program product for root cause analysis of alarms supporting continuous training

Abstract

The present invention discloses an alarm root cause analysis method, storage medium, device and computer program product supporting continuous training, including: collecting alarms occurring in the operation and maintenance process of a telecommunication network, normalizing the alarms, and constructing an alarm type causal relationship graph through a directed graph for the processed alarms; when an alarm occurs, normalizing the alarms, constructing a borderless alarm relationship graph, adding directed edges to the borderless alarm relationship graph according to the alarm type causal relationship graph and the equipment topology in the telecommunication network, and obtaining a directed alarm relationship graph; taking the occurring alarm as a starting point, using a DFS algorithm to reversely traverse the directed alarm relationship graph, obtaining several alarm propagation paths, summing the weights of all edges in each propagation path, and taking the three propagation paths with the largest sum of weights as the results of the alarm root cause analysis. The present invention improves the accuracy of alarm root cause analysis, narrows the scope of fault finding in the operation and maintenance process of a telecommunication network, and quickly locates faults.

Description

Technical Field

[0001] The present invention relates to the technical field of alarm root cause analysis, and in particular to an alarm root cause analysis method, storage medium, device and computer program product supporting continuous training. Background Art

[0002] In telecommunications networks, due to the large scale of the network and the close interconnection between devices, alarms are often needed to monitor the operation status during daily operation and maintenance. A single fault point in the network may trigger a large number of different types of alarms on multiple connected devices, which results in multiple alarms when a fault occurs. In this case, relying solely on manual troubleshooting can no longer meet current needs. Some methods or analysis systems are needed to assist operation and maintenance personnel in root cause analysis of the fault, narrow the scope of troubleshooting, and quickly repair the fault.

[0003] In the current technical solution, there is a method to achieve the root cause analysis of alarms by managing alarm root cause analysis rules and accepting real-time alarms to match the alarm root cause analysis rules. The main steps are: 1. Manually define alarm root cause judgment rules; 2. Accept real-time alarms and perform preprocessing; 3. Match the processed alarms with the previously defined alarm root cause judgment rules and output the analysis results. This method is more flexible by configuring rules on the page, and can be modified as needed. However, because the alarm root cause judgment rules need to be manually maintained, it has extremely high requirements for users and the subsequent maintenance of the alarm root cause judgment rules will bring a certain amount of workload.

[0004] Another method to implement root cause analysis is to analyze alarm data by combining AI algorithms and knowledge graphs. The main steps are as follows: 1. Construct real-time alarm data into an alarm relationship knowledge graph; 2. Find relevant historical alarm knowledge graphs; 3. Use the graph attention network GAT to embed and align the real-time alarm knowledge graph and the historical alarm knowledge graph. By comparing the similarity between the real-time alarm knowledge graph and the historical alarm knowledge graph, alarms greater than the set threshold are taken out to form a target alarm knowledge graph; 4. Determine the alarm root cause in the target alarm knowledge graph based on the set root factor. This method uses AI to combine knowledge graphs for root cause analysis and does not require labeled data, but does not consider further improving the accuracy of the analysis results by returning the subsequent alarm root cause analysis results to the database. Summary of the invention

[0005] In response to the problems existing in the prior art, the present invention provides an alarm root cause analysis method, storage medium, device and computer program product that support continuous training. By constructing an alarm type causal relationship diagram, the accuracy of alarm root cause analysis is improved, thereby performing fault propagation analysis on the alarm, narrowing the fault search scope during the operation and maintenance of the telecommunications network, and quickly locating the fault.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solution: a method for root cause analysis of alarms supporting continuous training, specifically comprising the following steps:

[0007] Step S1: Collect alarms generated during the operation and maintenance of the telecommunication network and perform normalization processing;

[0008] Step S2: construct an alarm type causal relationship graph through a directed graph using the normalized alarms;

[0009] Step S3: when an alarm occurs during the operation and maintenance of the telecommunication network, the alarm is normalized to construct a borderless alarm relationship graph, and directed edges are added to the borderless alarm relationship graph according to the alarm type causal relationship graph and the device topology in the telecommunication network to obtain a directed alarm relationship graph;

[0010] Step S4: Taking the alarm generated in step S3 as the starting point, use the DFS algorithm to reversely traverse the directed alarm relationship graph to obtain several alarm propagation paths, sum the weights of all edges in each propagation path, and take the three propagation paths with the largest sum of weights as the results of the alarm root cause analysis.

