WO2022021420A1

WO2022021420A1 - Root cause localization method and apparatus, and electronic device

Info

Publication number: WO2022021420A1
Application number: PCT/CN2020/106404
Authority: WO
Inventors: 黄晓光; 张伟; 岳晓贫; 兰宇
Original assignee: 华为技术有限公司
Priority date: 2020-07-31
Filing date: 2020-07-31
Publication date: 2022-02-03

Abstract

Embodiments of the present application provide a root cause localization method and a device, relating to the technical field of fault localization, and capable of improving the accuracy of root cause localization. The specific solution comprises: acquiring a tree network for a first performance indicator; acquiring a sampling value of the first performance indicator and sampling values of multiple second performance indicators obtained by means of sampling in a preset localization cycle; calculating, according to the sampling value of the first performance indicator and the sampling values of the multiple second performance indicators, influence factors of respective child nodes in relation to a parent node in the tree network; and determining, according to the influence factors of the respective child nodes in relation to the parent node in the tree network, a second performance indicator that has caused an abnormality in the first performance indicator in the tree network. The tree network comprises the first performance indicator and the multiple second performance indicators that influence the first performance indicator. The first performance indicator corresponds to a root node of the tree network, and the multiple second performance indicators correspond to nodes other than the root node in the tree network.

Description

A root cause locating method and device, and electronic equipment

technical field

The present application relates to the technical field of fault location, and in particular, to a root cause location method and device, and electronic equipment.

Background technique

In the operation and maintenance of the wireless communication system, monitoring the performance of the wireless communication system is an important part, which is mainly used for the location of performance faults and the root cause of performance faults in the wireless communication system. In this regard, the operation and maintenance personnel have set many indicators to evaluate the performance of the wireless communication system, and to locate the performance fault and the root cause of the performance fault of the wireless communication system.

Among them, these indicators may include key performance indicators (Key Performance Indicator, KPI) in terms of access, retention, and mobility, as well as service experience indicators such as rate and delay. The performance fault location is to locate the abnormal value of the above indicators. Root cause location of performance faults refers to analyzing the causes of abnormal values of the above indicators. After locating the performance fault and the root cause of the performance fault, the operation and maintenance personnel can eliminate the performance fault to ensure the user's wireless communication experience.

At present, in some solutions, the fault tree and the specific values of the above indicators can be used to locate performance faults, and the indicators with abnormal values (abbreviated as abnormal indicators) can be determined; Whether the specific value of the index is greater than the threshold value corresponding to the index, if the specific value of the index corresponding to a child node is greater than the threshold value corresponding to the index, it can be determined that the above-mentioned performance failure root cause is that the index corresponding to the child node is abnormal.

However, the performance fault root cause location is performed by comparing the threshold value, which may affect the accuracy of the performance fault root cause location due to the inaccurate setting of the threshold value. In addition, the indicators of each sub-node may have different degrees of influence on the abnormal indicators; it is relatively arbitrary to determine whether the indicators of each sub-node have an impact on the abnormal indicators by simply using the threshold judgment method, which will affect the accuracy of performance fault root cause location.

SUMMARY OF THE INVENTION

The present application provides a root cause locating method, device, and electronic device, which can improve the accuracy of root cause locating.

In a first aspect, the present application provides a root cause locating method, the root cause locating method includes: acquiring a tree-shaped network of a first performance index; The sampling value of the second performance index; according to the sampling value of the first performance index and the sampling values of a plurality of second performance indicators, calculate the influence factor of each child node in the tree network on the parent node; according to each child node in the tree network For the influence factor of the parent node, the second performance index in the tree network that causes the abnormality of the first performance index is determined.

The tree-shaped network includes a first performance indicator and a plurality of second performance indicators that affect the first performance indicator, the first performance indicator corresponds to the root node of the tree-shaped network, and the plurality of second performance indicators corresponds to one of the root nodes in the tree-shaped network. other nodes. The influence factor of each child node on the parent node is used to characterize the degree of influence of the performance index of each child node on the performance index of the parent node.

In the above method, the influence factor of each child node in the tree network on the parent node can be calculated according to the sampled value of the first performance index and the sampled value of a plurality of second performance indicators. The influence factor of a child node on the parent node is used to represent the influence degree of the performance index of the child node on the performance index of the parent node. Wherein, since the above-mentioned first performance index corresponds to the root node of the tree-shaped network, the above-mentioned multiple second performance indicators correspond to other nodes in the tree-shaped network except the root node; therefore, it can be concluded that the node corresponding to the above-mentioned second performance index can be is: a child node of the root node corresponding to the first performance index or a child node of the child node of the root node corresponding to the first performance index, and the like. It can be seen that the influence factor of each child node in the tree network on the parent node can reflect the degree of influence of the above-mentioned multiple second performance indicators on the first performance indicator. Therefore, according to the influence factor of each child node in the above-mentioned tree network on the parent node, the second performance index in the tree network that causes the abnormality of the first performance index can be determined.

The method provided by the present application does not require manual participation, and can determine, through calculation, the sampling value of the first performance index obtained by periodic sampling and the sampling values of a plurality of second performance indicators, which causes the first performance index in the tree network. Abnormal secondary performance indicator. In this way, compared with manual participation in root cause localization, the efficiency and accuracy of root cause localization can be improved, and the cost can also be reduced.

In a possible design manner, the preset positioning period includes Q sampling moments, where Q≥2, and Q is an integer. The above-mentioned acquisition of the sampling value of the first performance index and the sampling values of a plurality of second performance indicators obtained by sampling in the preset positioning period includes: at the qth sampling time of the preset positioning period, sampling to obtain the first performance index of the first performance index. The q sampled values and the qth sampled value of the plurality of second performance indicators. Among them, q takes values in sequence in {1,...,Q}.

In this design mode, the execution frequency of performance fault localization can be controlled by presetting the size of the localization period, and the number of first performance indicators and the number of multiple second performance indicators used for performance fault localization can be controlled.

In another possible design method, calculating the influence factor of each child node on the parent node in the tree network according to the sampled value of the first performance index and the sampled value of a plurality of second performance indicators includes: first, for the tree Each parent node in the shape network executes: the qth sampled value of the performance index of all child nodes of the first parent node is used as the input of the impact factor estimation model of the first parent node, and the impact of the first parent node is run. Factor estimation model, output the qth influence factor of each child node of the first parent node to the first parent node; then, for each child node of the first parent node: calculate the effect of the first child node on the first parent node. The average value of the Q influence factors is used to obtain the influence factor of the first child node on the first parent node; wherein, the first child node is any child node of the first parent node. Among them, the first parent node is any parent node in the tree network; the impact factor estimation models of different parent nodes are different; the impact factor estimation model of the first parent node has the performance of all child nodes from the first parent node From the sampling value of the index, the ability to extract the influence factor of each child node of the first parent node on the first parent node.

In this design method, for each parent node in the tree network, q takes values in {1, . value and the qth sampling value of each of the multiple second performance indicators, calculate the qth impact factor of each child node of each parent node on the parent node, that is, use the sampling value of the performance indicator collected at any sampling moment, Calculate the influence factor of the child node on the parent node at this adoption moment. Since the values of the performance indicators (including the first performance indicator and a plurality of second performance indicators) collected at different sampling times are different, the degree of influence between the performance indicators determined by the performance indicators with different values may also be different. Therefore, the influence factor of each sampling time is calculated by using the sampling values collected at each adopting time, and then the influence factors of all sampling time are averaged to determine the unique influence factor of each child node on the parent node. In this way, the method of calculating the influence factors of each sampling time separately and then averaging, compared with using the sampling values collected at all sampling times to calculate the unique influence factor of each child node on the parent node, the accuracy of the obtained influence factor is higher. .

Secondly, using the impact factor prediction model, the qth impact factor of all child nodes of each parent node on the parent node can be obtained at one time, thereby improving the efficiency of root cause location.

In another possible design method, calculating the influence factor of each child node on the parent node in the tree network according to the sampled value of the first performance index and the sampled value of a plurality of second performance indicators includes: first, for the tree Each parent node in the shape network executes: the qth sampled value of the performance index of all child nodes of the first parent node is used as the input of the impact factor estimation model of the first parent node, and the impact of the first parent node is run. The factor estimation model outputs the qth impact factor of each child node of the first parent node on the first parent node; then, if the pth sampling time in the Q sampling moments is the mutation moment of the performance index, the first Each child node of the parent node executes: calculate the average value of the first p-1 impact factors of the first child node to the first parent node, and obtain the impact factor of the first child node to the first parent node before mutation; calculate the first child node The average value of the p-th -Q-th impact factor of the node on the first parent node, and the impact factor of the first child node on the first parent node after the mutation is obtained; finally, the first child node before the mutation is calculated. The difference or ratio of the influence factor of , and the influence factor of the first child node to the first parent node after mutation, and the difference or ratio is used as the influence factor of the first child node to the first parent node.

Wherein, the first parent node is any parent node in the tree network; the impact factor estimation models of different parent nodes are different. The impact factor estimation model of the first parent node has the ability to extract the impact factor of each child node of the first parent node on the first parent node from the sampled values of performance indicators of all child nodes of the first parent node. p is a positive integer less than or equal to Q; the first child node is any child node of the first parent node.

It can be understood that the abnormality of the first performance index may be a sudden result. For the sudden abnormality of the first performance index, there is a large mutation in the sampling value of the first performance index and the sampling values of multiple second performance indicators, then the multiple second performance indicators before and after the mutation have a significant impact on the first performance The degree of influence of indicators may vary greatly. That is to say, there is a big difference between the influence factor of the first child node before the mutation on the first parent node and the influence factor of the first child node after the mutation on the first parent node. Therefore, in this design method, all the influencing factors of the first child node before and after mutation to the first parent node are averaged respectively.