[0011] Furthermore, the normalized alarms include: alarm type, alarm unique identifier, alarm occurrence location and alarm occurrence time.

[0012] Furthermore, step S2 includes the following sub-steps:

[0013] Step S2.1, constructing an edgeless directed graph with the alarm categories in the normalized alarms as nodes;

[0014] Step S2.2: construct a directed candidate graph according to the edge adding rule for all nodes in the edgeless directed graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function, and use the directed candidate graph with the highest likelihood score as the initial alarm type causal relationship graph for training;

[0015] Step S2.3, update the directed candidate graph according to the edge addition rule for all nodes in the alarm type causal relationship graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function;

[0016] Step S2.4: If the highest likelihood score exceeds the highest likelihood score of the previous iteration and the number of training iterations is less than the maximum number of iterations, the directed candidate graph with the highest likelihood score is used as the alarm type causal relationship graph, and step S2.3 is repeated until the number of training iterations reaches the maximum number of iterations or the highest likelihood score is less than or equal to the highest likelihood score of the previous iteration, and the optimal alarm type causal relationship graph in one round of iteration is obtained;

[0017] Step S2.5, using conditional probability to calculate the weight of each edge in the optimal alarm type causal relationship graph;

[0018] Step S2.6: Repeat steps S2.1-S2.6 in each iteration until the maximum number of iterations is reached to obtain the current optimal alarm type causal relationship diagram.

[0019] Furthermore, the edge addition rule satisfies:

[0020] i. Each node among all nodes forms a tuple with other nodes (node ​​A, node B);

[0021] ii. For each tuple (node ​​A, node B), if there is an edge from node A to node B in the alarm type causal relationship graph, but it is marked as an error in the feedback information, delete the edge from node A to node B; the feedback information refers to marking the edge in the alarm causal relationship graph as correct or wrong based on expert experience;

[0022] iii. If there is no edge from node A to node B in the alarm type causal relationship graph, and the edge from node A to node B is not marked incorrectly in the feedback information, and the edge from node B to node A in the feedback information is not marked correctly, add the edge from node A to node B in the alarm type causal relationship graph. If there is still an edge from node A to node B in the alarm type causal relationship, delete the edge from node A to node B.

[0023] Furthermore, the specific process of the THP likelihood function calculating the likelihood score of each directed candidate graph is as follows:

[0024]

[0025] Among them, V table is the collection of all alarm types in the operation and maintenance process of the telecommunication network, T represents the time collection of alarm collection, N represents all device nodes in the operation and maintenance process of the telecommunication network, and X n,v,t is the number of times the vth type of alarm occurs on the nth device node within time t, PA v Indicates the alarm pointing to the vth type in the directed candidate graph. is the PA within two hops from the nth device node within time t v Number of occurrences.

[0026] Furthermore, the specific process of step S2.5 is as follows:

[0027] A. For the two types of alarms on each edge of the optimal alarm type causal relationship graph, the number of occurrences of one alarm is counted as num(C), and within the alarm occurrence time, the number of occurrences of another alarm within two hops from the alarm generating device is counted as num(CD);

[0028] B. If the edge is marked as correct in the feedback information, the weight of the edge is weight(node ​​C, node D) = [num(CD) / num(C)+weight old (node ​​C, node D)] / 2; otherwise, the weight of the edge weight(node ​​C, node D) = (num(CD) / num(C), where weight old (Node C, Node D) represents the weight of the edge from Node C to Node D in the previous iteration.

[0029] Furthermore, step S3 includes the following sub-steps:

[0030] Step S3.1, when an alarm occurs during the operation and maintenance of the telecommunication network, the alarm is normalized;

[0031] Step S3.2, traverse the normalized alarms, form all different alarms into two-tuples (alarm 1, alarm 2), and form a boundless alarm relationship graph;

[0032] Step S3.3, for each tuple (alarm 1, alarm 2) in the edgeless alarm relationship graph, find the distance between the device where alarm 1 occurs and the device where alarm 2 occurs in the device topology. If the distance is less than 2 hops, and there is an edge between alarm 1 and alarm 2 in the alarm type causal relationship graph, and alarm 1 occurs before alarm 2, add an edge from alarm 1 to alarm 2 in the edgeless alarm relationship graph, and set the weight of the edge from alarm 1 to alarm 2 to the weight of the corresponding edge in the alarm type causal relationship graph.

[0033] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the alarm root cause analysis method supporting continuous training.

[0034] Furthermore, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the alarm root cause analysis method supporting continuous training is implemented.

[0035] Furthermore, the present invention also provides a computer program product, including a computer program, which implements the alarm root cause analysis method supporting continuous training when executed by a processor.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) The alarm root cause analysis method supporting continuous training of the present invention constructs an alarm type causal relationship graph through a directed graph based on feedback information and the THP method of unsupervised learning, thereby reducing the workload. At the same time, the iteration of the alarm type causal relationship graph in the training process is controlled by feedback information, thereby ensuring that no causal edges opposite to the feedback information appear in the training results. In addition, through this continuous feedback and continuous training method, the alarm type causal relationship graph simultaneously includes the information reflected in the alarm and the expert experience, thereby improving the accuracy of the alarm type causal relationship graph analysis, thereby continuously improving the accuracy of the alarm root cause analysis results;

[0038] (2) When performing alarm root cause analysis, directed edges are added to the edgeless alarm relationship graph based on the alarm type causal relationship graph and the equipment topology in the telecommunications network to obtain a directed alarm relationship graph. The alarm type causal relationship graph reflects the causal situation reflected by historical alarms with expert experience, and the equipment topology reflects the possibility of alarm propagation at the physical level. The two are used together to form a directed alarm relationship graph, which can greatly improve the credibility and accuracy of the alarm root cause analysis results.

[0039] In summary, by constructing a causal relationship diagram of alarm types and improving the accuracy of alarm root cause analysis, fault propagation analysis of alarms can be performed, which can narrow the scope of fault search during telecommunication network operation and maintenance and quickly locate faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of the alarm root cause analysis method supporting continuous training of the present invention;

[0041] Figure 2 The present invention is a flowchart for constructing a cause-effect relationship diagram of alarm types. DETAILED DESCRIPTION

[0042] The technical solution of the present invention is further explained below in conjunction with the accompanying drawings.

[0043] like Figure 1 The present invention is a flow chart of the alarm root cause analysis method supporting continuous training, and the alarm root cause analysis method specifically includes the following steps:

[0044] Step S1, collect alarms generated during the operation and maintenance of the telecommunication network. Since different types of alarms contain different information, the alarms need to be normalized before the root cause analysis. The normalized alarms include: alarm type, alarm unique identifier, alarm occurrence location and alarm occurrence time.

[0045] Step S2, construct an alarm type causal relationship graph through a directed graph for the normalized alarms. The present invention supports a continuously trained alarm root cause analysis method that constructs an alarm type causal relationship graph through a directed graph based on feedback information and unsupervised learning THP method, thereby reducing workload. At the same time, the iteration of the alarm type causal relationship graph during the training process is controlled by feedback information, ensuring that no causal edges that are opposite to the feedback information will appear in the training results. In addition, through this continuous feedback and continuous training method, the alarm type causal relationship graph simultaneously includes the information reflected in the alarm and the expert experience, thereby improving the accuracy of the alarm type causal relationship graph analysis, thereby continuously improving the accuracy of the alarm root cause analysis results. Figure 2 , specifically including the following sub-steps:

[0046] Step S2.1, constructing an edgeless directed graph with the alarm categories in the normalized alarms as nodes;

[0047] Step S2.2: construct a directed candidate graph according to the edge adding rule for all nodes in the edgeless directed graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function, and use the directed candidate graph with the highest likelihood score as the initial alarm type causal relationship graph for training;