Further, in this design method, the difference or ratio of the influence factor of the first child node on the first parent node before and after the mutation is calculated, and the difference or the ratio is used as the influence factor of the first child node on the first parent node. Since the difference or the ratio represents the change of the influence factor of the first child node on the first parent node before and after the mutation, and the greater the change of the influence factor of any child node on the parent node before and after the mutation, the The child node has a greater impact on causing the abnormality of the first performance index; therefore, the difference or the ratio is used for locating the root cause of the abnormality of the first performance index, which can improve the accuracy of locating the root cause of the abnormality of the first performance index.

In another possible design manner, before calculating the influence factor of each child node in the tree network on the parent node according to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, the method further includes: : Use the h-th sample value of the performance index of the first parent node and the h-th sample value of the performance index of all child nodes of the first parent node as training samples, train the preset attention model, and obtain the Impact factor prediction model. Among them, h takes values in sequence in {1,...,H}, H≥100, and H is an integer.

The above design method provides an implementation method for obtaining the influence factor prediction model of the first parent node.

In another possible design method, calculating the influence factor of each child node in the tree network on the parent node according to the sampled value of the first performance index and the sampled values of a plurality of second performance indicators, including: for the tree network Perform the following steps for each child node in to obtain the influence factor of each child node on the parent node: according to the Q sampling values of the performance index of the second child node, calculate the first residual of the performance index of the second child node; For the Q sampled values of the performance index of the parent node of the second child node, calculate the second residual of the performance index of the parent node of the second child node; calculate the correlation between the first residual and the second residual, and According to the correlation between the first residual and the second residual, the influence factor of the second child node on the parent node is obtained.

Wherein, the second child node is any child node of the tree network; the first residual is used to represent the fluctuation of the Q sample values of the performance index of the second child node. The second residual is used to represent the fluctuation of the Q sampled values of the performance index of the parent node of the second child node.

In the above design method, the first residual of the performance index of the second child node and the second residual of the performance index of the parent node of the second child node are calculated first; The correlation between the first residual and the second residual is determined, and the influence factor of the second child node on the parent node is determined according to the correlation between the first residual and the second residual. Since the residual represents the fluctuation of the Q sampling values of the performance index of any node, if the residual of the performance index of any node has a large correlation with the residual of the performance index of the parent node of any node, then The fluctuation of the Q sampling values representing the performance index of any node has a greater impact on the fluctuation of the Q sampling values of the performance index of the parent node of any node. However, if the fluctuation of the Q sampling values of the performance index of any node has a greater impact on the fluctuation of the Q sampling values of the performance index of the parent node of any node, it can be determined that the performance index of any node affects the parent node. The greater the impact of the node's sex index. Therefore, in the above design method, the influence factor of the second child node on the parent node is determined according to the correlation between the first residual and the second residual, and then the influence factor is used to determine the second performance that causes the abnormality of the first performance index The indicator refers to the degree of influence on the fluctuation of the Q sampling values of the performance index of the second child node on the fluctuation of the Q sampling values of the performance index of the second child node to determine the abnormality of the first performance index. The second performance index can improve the accuracy of locating the root cause of the abnormality of the first performance index.

In another possible design manner, calculating the first residual of the performance index of the second child node according to the Q sampled values of the performance index of the second child node includes: using a time series decomposition algorithm to calculate the first residual from the second child node From the Q sampling values of the performance index of the node, extract the trend component of the performance index of the second child node and the periodic component of the performance index of the second child node; subtract the Q sampling values of the performance index of the second child node from the The trend component and the period component are used to obtain the first residual of the performance index of the second child node. The trend component is used to represent the change trend of the Q sampled values of the performance index of the second child node, and the periodic component is used to represent the periodic change of the Q sampled values of the performance index of the second child node.

The above design method provides an implementation manner of obtaining the first residual of the performance index of the second child node, which can also be used to obtain the second residual of the performance index of the parent node of the second child node.

In another possible design method, determining the second performance index in the tree network that causes the first performance index to be abnormal according to the influence factors of each child node in the tree network on the parent node, including: The influence factor of each child node on the parent node is calculated, and the influence factor of the leaf node in the tree network to the root node is calculated; the performance indicators of N leaf nodes in the tree network are determined as the second performance that causes the abnormality of the first performance indicator index. Among them, the influence factor of N leaf nodes on the root node is greater than the influence factor of other leaf nodes in the tree network on the root node, N≥1, and N is an integer.

It can be understood that since the leaf nodes are terminal nodes in the tree network, the root cause of the abnormality of the first performance index exists in all the leaf nodes in the tree network. Therefore, in the above design method, the influence factor of all leaf nodes in the tree network on the root node is directly calculated; and then the performance index of the leaf node with the larger influence factor on the root node is determined as the second factor that causes the abnormality of the first performance index. Performance.

In another possible design method, the tree network includes a first preset node, the first preset node is not a leaf node, and the influence factor of the child nodes of the first preset node on the first preset node is less than or equal to the preset node. Set the threshold. The above-mentioned determining, according to the influence factor of each child node in the tree network on the parent node, determines the second performance index in the tree network that causes the first performance index to be abnormal, including: according to the influence factor of each child node in the tree network on the parent node , calculate the influence factor of the leaf node on the root node in the tree network, and calculate the influence factor of the first preset node on the root node; determine the performance indicators of the leaf node and the M nodes in the first preset node as causing The second performance index in which the first performance index is abnormal. Among them, the influence factor of M nodes on the root node is greater than the influence factor of other nodes in the tree network on the root node, M≥1, and M is an integer.

It can be understood that although the first preset node in the tree network is not a leaf node, the influence factor of the child nodes of the first preset node on the first preset node does not exceed the preset threshold, that is, the first preset node. The performance index of the child node of the first preset node has little influence on the performance index of the first preset node, it can be considered that the first preset node is similar to the leaf node, and the root cause of the abnormal first performance index may also exist in the tree network. in the first preset node in . Therefore, in the above design method, the influence factor of all leaf nodes in the tree network on the root node and the influence factor of the first preset node in the tree network on the root node are calculated; Among the nodes and the first preset node, a node with a larger influence factor on the root node is determined, and its performance index is the second performance index that causes the first performance index to be abnormal. In this way, both the leaf node that may include the root cause that causes the abnormality of the first performance index and the first preset node that may include the root cause that causes the abnormality of the first performance index are judged, and the first performance index that causes the abnormality can be determined more accurately. Abnormal secondary performance indicator.

In a second aspect, the present application provides a root cause locating device, the device comprising: an acquisition module configured to acquire a tree-shaped network of a first performance index, and obtain a sampling value of the first performance index obtained by sampling within a preset positioning period and sampling values of a plurality of second performance indicators; a calculation module for calculating the influence factor of each child node in the tree network on the parent node according to the sampling values of the first performance index and the sampling values of a plurality of second performance indicators; positioning The module is configured to determine the second performance index in the tree network that causes the abnormality of the first performance index according to the influence factors of each child node in the tree network on the parent node.

In a possible design manner, the preset positioning period includes Q sampling moments, where Q≥2, and Q is an integer. The acquisition module is specifically used to obtain the qth sampling value of the first performance index and the qth sampling value of the plurality of second performance indicators by sampling at the qth sampling time of the preset positioning cycle; wherein, q is in {1 , ..., Q} and take values in turn.

In another possible design manner, the calculation module is specifically used to: perform for each parent node in the tree network: take the qth sampled value of the performance index of all child nodes of the first parent node as the first The input of the influence factor estimation model of the parent node, run the influence factor estimation model of the first parent node, and output the qth influence factor of each child node of the first parent node on the first parent node; Execute for each child node: calculate the average value of Q influence factors of the first child node to the first parent node, and obtain the influence factor of the first child node to the first parent node. Wherein, the first parent node is any parent node in the tree network; the impact factor estimation models of different parent nodes are different. The impact factor estimation model of the first parent node has the ability to extract the impact factor of each child node of the first parent node on the first parent node from the sampled values of performance indicators of all child nodes of the first parent node. The first child node is any child node of the first parent node.

In another possible design method, the calculation module is specifically used for: first, for each parent node in the tree network, execute: take the qth sampled value of the performance index of all the child nodes of the first parent node as Input the influence factor estimation model of the first parent node, run the influence factor estimation model of the first parent node, and output the qth influence factor of each child node of the first parent node on the first parent node; then, if Q The p-th sampling time in the sampling time is the mutation time of the performance index, which is executed for each child node of the first parent node: calculate the average value of the first p-1 impact factors of the first child node to the first parent node, Obtain the influence factor of the first child node on the first parent node before the mutation; The influence factor of the parent node; finally, calculate the difference or ratio of the influence factor of the first child node to the first parent node before the mutation and the influence factor of the first child node to the first parent node after the mutation, and use the difference or ratio as The influence factor of the first child node on the first parent node.

In another possible design manner, the device further includes: a model training module, configured to calculate, according to the sampled value of the first performance index and the sampled value of a plurality of second performance indicators, the relationship between each child node in the tree network to the parent Before the impact factor of the node, the h-th sample value of the performance index of the first parent node and the h-th sample value of the performance index of all child nodes of the first parent node are used as training samples, and the preset attention model is trained to obtain The impact factor prediction model of the first parent node. Among them, h takes values in sequence in {1,...,H}, H≥100, and H is an integer.

In another possible design manner, the calculation module is specifically configured to perform the following steps for each child node in the tree network, so as to obtain the influence factor of each child node on the parent node: Q according to the performance index of the second child node calculate the first residual of the performance index of the second child node; according to the Q sampling values of the performance index of the parent node of the second child node, calculate the second residual of the performance index of the parent node of the second child node difference; calculate the correlation between the first residual and the second residual, and obtain the influence factor of the second child node on the parent node according to the correlation between the first residual and the second residual. Wherein, the second child node is any child node of the tree network. The first residual is used to represent the fluctuation of the Q sampled values of the performance index of the second child node. The second residual is used to represent the fluctuation of the Q sampled values of the performance index of the parent node of the second child node.