[0048] The margining rules involved in the present invention are:

[0049] i. Each node among all nodes forms a tuple with other nodes (node ​​A, node B);

[0050] ii. For each tuple (node ​​A, node B), if there is an edge from node A to node B in the alarm type causal relationship graph, but it is marked as an error in the feedback information, delete the edge from node A to node B; the feedback information refers to marking the edge in the alarm causal relationship graph as correct or wrong based on expert experience;

[0051] iii. If there is no edge from node A to node B in the alarm type causal relationship graph, and the edge from node A to node B is not marked incorrectly in the feedback information, and the edge from node B to node A in the feedback information is not marked correctly, add the edge from node A to node B in the alarm type causal relationship graph. If there is still an edge from node A to node B in the alarm type causal relationship, delete the edge from node A to node B.

[0052] The edge adding rule in the present invention controls the iteration of the alarm type causal relationship graph during the training process through feedback information, ensuring that causal edges that are opposite to the feedback information will not appear in the training results. Moreover, through this continuous feedback and continuous training method, the alarm type causal relationship graph simultaneously includes the information reflected in the alarm and the expert experience, thereby improving the accuracy of the alarm type causal relationship graph analysis, thereby continuously improving the accuracy of the alarm root cause analysis results.

[0053] The specific process of calculating the likelihood score of each directed candidate graph by the THP likelihood function in the present invention is:

[0054]

[0055] Among them, V table is the collection of all alarm types in the operation and maintenance process of the telecommunication network, T represents the time collection of alarm collection, N represents all device nodes in the operation and maintenance process of the telecommunication network, and X n,v,t is the number of times the vth type of alarm occurs on the nth device node within time t, PA v Indicates the alarm pointing to the vth type in the directed candidate graph. is the PA within two hops from the nth device node within time t v Number of occurrences.

[0056] Step S2.3, update the directed candidate graph according to the edge addition rule for all nodes in the alarm type causal relationship graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function;

[0057] Step S2.4: If the highest likelihood score exceeds the highest likelihood score of the previous iteration and the number of training iterations is less than the maximum number of iterations, the directed candidate graph with the highest likelihood score is used as the alarm type causal relationship graph, and step S2.3 is repeated until the number of training iterations reaches the maximum number of iterations or the highest likelihood score is less than or equal to the highest likelihood score of the previous iteration, and the optimal alarm type causal relationship graph in one round of iteration is obtained;

[0058] Step S2.5: In actual telecommunication network operation and maintenance, there are some alarm types that do not appear frequently. If the existing alarm analysis method is used, it is likely that these alarm types will not be included in the next round of training iterations because continuous training is not supported. At the same time, the most recent alarm data used for training can usually reflect the latest status of alarm analysis. Therefore, the present invention proposes continuous training, applies the weights of the previous round to the calculation of the edge weights of the next round, and uses conditional probability to calculate the weight of each edge in the optimal alarm type causal relationship graph, so that in the next round of iterations, the accuracy of alarm analysis can be improved, and the weights can be updated to reflect the latest status of alarm analysis; specifically:

[0059] A. For the two types of alarms on each edge of the optimal alarm type causal relationship graph, the number of occurrences of one alarm is counted as num(C), and within the alarm occurrence time, the number of occurrences of another alarm within two hops from the alarm generating device is counted as num(CD);

[0060] B. If the edge is marked as correct in the feedback information, the weight of the edge is weight(node ​​C, node D) = [num(CD) / num(C)+weightold (node ​​C, node D)] / 2; otherwise, the weight of the edge weight(node ​​C, node D) = (num(CD) / num(C), where weight old (Node C, Node D) represents the weight of the edge from Node C to Node D in the previous iteration.

[0061] Step S2.6: Repeat the process of steps S2.1-S2.6 in each iteration to obtain the current optimal alarm type causal relationship diagram.