In another possible design method, the calculation module is specifically used for: using a time series decomposition algorithm to extract the trend component of the performance index of the second child node and the first sub-node from the Q sampling values of the performance index of the second child node. The periodic component of the performance index of the second child node; the Q sampled values of the performance index of the second child node are subtracted from the trend component and the periodic component to obtain the first residual of the performance index of the second child node. The trend component is used to represent the change trend of the Q sampled values of the performance index of the second child node, and the periodic component is used to represent the periodic change of the Q sampled values of the performance index of the second child node.

In another possible design method, the positioning module is specifically used to: calculate the influence factor of the leaf node on the root node in the tree network according to the influence factor of each child node in the tree network on the parent node;

The performance indicators of the N leaf nodes in the tree network are determined as the second performance indicators that cause the first performance indicator to be abnormal. Among them, the influence factor of N leaf nodes on the root node is greater than the influence factor of other leaf nodes in the tree network on the root node, N≥1, and N is an integer.

In another possible design method, the tree network includes a first preset node, the first preset node is not a leaf node, and the influence factor of the child nodes of the first preset node on the first preset node is less than or equal to the preset node. Set the threshold. The positioning module is specifically used for: calculating the influence factor of the leaf node on the root node in the tree network according to the influence factor of each child node in the tree network on the parent node, and calculating the influence factor of the first preset node on the root node; The performance indicators of M nodes in the leaf node and the first preset node are determined as the second performance indicators that cause the first performance indicator to be abnormal. Among them, the influence factor of M nodes on the root node is greater than the influence factor of other nodes in the tree network on the root node, M≥1, and M is an integer.

In a third aspect, the present application provides an electronic device, the electronic device comprising: a processor and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions, such that the The electronic device executes the root cause locating method described in the first aspect and any possible design manners thereof.

In a fourth aspect, the present application provides a computer-readable storage medium, where computer instructions are stored on the computer-readable storage medium, and when the computer instructions are executed on an electronic device, the electronic device is made to perform the first aspect and any of the methods described above. A possible design approach described in the root cause localization method.

In a fifth aspect, the present application provides a computer program product that, when the computer program product is run on an electronic device, causes the electronic device to execute the root cause described in the first aspect and any possible design manner thereof positioning method.

For the technical effects brought by the second aspect of the present application and any possible design methods thereof, as well as the third aspect, the fourth aspect, and the fifth aspect, reference may be made to the technologies brought about by different design methods in the above-mentioned first aspect The effect will not be repeated here.

Description of drawings

FIG. 1 is a flowchart 1 of a root cause locating method provided by an embodiment of the present application;

FIG. 2 is a schematic structural diagram of a tree network of a wireless communication system according to an embodiment of the present application;

FIG. 3 is a flow chart 2 of a root cause locating method provided by an embodiment of the present application;

FIG. 4 is a flowchart 3 of a root cause locating method provided by an embodiment of the present application;

5 is a schematic structural diagram of an impact factor prediction model of a parent node provided by an embodiment of the present application;

6 is a flowchart four of a root cause locating method provided by an embodiment of the present application;

7 is a flowchart 5 of a root cause locating method provided by an embodiment of the present application;

8 is a flowchart of a method for calculating the influence factor of a child node on a parent node provided by an embodiment of the present application;

9 is a flowchart 6 of a root cause locating method provided by an embodiment of the present application;

FIG. 10 is a schematic structural diagram of a tree network of a first performance index provided by an embodiment of the present application;

FIG. 11 is a flowchart seven of a root cause locating method provided by an embodiment of the present application;

12 is a flowchart of a method for determining a second performance indicator that causes an abnormality of the first performance indicator according to an embodiment of the present application;

13 is a schematic structural diagram of a root cause locating device according to an embodiment of the present application;

FIG. 14 is a schematic structural diagram of an electronic device according to an embodiment of the present application.

detailed description

The "first" and "second" etc. described in the embodiments of the present application are used to distinguish different objects, or used to distinguish different processing of the same object, rather than being used to describe a specific order of the objects. For example, the first performance indicator and the second performance indicator are different performance indicators.

In order to ensure the normal operation of any kind of equipment or system, the operation and maintenance (ie, operation and maintenance) of the equipment or the system is very important. The operation and maintenance of the device or the system mainly includes fault location and root cause location of the device or system. Among them, fault location refers to determining the fault of the device or system; fault root cause location refers to determining the cause of the failure of the device or system.

At present, the failures of the equipment or the system are mainly divided into two types, namely performance failures and hardware failures. The performance failure refers to the abnormal value of some performance indicators of the device or system; the performance indicators may include the above KPI indicators and the above service experience indicators. Hardware failure refers to an abnormality in the hardware of the device or system; hardware failure may be caused by hardware damage or incorrect operation.

In the embodiment of the present application, the root cause of the fault is mainly located for the performance fault of the device or the system, that is, the cause of the abnormal value of the performance index is determined. The performance index of the numerical anomaly is referred to as the abnormal performance index for short. A hardware failure can also lead to a performance failure; therefore, the cause of the abnormal value of the abnormal performance indicator may be other performance indicators other than the abnormal performance indicator of the device or the system, or a hardware failure.

Specifically, for the performance fault of the device or the system, some performance indicators of the device or the system can be monitored, and according to the values of the performance indicators, the performance fault location and the root cause of the performance fault can be located for the device or the system. After locating the performance fault and the root cause of the performance fault, the operation and maintenance personnel can eliminate the performance fault to ensure the user experience of the device or the system. Wherein, the performance fault location is to specify a performance index (ie, an abnormal performance index) with an abnormal value among all the performance indexes of the device or system. Root cause location of performance faults refers to analyzing the causes of numerical anomalies of abnormal performance indicators.

In the related scheme, performance fault location and root cause location of performance faults are performed manually, and abnormal performance indicators and performance indicators that cause numerical abnormality of abnormal performance indicators are determined. Due to the limited manual speed, there is a problem that the efficiency of root cause localization is low. In addition, some devices or systems have a huge number of performance indicators, and it is easy to make mistakes when manually locating the root cause, which reduces the accuracy of root cause locating.

For example, in the operation and maintenance of a wireless communication system, there are a large number of performance indicators through the wireless communication system, and the number of abnormal performance indicators is also large. Taking the wireless communication system of a certain operator as an example, the number of abnormal performance indicators per day accounts for about 3% of the total number of network elements in the wireless communication system. The number of performance metrics analyzed is greater.

Further, the related scheme uses the fault tree of abnormal performance indicators to locate the root cause of abnormal performance indicators. Among them, the fault tree is a directed acyclic graph (DAG), which consists of nodes representing performance indicators and directed edges connecting these nodes, and the directed edges between nodes represent the mutual relationship between nodes ( from the parent node to its child node). The root node of the fault tree corresponds to the abnormal performance index, and other nodes of the fault tree except the root node correspond to other performance indicators that have an impact on the abnormal performance index.

Specifically, it is respectively judged whether the value of the performance index of each child node in the fault tree is greater than the threshold value corresponding to the performance index. If the value of the performance index corresponding to a child node is greater than the threshold value corresponding to the performance index, then It can be determined that the performance index of the child node causes the numerical value of the abnormal performance index to be abnormal.

However, using the method of comparing the threshold value for root cause localization may affect the accuracy of root cause localization due to the inaccurate setting of the threshold value.

On the one hand, the performance indicators of each sub-node may have different degrees of influence on the abnormal performance indicators, and it is rather arbitrary to determine whether the performance indicators of each sub-node have an impact on the abnormal performance indicators by simply using the threshold value judgment method, which may Some performance indicators whose impact on abnormal performance indicators is less than the corresponding threshold value are filtered out, and some performance indicators that have no effect on abnormal performance indicators may also be determined as the root cause of abnormal performance indicators. In this way, the accuracy of root cause location will be affected.

On the other hand, since the distance between each child node and the root node is different; therefore, different thresholds need to be set for different child nodes to judge whether the performance index of the child node has an impact on the abnormal performance index. The farther the child node is from the root node, the less obvious the influence relationship between the child node and the root node is, and the more difficult it is to determine the threshold value corresponding to the child node. Moreover, if the threshold value corresponding to the child node is not set accurately, the accuracy of root cause location will be affected.

Further, it is difficult to determine the threshold value of the performance index of each node on the fault tree, and setting the threshold value of the performance index of all the nodes in the fault tree also requires a lot of manpower and cost.

Embodiments of the present application provide a root cause locating method, device, and electronic device, and the root cause locating method can improve the accuracy of root cause locating. The root cause locating method can be applied to any scenario that requires root cause locating, for example, a wireless communication system, or other scenarios where the influence relationship between performance indicators can be determined.

The implementation of the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

Referring to FIG. 1 , a root cause location method provided by this embodiment of the present application may include S101-S104.

S101. The electronic device obtains a tree-shaped network of a first performance index; wherein, the tree-shaped network includes a first performance index and a plurality of second performance indicators that affect the first performance index, and the first performance index corresponds to a root node of the tree-shaped network, The plurality of second performance indicators correspond to other nodes in the tree network except the root node.

Wherein, the first performance index is a performance index that is abnormal and needs root cause localization. The tree network of the first performance indicator is a directed acyclic graph DAG, which consists of nodes representing performance indicators and directed edges connecting these nodes. The directed edges between nodes represent the influence relationship between nodes (directed by child nodes). its parent node). Any node other than the root node in the tree network corresponds to at least one second performance index, and the performance indexes corresponding to different nodes other than the root node in the tree network are different. A tree network can also be called a fault tree.

The first performance indicator may be any one of all performance indicators of the electronic device, any one of all performance indicators of a wireless communication network, or any one of all performance indicators of other electronic devices, and so on. The plurality of second performance indicators may include a plurality of performance indicators that may affect the first performance indicator. The plurality of second performance indicators are different from the first performance indicators.

For example, the first performance indicator is the user rate, and the multiple performance indicators that affect the first performance indicator may include: physical resource block (Physical Resource Block, PRB) utilization that directly affects the first performance indicator, PRB utilization Distribution ratio, and control channel element (control channel element, CCE) utilization rate, etc.; the path loss that indirectly affects the first performance index, etc.

In this embodiment of the present application, the electronic device may receive a tree network corresponding to the first performance index input by the operation and maintenance personnel. Alternatively, the electronic device may receive a tree-shaped network corresponding to the first performance index from other electronic devices. Alternatively, the electronic device may receive an abnormal performance index that needs root cause localization, and use the performance index as the first performance index; The index constructs a tree network of the first performance index.

Wherein, the tree-shaped network may be constructed according to artificial experience, or the tree-shaped network of the obtained first performance index may be constructed according to the relationship between all performance indicators of the electronic device.

Exemplarily, as shown in FIG. 2, a tree-shaped network of a wireless communication system includes a root node and a plurality of child nodes. The first performance indicator corresponding to the root node is the user rate. The plurality of child nodes include a child node representing resource capacity, a child node representing air interface quality, a child node representing some other performance indicators, a child node representing weak coverage, and a child node representing signal interference situation. The performance indicators corresponding to child nodes representing resource capacity include physical resource block (Physical Resource Block, PRB) utilization, distribution ratio of PRB utilization, and control channel element (control channel element, CCE) utilization. The performance indicators corresponding to the sub-nodes representing the air interface quality include block error rate (block error rate, BLER), average scheme level of modulation and coding strategy (Modulation and Coding Scheme, MCS), and channel quality indicator (Channel Quality Indicator, CQI) . The performance indicators corresponding to the child nodes representing some other performance indicators include uplink interference power and handover success rate. The performance index corresponding to the sub-node representing the weak coverage includes the path loss and the average value of the distance between all the user equipments and the base station in the wireless communication system. The performance index corresponding to the child node representing the signal interference situation includes the uplink interference power.

S102. The electronic device acquires the sampled value of the first performance index and the sampled values of a plurality of second performance indicators that are sampled within a preset positioning period.

The electronic device may acquire the sampled value of the first performance index and the sampled values of a plurality of second performance indicators collected in a preset positioning period before the current time and adjacent to the current time.

The preset positioning period may be one day or seven days, and so on. The sampled value of the first performance indicator may include one or more sampled values; and the sampled value of each second performance indicator in the plurality of second performance indicators may include one or more sampled values.

In some embodiments, the electronic device may also collect sampling values of all performance indicators (including the first performance index and a plurality of second performance indicators) of the electronic device in real time according to a preset sampling period, and combine the collected sampling values and The corresponding sampling time is saved. The specific process of performing S102 by the electronic device may include: acquiring all sampled values of the first performance index and all sampled values of a plurality of second performance indicators at the sampling time within the above-mentioned preset positioning period.

The preset sampling period may be one hour or one day, and so on. The preset positioning period may include one or more preset sampling periods, and each preset sampling period corresponds to a sampling moment.

S103, the electronic device calculates the influence factor of each child node on the parent node in the tree network according to the sampled value of the first performance index and the sampled value of a plurality of second performance indexes; wherein, the influence factor of each child node on the parent node is calculated as It is used to characterize the degree of influence of the performance index of each child node on the performance index of the parent node.

The electronic device may perform for each child node in the tree network: using the sampled value of the performance index of the child node and the sampled value of the performance index of the parent node of the child node, calculate the influence factor of the child node on its parent node. The child node may have one or more parent nodes, and the child node may also have one or more influence factors on its parent node. The influence factor of the child node to its parent node is in a one-to-one correspondence with the parent node of the child node.

Alternatively, the electronic device may also perform for each parent node in the tree network: using the sampled value of the performance index of the parent node and the sampled value of the performance index of the child nodes of the parent node, calculate the pair of child nodes of the parent node. The influence factor of this parent node. The parent node may have one or more child nodes, and the parent node's child nodes may also have one or more influence factors on the parent node. The influence factors of the child nodes of the parent node to the parent node are in a one-to-one correspondence with the child nodes of the parent node.

Among them, the relationship strength of the influence relationship between the nodes, or the influence degree of the performance index between the nodes can be represented by the influence factor.

S104. The electronic device determines, according to the influence factors of each child node in the tree network on the parent node, the second performance index in the tree network that causes the first performance index to be abnormal.

According to the influence factor of each child node on the parent node, the electronic device can select a node with a greater impact on the root node; the performance index of the node with a greater impact on the root node is the first performance index that causes the abnormality of the first performance index. Two performance indicators. Furthermore, the electronic device may be repaired for the second performance index causing the abnormality of the first performance index, so as to solve the problem of the abnormality of the first performance index by adjusting the second performance index causing the abnormality of the first performance index.

It can be understood that, in the above method, the influence factor of each child node in the tree network on the parent node can be calculated according to the sampled value of the first performance index and the sampled value of a plurality of second performance indicators. The influence factor of a child node on the parent node is used to represent the influence degree of the performance index of the child node on the performance index of the parent node. Wherein, since the above-mentioned first performance index corresponds to the root node of the tree-shaped network, the above-mentioned multiple second performance indicators correspond to other nodes in the tree-shaped network except the root node; therefore, it can be concluded that the node corresponding to the above-mentioned second performance index can be is: a child node of the root node corresponding to the first performance index or a child node of the child node of the root node corresponding to the first performance index, and the like. It can be seen that the influence factor of each child node in the tree network on the parent node can reflect the degree of influence of the above-mentioned multiple second performance indicators on the first performance indicator. Therefore, according to the influence factor of each child node in the above-mentioned tree network on the parent node, the second performance index in the tree network that causes the abnormality of the first performance index can be determined.

Further, if more second performance indicators that affect the first performance indicator are obtained, as long as the impact paths of more second performance indicators that affect the first performance The tree-shaped network of performance indicators is expanded, so that the tree-shaped network of the first performance indicator is more refined, and then a more accurate second performance indicator that causes the abnormality of the first performance indicator can be determined. Since the electronic device can automatically calculate the influence factor of each child node in the tree network on the parent node, expanding the tree network of the first performance index will not increase the difficulty of root cause localization.

In this embodiment of the present application, the preset positioning period includes Q sampling moments, where Q≥2, and Q is an integer. The electronic device obtains the qth sampling value of the first performance index and the qth sampling value of the plurality of second performance indicators by sampling at the qth sampling time of the preset positioning period; Q} takes values in turn.

Wherein, the qth sampled value of the plurality of second performance indicators includes the qth sampled value of each second performance indicator. The sampled values of the first performance indicator sampled in the preset positioning period include Q sampled values of the first performance indicator; the sampled values of multiple second performance indicators sampled in the preset positioning period include the sampling values of each second performance indicator. Q sample values.

In this embodiment of the present application, the electronic device may first calculate the number of samples in the tree network at each sampling moment according to the Q sampled values of the first performance indicator and the Q sampled values of each of the second performance indicators in the plurality of first performance indicators. The influence factor of each child node on the parent node is calculated, and then the influence factor of each child node on the parent node in the tree network within the preset positioning period (ie the influence factor of each child node on the parent node in the tree network) is calculated.

In the embodiment of the present application, for each parent node in the tree network, the electronic device first calculates the influence factor of each child node on the parent node at each sampling moment, and then calculates the impact factor of each child node on the parent node within the preset positioning period The impact factor (that is, the impact factor of each child node on the parent node). Specifically, as shown in FIG. 3 , S103 may include S301-S302. To illustrate the process of calculating the influence factor of each parent node's child nodes on the parent node more clearly, each parent node is referred to as the first parent node.

S301, the electronic device performs for each parent node in the tree network: taking the qth sampled value of the performance index of all child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, and running The influence factor estimation model of the first parent node outputs the qth influence factor of each child node of the first parent node on the first parent node; wherein, q takes values in sequence in {1,...,Q}.

Wherein, the first parent node is any parent node in the tree network; the impact factor estimation models of different parent nodes are different. The impact factor prediction model of the first parent node has the ability to extract the impact factor of each child node of the first parent node on the first parent node from the sampling values of the performance indicators of all child nodes of the first parent node.

In the embodiment of the present application, the electronic device pre-stores the influence factor prediction model of each parent node in the tree network. The impact factor estimation model of the first parent node has the function of extracting the impact factor of each child node of the first parent node on the first parent node and the first parent node from the sampling values of the performance indicators of all child nodes of the first parent node. The ability to estimate the performance indicators of the parent node.

In the embodiment of the present application, after the electronic device sequentially takes values from {1, .

S302, the electronic device performs for each child node of the first parent node: calculate the average value of Q influence factors of the first child node on the first parent node, and obtain the influence factor of the first child node on the first parent node; wherein, The first child node is any child node of the first parent node.

It should be noted that, in order to explain the process of calculating the influence factor of each child node on the first parent node more clearly, each parent node when calculating the influence factor on the first parent node is referred to as the first parent node.

The electronic device divides the sum of the Q influence factors of the first child node to the first parent node by Q to obtain the average value of the Q influence factors of the first child node to the first parent node, and uses the average value as the first The influence factor of the child node on the first parent node. Furthermore, the electronic device can obtain the influence factors of each child node of the first parent node on the first parent node.

It can be understood that, due to the different values of the performance indicators (including the first performance indicator and multiple second performance indicators) collected at different sampling times, the degree of influence between the performance indicators determined by the performance indicators with different values may also be different. Therefore, the electronic device calculates the influence factor of each sampling time by using the sampling values collected at each adoption time, and then averages the influence factors of all the sampling time to determine the unique influence factor of each child node on the parent node. In this way, the method of calculating the influence factors of each sampling time separately and then averaging, compared with using the sampling values collected at all sampling times to calculate the unique influence factor of each child node on the parent node, the accuracy of the obtained influence factor is higher. .

In this embodiment of the present application, the abnormality of the first performance index may be a result of slow deterioration of the first performance index, or a sudden result.