[0062] Step S3: When an alarm occurs during the operation and maintenance of the telecommunications network, the alarm is normalized to construct a boundless alarm relationship graph. According to the alarm type causal relationship graph and the equipment topology in the telecommunications network, directed edges are added to the boundless alarm relationship graph to obtain a directed alarm relationship graph, wherein the alarm type causal relationship graph reflects the causal situation reflected by historical alarms with expert experience, and the equipment topology reflects the possibility of alarm propagation at the physical level. The two are used together to form a directed alarm relationship graph, which can greatly improve the credibility and accuracy of the alarm root cause analysis results. Specifically, it includes the following sub-steps:

[0063] Step S3.1, when an alarm occurs during the operation and maintenance of the telecommunication network, the alarm is normalized;

[0064] Step S3.2, traverse the normalized alarms, form all different alarms into two-tuples (alarm 1, alarm 2), and form a boundless alarm relationship graph;

[0065] Step S3.3, for each tuple (alarm 1, alarm 2) in the edgeless alarm relationship graph, find the distance between the device where alarm 1 occurs and the device where alarm 2 occurs in the device topology. If the distance is less than 2 hops, and there is an edge between alarm 1 and alarm 2 in the alarm type causal relationship graph, and alarm 1 occurs before alarm 2, add an edge from alarm 1 to alarm 2 in the edgeless alarm relationship graph, and set the weight of the edge from alarm 1 to alarm 2 to the weight of the corresponding edge in the alarm type causal relationship graph.

[0066] Step S4: Taking the alarm generated in step S3 as the starting point, use the DFS algorithm to reversely traverse the directed alarm relationship graph to obtain several alarm propagation paths, sum the weights of all edges in each propagation path, and take the three propagation paths with the largest sum of weights as the results of the alarm root cause analysis.

[0067] The alarm root cause analysis method supporting continuous training of the present invention improves the accuracy of alarm root cause analysis by constructing an alarm type causal relationship diagram, thereby performing fault propagation analysis on the alarm, which can narrow the fault search scope during the operation and maintenance of the telecommunications network, assist the operation and maintenance personnel to quickly locate the fault, thereby shortening the fault repair time and ensuring the stable operation of the telecommunications network.

[0068] In a technical solution of the present invention, a computer-readable storage medium is also provided, storing a computer program, wherein the computer program enables a computer to execute the alarm root cause analysis method supporting continuous training.

[0069] In a technical solution of the present invention, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the alarm root cause analysis method supporting continuous training is implemented.

[0070] In a technical solution of the present invention, a computer program product is also provided, including a computer program, and when the computer program is executed by a processor, the alarm root cause analysis method supporting continuous training is implemented.

[0071] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0072] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0073] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for root cause analysis of alarms supporting continuous training, characterized in that: The specific steps include: Step S1: Collect alarms generated during the operation and maintenance of the telecommunication network and perform normalization processing; Step S2: construct an alarm type causal relationship graph through a directed graph using the normalized alarms; Step S3: when an alarm occurs during the operation and maintenance of the telecommunication network, the alarm is normalized to construct a borderless alarm relationship graph, and directed edges are added to the borderless alarm relationship graph according to the alarm type causal relationship graph and the device topology in the telecommunication network to obtain a directed alarm relationship graph; Step S4: Taking the alarm generated in step S3 as the starting point, use the DFS algorithm to reversely traverse the directed alarm relationship graph to obtain several alarm propagation paths, sum the weights of all edges in each propagation path, and take the three propagation paths with the largest sum of weights as the results of the alarm root cause analysis.

2. The alarm root cause analysis method supporting continuous training according to claim 1, characterized in that: The normalized alarm includes: alarm type, alarm unique identifier, alarm occurrence location and alarm occurrence time.