Wherein, for the abnormality of the first performance indicator caused by the slow deterioration of the first performance indicator, the changes of the sampling values of the first performance indicator and the plurality of second performance indicators are both small, and the plurality of second performance indicators at different sampling times have little effect on the first performance indicator. The degree of influence of the performance index does not change much; therefore, the method of averaging the Q influence factors of the first child node on the first parent node in S302 can be used to calculate the influence of the first child node on the first parent node factor.

Wherein, for the sudden abnormality of the first performance index, there is a large mutation in the sampling values of the first performance index and the plurality of second performance indicators, then the plurality of second performance indicators before and after the mutation have a significant effect on the first performance index. The change of the influence degree of , that is to say, the influence factor of the first child node before the mutation on the first parent node and the influence factor of the first child node after the mutation on the first parent node are quite different. Moreover, since the abnormality of the first performance index is sudden, the greater the change of the influence factor of any child node on the parent node before and after the mutation, the greater the influence of any child node on the abnormality of the first performance index. . Therefore, for the sudden abnormality of the first performance index, after obtaining the influence factor of each child node on the parent node in the tree network at each sampling time, obtain the change of the influence factor of each child node on the parent node before and after the mutation , and calculate the influence factor of each child node on the parent node in the tree network according to the change.

In the embodiment of the present application, for the sudden first performance index abnormality, the electronic device calculates the specific process of the influence factor of the child node on the parent node for each parent node in the tree network, as shown in FIG. 4 , S103 can be Including S401-S402. To illustrate the process of calculating the influence factor of each parent node's child nodes on the parent node more clearly, each parent node is referred to as the first parent node.

S401, the electronic device performs for each parent node in the tree network: taking the qth sampling value of the performance index of all the child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, and running The influence factor estimation model of the first parent node outputs the qth influence factor of each child node of the first parent node on the first parent node; wherein, q takes values in sequence in {1,...,Q}.

It should be noted that, for the specific process of S401, reference may be made to the detailed description of the foregoing S301, which is not repeated in this embodiment of the present application.

S402. If the p-th sampling time in the Q sampling times is the sudden change of the performance index, the electronic device performs for each child node of the first parent node: calculate the first p-1 times of the first child node to the first parent node Calculate the average value of the impact factors to obtain the impact factor of the first child node on the first parent node before the mutation; The influence factor of a child node on the first parent node; calculate the difference or ratio between the influence factor of the first child node on the first parent node before the mutation and the influence factor of the first child node on the first parent node after the mutation. The value or ratio is used as the influence factor of the first child node to the first parent node.

Wherein, p is a positive integer less than or equal to Q; the first child node is any child node of the first parent node.

The difference value may be obtained by subtracting the influence factor of the first child node on the first parent node after the mutation from the influence factor of the first child node on the first parent node before the mutation, and the ratio may be the first child node after the mutation. It is obtained by dividing the impact factor on the first parent node by the impact factor on the first parent node of the first child node before mutation.

The electronic device determines that the abnormality of the first performance index is sudden when receiving the sudden change of the performance index as the pth sampling time. The electronic device divides the sum of the first p-1 influence factors of the first child node to the first parent node by the first difference, and the first difference is equal to p-1, to obtain the first child node to the first parent node before mutation Then divide the sum of the p-th - Q-th influencing factors of the first child node to the first parent node by the second difference, and the second difference is equal to Qp, to obtain the first child node after mutation to the first The influence factor of a parent node. Finally, the electronic device subtracts the difference between the influence factor of the first child node on the first parent node after the mutation and the influence factor of the first child node on the first parent node before the mutation, as the difference between the first child node and the first parent node. Influence factor; or, divide the influence factor of the first child node to the first parent node after the mutation by the ratio of the influence factor of the first child node to the first parent node before the mutation, as the ratio of the first child node to the first parent node Impact factor. Furthermore, the electronic device can obtain the influence factors of each child node of the first parent node on the first parent node.

It can be understood that the abnormality of the first performance index may be a sudden result. For the sudden abnormality of the first performance index, there is a large mutation in the sampling value of the first performance index and the sampling values of multiple second performance indicators, then the multiple second performance indicators before and after the mutation have a significant impact on the first performance The degree of influence of indicators may vary greatly. That is to say, there is a big difference between the influence factor of the first child node before the mutation on the first parent node and the influence factor of the first child node after the mutation on the first parent node. Therefore, the electronic device averages all the influence factors of the first child node on the first parent node before and after the mutation, respectively.

Further, the electronic device calculates the difference or ratio of the influence factor of the first child node to the first parent node before and after the mutation, and uses the difference or the ratio as the influence factor of the first child node to the first parent node. Since the difference or the ratio represents the change of the influence factor of the first child node on the first parent node before and after the mutation, and the greater the change of the influence factor of any child node on the parent node before and after the mutation, the The child node has a greater impact on causing the abnormality of the first performance index; therefore, the difference or the ratio is used for locating the root cause of the abnormality of the first performance index, which can improve the accuracy of locating the root cause of the abnormality of the first performance index.

In the embodiment of the present application, before S301-S302 and S401-S402, the electronic device adopts the h-th sample value of the performance index of the first parent node in the above tree network and the performance index of all child nodes of the first parent node. The h-th sampling value of is used as a training sample to train the preset attention model (Attention Model) to obtain the influence factor prediction model of the first parent node; where h takes values in sequence in {1,...,H}, H≥100, H is an integer.

Wherein, for the tree network of the first performance index, the electronic device can collect sample values of the first performance index and multiple sample values of the second performance index that the electronic device is applied to in various usage scenarios, so as to obtain the first performance index H sample values of , and H sample values of each second performance indicator in the plurality of second performance indicators. The H sample values of the first performance index and the H sample values of each second performance index are in one-to-one correspondence.

The electronic device may arrange the H sample values of the first performance index and the H sample values of each second performance index into the first sample value, the second sample, ... , and the H th sample, the H sample values of the first performance index and the H sample values of each second performance index may be respectively arranged according to other arrangement rules.

It should be noted that h takes values in sequence in {1,...,H}, and performs the process of training the preset attention model. The preset attention model used for training when h takes a value in {1, . It is assumed that the attention model, for example, the preset attention model used for training when the value of h is 2 is the preset attention model obtained by training when the value of h is 1. When the value of h is H, the preset attention model obtained by training is the influence factor prediction model of the first parent node.

In the embodiment of the present application, the electronic device uses the h-th sampled value of the performance indicators of all child nodes of the first parent node as the input of the preset attention model, runs the preset attention module, and outputs the performance of the first parent node. The h-th estimated value of the index; then calculate the absolute value of the difference between the h-th estimated value of the performance index of the first parent node and the h-th sample value of the performance index of the first parent node; then, The parameters of the preset attention model are adjusted according to the absolute value of the difference, so as to reduce the absolute value of the difference.

Exemplarily, using the preset attention model of the first parent node, the structure of the influence factor estimation model of the first parent node obtained by training is shown in Figure 5, assuming that all child nodes of the first parent node Y include child nodes. 1. Child node 2 and child node 3. Take the h-th sample value x1 of the performance index of child node 1, the h-th sample value x2 of the performance index of child node 2, and the h-th sample value x3 of the performance index of child node 3 as the input of the preset attention model . On the one hand, the h-th sample value of the performance index of each child node is input to the corresponding Multilayer Perceptron (MLP); then the output of the MLP is normalized by the corresponding normalization function (for example, the softmax function). Unification, the influence factor f1 of each child node on the first parent node Y is obtained. It includes: the influence factor f1(x1) of the child node x1 on the first parent node Y, the influence factor f1(x2) of the child node x2 on the first parent node Y, the influence factor f1 of the child node x3 on the first parent node Y (x3). On the other hand, the h-th sample value of the performance index of each child node is multiplied by the influence factor of the corresponding child node on the first parent node, and all the multiplied values obtained include x1*f1(x1), x2* f1(x2), x3*f1(x3); the multiplied values of all child nodes are input to an MLP, and the MLP outputs the h-th estimated value f2(X) of the performance index of the first parent node.

It should be noted that FIG. 5 is only a schematic diagram of the influence factor estimation model of a parent node, and the embodiment of the present application does not limit the influence factor estimation model of any parent node. In addition, since the number of child nodes of different parent nodes is different, the structure of the influence factor prediction model of different parent nodes is also different. Since the performance indicators of different parent nodes are different, and the performance indicators of the child nodes of different parent nodes are different, the parameters of the influence factor prediction models of different parent nodes are also different.

In the embodiment of the present application, the influence factor estimation model of the first parent node has the function of extracting the effect of each child node of the first parent node on the first parent node from the sampling values of the performance indicators of all child nodes of the first parent node. Influence factor, and the ability to estimate the performance index of the first parent node. Since the estimated value using the performance index of the first parent node is extracted from the sampling values of the performance index of all child nodes of the first parent node by using the impact factor prediction model of the first parent node, if the first parent node If there is a big difference between the estimated value of the performance index of the first parent node and the sample value of the performance index of the first parent node, it can be considered that there are factors other than the performance index of all the child nodes of the first parent node. The performance indicators of all child nodes of the parent node have a greater impact on the sample values of the performance indicators of the first parent node. Other factors may refer to hardware failures or performance metrics not included in the tree network. Therefore, in addition to determining the second performance index in the tree network that causes the first performance index to be abnormal, the electronic device may also determine other factors that cause the first performance index to be abnormal. Specifically, referring to FIG. 6 , the electronic device may execute S601-S604 without executing S103-S104 after S102. In order to illustrate the process of calculating the influence factor of each parent node's child nodes or other factors on the parent node more clearly, each parent node is referred to as the first parent node.

S601, the electronic device performs for each parent node in the tree network: taking the qth sampled value of the performance index of all the child nodes of the first parent node as the input of the influence factor estimation model of the first parent node, and running The impact factor estimation model of the first parent node outputs the qth impact factor of each child node of the first parent node on the first parent node and the qth estimated value of the performance index of the first parent node; The absolute value of the difference between the q-th estimated value of the performance indicator of a parent node and the q-th sampled value of the performance indicator of the first parent node, divided by the q-th sampled value of the performance indicator of the first parent node, Get the qth influencing factor of other factors of the first parent node on the first parent node.

Among them, q takes values in sequence in {1,...,Q}; other factors refer to factors other than the performance indicators of all child nodes of the first parent node.

It should be noted that, for the specific process of the same part in S601 as that of S301, reference may be made to the detailed description of the same part in S301, which is not repeated in this embodiment of the present application.

S602. The electronic device performs for each child node of the first parent node: calculating the average value of Q influence factors of the first child node on the first parent node to obtain the influence factor of the first child node on the first parent node.

The first child node is any child node of the first parent node.

It should be noted that, for the specific process of S602, reference may be made to the detailed description of S301, which is not repeated in this embodiment of the present application.

S603. The electronic device calculates the average value of Q influence factors of other factors of the first parent node on the first parent node, and obtains the influence factors of other factors of the first parent node on the first parent node.

S604. The electronic device determines, according to the influence factors of each child node in the tree network on the parent node and the influence factors of other factors of each parent node on the corresponding parent node, the first performance index in the tree network that causes the first performance index to be abnormal. The second performance index or other factors that cause the first performance index to be abnormal.

It should be noted that other factors of each parent node may be equal to the performance index of a new child node of the parent node. In this case, for the specific process of S604, please refer to the detailed description of S104, which is not repeated in this embodiment of the present application. .

It can be understood that, in addition to calculating the influence factor of each child node in the tree network on the parent node, the electronic device also calculates the influence factor of other factors of each parent node on the corresponding parent node. The problem of inaccuracy of the second performance index determined from the plurality of second performance indexes and causing the abnormality of the first performance index can be avoided because the obtained multiple second performance indexes are not comprehensive enough. It is also possible to avoid the problem of inaccuracy of the second performance index determined from a plurality of second performance indicators that causes the first performance index to be abnormal when the first performance index is abnormal due to a hardware failure. Therefore, determining the root cause of the abnormality of the first performance index from the plurality of second performance indicators and other factors is more accurate than determining the root cause of the abnormality of the first performance index from the plurality of second performance indicators.

In this embodiment of the present application, for a sudden first performance indicator abnormality, the electronic device may determine other than the second performance indicator in the tree network that causes the first performance indicator to be abnormal, and may also determine other performance indicators that cause the first performance indicator to be abnormal. factor. Specifically, referring to FIG. 7 , the electronic device may execute S701-S704 without executing S103-S104 after S102. In order to more clearly illustrate the process of calculating the influence factor of each parent node's child nodes or other factors on the parent node, each parent node is referred to as the first parent node.

S701, the electronic device performs for each parent node in the tree network: taking the qth sampled value of the performance index of all the child nodes of the first parent node as the input of the influence factor estimation model of the first parent node, and running The impact factor estimation model of the first parent node outputs the qth impact factor of each child node of the first parent node on the first parent node and the qth estimated value of the performance index of the first parent node; The absolute value of the difference between the q-th estimated value of the performance indicator of a parent node and the q-th sampled value of the performance indicator of the first parent node, divided by the q-th sampled value of the performance indicator of the first parent node, Get the qth influencing factor of other factors of the first parent node on the first parent node.

Among them, q takes values in sequence in {1,...,Q}.

It should be noted that, for the specific process of S701, reference may be made to the detailed description of the foregoing S601, which is not repeated in this embodiment of the present application.

S702. If the p-th sampling time in the Q sampling times is the sudden change of the performance index, the electronic device performs for each child node of the first parent node: calculate the first p-1 times of the first child node to the first parent node Calculate the average value of the impact factors to obtain the impact factor of the first child node on the first parent node before the mutation; The influence factor of a child node on the first parent node; calculate the difference or ratio between the influence factor of the first child node on the first parent node before the mutation and the influence factor of the first child node on the first parent node after the mutation. The value or ratio is used as the influence factor of the first child node to the first parent node.

It should be noted that, for the specific process of S702, reference may be made to the detailed description of S402, which is not repeated in this embodiment of the present application.

S703. The electronic device calculates the average value of the first p-1 influence factors of the other factors of the first parent node on the first parent node, and obtains the influence factors of other factors of the first parent node before the mutation on the first parent node; calculate the first parent node. The average value of the p-th to Q-th influencing factors of the other factors of a parent node on the first parent node, to obtain the influence factors of other factors of the first parent node after the mutation on the first parent node; calculate the first parent node before the mutation The difference or ratio between the influence factors of other factors of the node on the first parent node and the influence factors of other factors of the first parent node on the first parent node after mutation, and the difference or ratio is taken as the pair of other factors of the first parent node. The influence factor of the first parent node.

It should be noted that, for the specific process of S703, reference may be made to the detailed description of S402, which is not repeated in this embodiment of the present application.

S704. The electronic device determines, according to the influence factor of each child node in the tree network on the parent node, and the influence factor of other factors of each parent node on the corresponding parent node, the first performance index in the tree network that causes the abnormality of the first performance index. The second performance index or other factors that cause the first performance index to be abnormal.

It should be noted that other factors of each parent node may be equivalent to the performance index of a new child node of the parent node. In this case, for the specific process of S704, reference may be made to the detailed description of S104, which is not repeated in this embodiment of the present application. .

It can be understood that, for the sudden abnormality of the first performance index, the electronic device can also determine the root cause of the abnormality of the first performance index from a plurality of second performance indicators and other factors. The problem of inaccuracy of the second performance index determined from the plurality of second performance indexes and causing the abnormality of the first performance index can be avoided because the obtained multiple second performance indexes are not comprehensive enough. It is also possible to avoid the problem of inaccuracy of the second performance index determined from a plurality of second performance indicators that causes the first performance index to be abnormal when the first performance index is abnormal due to a hardware failure. Therefore, determining the root cause of the abnormality of the first performance index from the plurality of second performance indicators and other factors is more accurate than determining the root cause of the abnormality of the first performance index from the plurality of second performance indicators.

In this embodiment of the present application, the electronic device first calculates the influence factor of each child node on the parent node at each sampling time, and then calculates the influence factor of each child node on the parent node within the preset positioning period (that is, the influence factor of each child node on the parent node). The influence factor of the node), or for each child node in the tree network, the influence factor of the child node on the parent node can be directly calculated.

Specifically, since the Q sampled values of the performance indicators of each node in the tree network are all arrays arranged in the order of the sampling moments, it can be known that the Q sampled values can be represented by the The trend component of the change trend, the periodic component that occurs periodically in the Q sampled values, and the residual characterizing the fluctuation of the Q sampled values are composed. Since the residual represents the fluctuation of the Q sampling values of the performance index of any node, if the residual of the performance index of any node has a large correlation with the residual of the performance index of the parent node of any node, then The fluctuation of the Q sampling values representing the performance index of any node has a greater impact on the fluctuation of the Q sampling values of the performance index of the parent node of any node. Therefore, the influence factor of any node on the parent node can be determined by using the correlation between the residual of the performance index of any node and the residual of the performance index of the parent node of any node.

In this embodiment of the present application, the electronic device may determine the influence factor of any node on the parent node according to the correlation of the residuals. Specifically, as shown in FIG. 8 , the electronic device may execute S801-S803 for each child node in the tree network to calculate the influence factor of each child node on the parent node. In order to illustrate the process of calculating the influence factor of each child node on the parent node more clearly, each child node is referred to as a second child node.

S801. The electronic device calculates the first residual of the performance index of the second child node according to the Q sampled values of the performance index of the second child node; wherein the second child node is any child node of the tree network; the first The residual is used to characterize the fluctuation of the Q sampled values of the performance index of the second child node.

Wherein, the residual of the performance index of the second child node is referred to as the first residual. The Q sampled values of the performance indicator of the second child node may include Q sampled values of one performance indicator, or Q sampled values of each performance indicator. The first residual of the performance index of the second child node may include the first residual of one performance index, or the first residual of each performance index.

In this embodiment of the present application, if the Q sampled values of the performance indicators of the second subnode include Q sampled values of each performance indicator, the electronic device calculates the performance according to the Q sampled values of each performance indicator of the second subnode The first residual of the indicator.

In this embodiment of the present application, the electronic device may use a time series decomposition algorithm to extract the trend component of the performance index of the second child node and the trend component of the performance index of the second child node from the Q sampled values of the performance index of the second child node. Periodic component; then subtract the trend component and the periodic component from the Q sampled values of the performance index of the second child node to obtain the first residual of the performance index of the second child node. The trend component is used to represent the change trend of the Q sampled values of the performance index of the second child node, and the periodic component is used to represent the periodic change of the Q sampled values of the performance index of the second child node.

Among them, the time series decomposition algorithm includes the seasonal trend decomposition method based on local weighted regression (Seasonal-Trend decomposition procedure based on Loess, STL decomposition method).

Specifically, the electronic device performs the following steps for each performance indicator of the second child node to obtain the first residual of the performance indicator: from the Q sampled values of the performance indicator, extracting the trend component of the performance indicator and the Periodic component; then subtract the trend component and the periodic component from the Q sampled values of the performance index to obtain the first residual of the performance index.

In the embodiment of the present application, the electronic device may use H sample values of the performance index of the second sub-node in advance to train a differential integrated moving average autoregressive model (Autoregressive Integrated Moving Average model, ARIMA model) based on the time series decomposition algorithm, and obtain The components of the second child node fit the model. The component fitting model has the capability of extracting a trend component of the performance index of the second child node and a periodic component of the performance index of the second child node from a plurality of sampled values of the performance index of the second child node.

It should be noted that, in addition to the above ARIMA model, the electronic device can also train other models based on the time series decomposition algorithm to obtain the component fitting model of the second child node, for example, an autoregressive model (Autoregressive model, AR model), this The application examples are not limited.