3. The alarm root cause analysis method supporting continuous training according to claim 2, characterized in that: Step S2 includes the following sub-steps: Step S2.1, constructing an edgeless directed graph with the alarm categories in the normalized alarms as nodes; Step S2.2: construct a directed candidate graph according to the edge adding rule for all nodes in the edgeless directed graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function, and use the directed candidate graph with the highest likelihood score as the initial alarm type causal relationship graph for training; Step S2.3, update the directed candidate graph according to the edge addition rule for all nodes in the alarm type causal relationship graph, and calculate the likelihood score of each directed candidate graph according to the THP likelihood function; Step S2.4: If the highest likelihood score exceeds the highest likelihood score of the previous iteration and the number of training iterations is less than the maximum number of iterations, the directed candidate graph with the highest likelihood score is used as the alarm type causal relationship graph, and step S2.3 is repeated until the number of training iterations reaches the maximum number of iterations or the highest likelihood score is less than or equal to the highest likelihood score of the previous iteration, and the optimal alarm type causal relationship graph in one round of iteration is obtained; Step S2.5, using conditional probability to calculate the weight of each edge in the optimal alarm type causal relationship graph; Step S2.6: Repeat steps S2.1-S2.6 in each iteration until the maximum number of iterations is reached to obtain the current optimal alarm type causal relationship diagram.

4. The method for root cause analysis of alarms supporting continuous training according to claim 3, characterized in that: The margining rule satisfies: i. Each node among all nodes forms a tuple with other nodes (node ​​A, node B); ii. For each tuple (node ​​A, node B), if there is an edge from node A to node B in the alarm type causal relationship graph, but it is marked as an error in the feedback information, delete the edge from node A to node B; the feedback information refers to marking the edge in the alarm causal relationship graph as correct or wrong based on expert experience; iii. If there is no edge from node A to node B in the alarm type causal relationship graph, and the edge from node A to node B is not marked incorrectly in the feedback information, and the edge from node B to node A in the feedback information is not marked correctly, add the edge from node A to node B in the alarm type causal relationship graph. If there is still an edge from node A to node B in the alarm type causal relationship, delete the edge from node A to node B.

5. The alarm root cause analysis method supporting continuous training according to claim 3, characterized in that: The specific process of the THP likelihood function calculating the likelihood score of each directed candidate graph is: Among them, V table is the collection of all alarm types in the operation and maintenance process of the telecommunication network, T represents the time collection of alarm collection, N represents all device nodes in the operation and maintenance process of the telecommunication network, and X n,v,t is the number of times the vth type of alarm occurs on the nth device node within time t, PA v Indicates the alarm pointing to the vth type in the directed candidate graph. is the PA within two hops from the nth device node within time t v Number of occurrences.

6. The alarm root cause analysis method supporting continuous training according to claim 4, characterized in that: The specific process of step S2.5 is: A. For the two types of alarms on each edge of the optimal alarm type causal relationship graph, the number of occurrences of one alarm is counted as num(C), and within the alarm occurrence time, the number of occurrences of another alarm within two hops from the alarm generating device is counted as num(CD); B. If the edge is marked as correct in the feedback information, the weight of the edge is weight(node ​​C, node D) = [num(CD) / num(C)+weight old (node ​​C, node D)] / 2; otherwise, the weight of the edge weight(node ​​C, node D) = (num(CD) / num(C), where weight old (Node C, Node D) represents the weight of the edge from Node C to Node D in the previous iteration.

7. The alarm root cause analysis method supporting continuous training according to claim 6, characterized in that: Step S3 includes the following sub-steps: Step S3.1, when an alarm occurs during the operation and maintenance of the telecommunication network, the alarm is normalized; Step S3.2, traverse the normalized alarms, form all different alarms into two-tuples (alarm 1, alarm 2), and form a boundless alarm relationship graph; Step S3.3, for each tuple (alarm 1, alarm 2) in the edgeless alarm relationship graph, find the distance between the device where alarm 1 occurs and the device where alarm 2 occurs in the device topology. If the distance is less than 2 hops, and there is an edge between alarm 1 and alarm 2 in the alarm type causal relationship graph, and alarm 1 occurs before alarm 2, add an edge from alarm 1 to alarm 2 in the edgeless alarm relationship graph, and set the weight of the edge from alarm 1 to alarm 2 to the weight of the corresponding edge in the alarm type causal relationship graph.

8. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables a computer to execute the alarm root cause analysis method supporting continuous training as described in any one of claims 1-7.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the alarm root cause analysis method supporting continuous training as described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the alarm root cause analysis method supporting continuous training described in any one of claims 1 to 7 is implemented.

Images