In this embodiment of the present application, the electronic device may input the Q sampled values of the performance indicators of the second child node into the component fitting model of the second child node, run the component fitting model of the second child node, and output the component fitting model of the second child node. The trend component of the performance indicator and the periodic component of the performance indicator of the second child node.

Specifically, the electronic device performs the following steps for each performance index of the second child node to obtain the trend component and periodic component of the performance index: inputting the Q sampled values of the performance index into the component fitting model of the second child node, Run the component fitting model of the second child node, and output the trend component of the performance index and the periodic component of the performance index.

S802. The electronic device calculates a second residual of the performance index of the parent node of the second child node according to the Q sampled values of the performance index of the parent node of the second child node, where the second residual is used to characterize the performance index of the second child node. The fluctuation of the Q sampled values of the parent node's performance index.

The residual of the performance index of the parent node of the second child node is referred to as the second residual. The parent node of the second child node may include one parent node or multiple parent nodes. The second residual of the performance index of the parent node of the second child node may include the second residual of the performance index of one parent node of the second child node, or the second residual of the performance index of each parent node of the second child node Difference.

The Q sampled values of the performance indicator of any parent node of the second child node may include Q sampled values of one performance indicator, or Q sampled values of each performance indicator. The second residual of the performance index of any parent node of the second child node may include the second residual of one performance index, or the second residual of each performance index.

In this embodiment of the present application, if the Q sampled values of the performance indicators of any parent node of the second child node include Q sampled values of multiple performance indicators, the electronic device will perform an analysis on each parent node of the second child node. The Q sampled values of the performance indicator are used to calculate the second residual of the performance indicator.

It should be noted that the process of calculating the second residual of the performance index of the parent node of the second child node by the electronic device is the same as the process of calculating the first residual of the performance index of the second child node. I won't go into details.

S803. The electronic device calculates the influence factor of the second child node on the parent node according to the preset correlation coefficient algorithm, the first residual and the second residual.

The parent node of the second child node includes one parent node or multiple parent nodes, and the influence factor of the second child node on the parent node includes the influence factor of the second child node on a parent node or the second child node on each parent node impact factor.

In the embodiment of the present application, for each parent node of the second child node, the electronic device performs the calculation according to the preset correlation coefficient algorithm, the first residual of the performance index of the second child node, and the performance of each parent node of the second child node. The second residual of the indicator calculates the influence factor of the second child node on each parent node.

Specifically, if the first residual of the performance index of the second child node includes the first residual of one performance index, the second residual of the performance index of a parent node of the second child node includes the second residual of one performance index difference, the electronic device calculates the correlation between the first residual of the one performance index and the second residual of the one performance index, and uses the correlation as the influence factor of the second child node on the parent node.

If the first residual of the performance index of the second child node includes the first residual of multiple performance indexes, and the second residual of the performance index of a parent node of the second child node includes the second residual of one performance index, The electronic device calculates the correlation between the first residual of each performance index of the second child node and the second residual of this one performance index, and obtains multiple correlations one-to-one corresponding to the multiple performance indexes of the second child node ; Then average the multiple correlations to obtain the influence factor of the second child node on this parent node.

If the first residual of the performance index of the second child node includes the first residual of one performance index, and the second residual of the performance index of a parent node of the second child node includes the second residual of multiple performance indexes, The electronic device calculates the correlation between the first residual of one performance index of the second child node and the second residual of each performance index, and obtains multiple correlations one-to-one corresponding to the multiple performance indexes of the parent node; The multiple correlations are averaged to obtain the influence factor of the second child node on this one parent node.

If the first residual of the performance index of the second child node includes the first residual of multiple performance indexes, the second residual of the performance index of a parent node of the second child node includes the second residual of multiple performance indexes , the electronic device calculates the correlation between the first residual of each performance index of the second child node and the second residual of each performance index, and obtains multiple correlations corresponding to each performance index of the second child node; Average multiple correlations corresponding to each performance index of the second child node to obtain an average value corresponding to each performance index of the second child node; then, average the average values corresponding to all performance indexes of the second child node , to get the influence factor of each performance index of the second child node on this parent node.

In the embodiment of the present application, the calculation formula of the preset correlation system may be as shown in formula (1):

Among them, U ₁ is the first residual, U ₂ is the second residual; r(U ₁ , U ₂ ) is the correlation between the first residual and the second residual (or called correlation coefficient); Cov(U ₁ , U ₂ ) is the covariance of U ₁ and U ₂ ; Var[U ₁ ] is the variance of U ₁ ; Var[U ₂ ] is the variance of U ₂ .

It can be understood that the electronic device determines the influence factor of the second child node on the parent node according to the correlation between the first residual and the second residual, and then uses the influence factor to determine the second performance that causes the first performance index to be abnormal. The indicator refers to the degree of influence on the fluctuation of the Q sampling values of the performance index of the second child node on the fluctuation of the Q sampling values of the performance index of the second child node to determine the abnormality of the first performance index. the second performance indicator. However, if the fluctuation of the Q sampling values of the performance index of any node has a greater impact on the fluctuation of the Q sampling values of the performance index of the parent node of any node, it can be determined that the performance index of any node affects the parent node. Therefore, determining the influence factor of the second child node on the parent node according to the correlation between the first residual and the second residual can improve the location of the root cause of the abnormal first performance index. accuracy.

In the embodiment of the present application, after the electronic device calculates the influence factors of all the child nodes in the tree network on the parent node, the breadth-first algorithm (also called breadth-first algorithm) can be used to first calculate the influence factors of each child node in the tree network For the influence factor of the parent node, calculate the influence factor of each leaf node in the tree network on the root node; then determine the node with the largest influence factor on the root node from all the leaf nodes in the tree network, and determine the influence on the root node. The performance indicator of the node with the largest factor is the second performance indicator that causes the first performance indicator to be abnormal. Specifically, as shown in FIG. 9 , S104 may include S901-S902.

S901. The electronic device calculates the influence factor of the leaf node on the root node in the tree network according to the influence factor of each child node in the tree network on the parent node.

The electronic device calculates the influence factor of each leaf node on the root node for each leaf node in the tree network. The influence factor of each leaf node on the root node may include one influence factor or multiple influence factors. The total number of influence factors of each leaf node to the root node is equal to the total number of paths from each leaf node to the root node.

In the embodiment of the present application, for each leaf node, the electronic device multiplies all the influence factors on each path from the leaf node to the root node to obtain an influence factor of the leaf node on the root node.

Exemplarily, as shown in the tree network of the first performance index as shown in FIG. 10 , the tree network includes a root node A and multiple child nodes, and the multiple child nodes include node B, node C, node D, node E, and node F. , Node G, and Node H. The root node A is the parent node of node B, node C, and node D. Node B is the parent of Node E and Node F. Node C is the parent of node F. Node D is the parent of Node G and Node H. The influence factor of node B on root node A is 0.6, the influence factor of node C on root node A is 0.3, the influence factor of node D on node A is 0.5, the influence factor of node E on node B is 0.6, and the influence factor of node F on node A is 0.6. The influence factor of B is 0.7, the influence factor of node F on node C is 0.1, the influence factor of node G on node D is 0.3, and the influence factor of node H on node D is 0.2.

Furthermore, the influence factor of node E on node B and the influence factor of node B on root node A can be multiplied, and the influence factor of node E on root node A can be obtained as 0.36. Multiply the impact factor of node F on node B and the impact factor of node B on root node A, and get an impact factor of node F on root node A as 0.42; The influence factor of node A is multiplied, and another influence factor of node F to root node A is obtained as 0.03. Multiplying the influence factor of node G on node D and the influence factor of node D on root node A, another influence factor of node G on root node A is obtained as 0.15. Multiplying the influence factor of node H on node D and the influence factor of node D on root node A, the other influence factor of node H on root node A is 0.10.

S902. The electronic device determines the performance indexes of the N leaf nodes in the tree network as the second performance indexes that cause the first performance index to be abnormal; wherein, the influence factor of the N leaf nodes on the root node is greater than that of other nodes in the tree network The influence factor of the leaf node on the root node, N≥1, N is an integer.

The electronic device may select N leaf nodes from all the leaf nodes in the tree network, and determine the performance indicators of the N leaf nodes as the second performance indicators that cause the first performance indicator to be abnormal.

In some embodiments, the electronic device may sort all leaf nodes in descending order of shadow factors according to the influence factors of all leaf nodes on the parent node, and then select the top N leaf nodes from the sorted leaf nodes. Wherein, if the influence factor of a leaf node to the root node includes multiple influence factors, the largest influence factor of the leaf node to the root node can be used for sorting.

In this embodiment of the present application, after S902, the electronic device may further acquire path information from each of the N leaf nodes to the root node. The path information may include connection relationships of all nodes on the path from a leaf node to the root node, and performance indicators of all nodes on the path from a leaf node to the root node. Further, the electronic device or the user can adjust the second performance index that causes the first performance index to be abnormal according to the path information and the second performance index that causes the first performance index to be abnormal, so as to solve the problem of the first performance index being abnormal.

Wherein, if there are multiple paths from a leaf node to the root node, the path information from the leaf node to the root node may include the path information corresponding to the maximum influence factor of the leaf node on the root node, or may include the leaf node to the root node. multiple path information.

Exemplarily, for the tree network shown in FIG. 10 , when N is equal to 1, the second performance indicator determined by the electronic device that causes the first performance indicator to be abnormal is the performance indicator of node F. The path information from node F to root node A may include: information indicating that node F points to node B and node B points to node A, the performance index of node F, the performance index of node B and the performance index of node A.

It can be understood that since the leaf nodes are terminal nodes in the tree network, the root cause of the abnormality of the first performance index exists in all the leaf nodes in the tree network. Therefore, the electronic device directly calculates the influence factors of all leaf nodes in the tree network on the root node; and then determines that the performance index of the leaf node with a larger influence factor on the root node is the second performance index that causes the abnormality of the first performance index.

In the embodiment of the present application, the tree network includes a first preset node, the first preset node is not a leaf node, and the influence factor of the child nodes of the first preset node on the first preset node is less than or equal to a preset threshold. Since the influence factor of the child nodes of the first preset node on the first preset node is less than or equal to the preset threshold, the first preset node can be considered as an indecomposable node. Then, the influence factor of each leaf node in the tree network on the root node is calculated, and the influence factor of the first preset node on the root node is also calculated. Then determine the node with the largest influence factor on the root node from all the leaf nodes and the first preset node of the tree network, and determine that the performance index of the node with the largest influence factor on the root node is the second one that causes the first performance index to be abnormal. Performance. Specifically, as shown in FIG. 11 , S104 may include S1101-S1102.

S1101. The electronic device calculates the influence factor of the leaf node on the root node in the tree network according to the influence factor of each child node on the parent node in the tree network, and calculates the influence factor of the first preset node on the root node.

Wherein, the tree network may include one or more first preset nodes. The electronic device calculates the influence factor of each first preset node in the tree network on the root node.

It should be noted that the process of calculating the influence factor of the first preset node on the root node by the electronic device is the same as the process of calculating the influence factor of the leaf node on the root node. For the specific process of S1101, you can refer to the detailed description of the above-mentioned S901. The application examples are not repeated here.

In this embodiment of the present application, before S1101, the electronic device may, for other nodes in the tree network except the leaf node and the root node, determine whether the influence factor of each child node of each other node on the parent node is less than or equal to the predetermined factor. Set the threshold. If the influence factor of all child nodes of each other node on the parent node is less than or equal to the preset threshold, then each other node is determined to be a first preset node. If the influence factor of at least one child node of each other node on the parent node is greater than the preset threshold, it is determined that each other node is not the first preset node.

Exemplarily, assuming that the preset threshold is 0.3, for the tree network shown in FIG. 10 , since the influence factor of the child node F of node C on node C is 0.1, which is less than 0.3; The influence factor of node D is 0.3, and the influence factor of node D's child node H on node D is 0.2, which is less than 0.3. Therefore, the tree network includes two first preset nodes, namely node C and node D.

S1102. The electronic device determines the performance indicators of the M nodes in the leaf node and the first preset node as the second performance indicator that causes the first performance indicator to be abnormal; wherein, the influence factor of the M nodes on the root node is greater than the tree shape The influence factor of other nodes in the network on the root node, M≥1, M is an integer.

The electronic device may select M nodes from all leaf nodes and all first preset nodes in the tree network; and then determine the performance indicators of the M nodes as the second performance indicators that cause the first performance indicator to be abnormal. Wherein, if the influence factor of a leaf node or a first preset node on the root node includes multiple influence factors, the maximum influence factor of the leaf node or the first preset node on the root node can be used to select M node.

It should be noted that the specific process in which the electronic device selects M nodes from all leaf nodes and all first preset nodes can participate in the above detailed description of selecting N leaf nodes from all leaf nodes in the tree network. This embodiment of the present application will not be repeated here.

In this embodiment of the present application, after S1102, the electronic device may further acquire path information from each of the M nodes to the root node. The path information may include connection relationships of all nodes on the path from a node to the root node, and performance indicators of all nodes on the path from a node to the root node. Further, the electronic device or the user can adjust the second performance index that causes the first performance index to be abnormal according to the path information and the second performance index that causes the first performance index to be abnormal, so as to solve the problem of the first performance index being abnormal.

Wherein, if there are multiple paths from a node to the root node, the path information from the node to the root node may include the path information corresponding to the maximum influence factor of the node on the root node, or may include multiple paths from the node to the root node information.

It can be understood that although the first preset node in the tree network is not a leaf node, the influence factor of the child nodes of the first preset node on the first preset node does not exceed the preset threshold, that is, the first preset node. The performance index of the child node of the first preset node has little influence on the performance index of the first preset node, it can be considered that the first preset node is similar to the leaf node, and the root cause of the abnormal first performance index may also exist in the tree network. in the first preset node in . Therefore, the electronic device calculates the influence factor of all leaf nodes in the tree network on the root node, and the influence factor of the first preset node in the tree network on the root node; In a preset node, a node with a larger influence factor on the root node is determined, and its performance index is the second performance index that causes the abnormality of the first performance index. In this way, both the leaf node that may include the root cause that causes the abnormality of the first performance index and the first preset node that may include the root cause that causes the abnormality of the first performance index are judged, and the first performance index that causes the abnormality can be determined more accurately. Abnormal secondary performance indicator.

In the embodiment of the present application, after the electronic device calculates the influence factors of all the child nodes in the tree network on the parent node, the depth-first algorithm can be used to search the tree network from the root node to the parent node level by level. The node with the largest influence factor is determined until the found node is a leaf node or the first preset node, and the performance index of the found node is determined to be the second performance index that causes the first performance index to be abnormal. Specifically, taking the found node as a leaf node, and determining that the performance index of the found node is the second performance index causing the abnormality of the first performance index as an example, the specific process of S104 is described. As shown in FIG. 12, S104 may include S1201-S1208.

S1201. The electronic device takes the root node as the second parent node, and determines the child node with the largest impact factor on the second parent node according to the influence factors of each child node in the tree network on the parent node.

S1202. The electronic device determines whether the child node with the largest influence factor on the second parent node is a leaf node.

If the electronic device determines that the child node with the greatest influence factor on the second parent node is not a leaf node, execute S1203; if it is determined that the child node with the greatest influence factor on the second parent node is a leaf node, execute S1204.

S1203. The electronic device takes the child node with the largest influence factor on the second parent node as the second parent node, and determines the child node with the largest influence factor on the second parent node according to the influence factors of each child node in the tree network on the parent node. child node.

After S1203, the electronic device continues to execute S1202.

S1204, the electronic device takes the child node with the largest influence factor of the second parent node as the child node to be selected.

S1205. The total number of child nodes to be selected obtained by the electronic device statistics.

S1206, the electronic device determines whether the total number of child nodes to be selected is less than N.

Among them, N≥1, N is an integer.

If the electronic device determines that the total number of child nodes to be selected is less than N, execute S1207; if it is determined that the total number of child nodes to be selected is greater than or equal to N, execute S1208.

S1207, the electronic device deletes the child node with the largest influence factor on the second parent node from the tree network.

The electronic device also deletes all parent nodes without child nodes from the tree network caused by deleting the child node with the largest influence factor on the second parent node.

Exemplarily, assuming that N is equal to 2, for the tree-shaped network shown in FIG. 10, if S1201-S1207 are executed, the node E is deleted from the tree-shaped network. For the tree-shaped network after deleting node E, continue to execute S1201-S1207, and delete node F from the tree-shaped network. Since node B has no child nodes after node F is deleted, node B is also deleted from the tree-shaped network.

After S1207, the electronic device continues to execute S1201.

S1208. The electronic device uses the obtained performance index of the child node to be selected as the second performance index that causes the first performance index to be abnormal.

It can be understood that the electronic device can search the tree network from the root node to the node with the largest influence factor on the parent node, until the found node is a leaf node, and determine that the performance index of the found node is: The second performance index that causes the first performance index to be abnormal can also determine the path information from the found node to the root node.

It can be understood that, in order to realize the above-mentioned functions, the above-mentioned electronic device or root cause locating apparatus includes corresponding hardware structures and/or software modules for executing each function. Those skilled in the art should easily realize that, in conjunction with the units and algorithm steps of the examples described in the embodiments disclosed herein, the embodiments of the present application can be implemented in hardware or a combination of hardware and computer software. Whether a function is performed by hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Skilled persons may use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the embodiments of the present application.

In this embodiment of the present application, the electronic device or the root cause locating device may be divided into functional modules according to the foregoing method examples. For example, each functional module may be divided corresponding to each function, or two or more functions may be integrated into one processing unit. in the module. The above-mentioned integrated modules can be implemented in the form of hardware, and can also be implemented in the form of software function modules. It should be noted that, the division of modules in the embodiments of the present application is schematic, and is only a logical function division, and there may be other division manners in actual implementation.

In the case where each functional module is divided according to each function, as shown in FIG. 13 , an embodiment of the present application provides a root cause locating device 1300 .

Wherein, the obtaining module 1301 is used to support the root cause locating device 1300 to perform S101-S102 in the above method embodiments, and/or other processes for the techniques described herein. The computing module 1302 is used to support the root cause locating device 1300 to perform S103, S301-S302, S401-S402, S601-S603, S701-S703, S801-S803 in the above method embodiments, and/or for the techniques described herein other processes. The location module 1303 is used to support the root cause location device 1300 to perform S104, S604, S704, S901-S902, S1101-S1102, S1201-S1208, and/or other processes for the techniques described herein in the above method embodiments.

Further, the above root cause locating apparatus 1300 may further include: a model training module. The model training module is used to support the root cause locating apparatus 1300 to perform the process of training a preset attention model in the above method embodiments, and/or other processes for the techniques described herein.

Of course, the above root cause locating device 1300 includes but is not limited to the unit modules listed above. For example, the root cause locating apparatus 1300 may further include a storage unit for storing sampled values of performance indicators. In addition, the specific functions that can be realized by the above functional units also include but are not limited to the functions corresponding to the method steps described in the above examples. For the detailed description of other units of the root cause locating device 1300, please refer to the detailed description of the corresponding method steps. This embodiment of the present application will not be repeated here.

In the case of using an integrated unit, the above-mentioned acquisition module 1301 , calculation module 1302 , and location module 1303 can be integrated into one processing module, and the above-mentioned storage unit can be a storage module of the root cause location device 1300 .

FIG. 13 shows a possible schematic structural diagram of the electronic device involved in the above embodiment. The electronic electronic device 1400 includes: a processing module 1401 , a storage module 1402 and a communication module 1403 .

The processing module 1401 is used to control and manage the electronic device 1400 . The storage module 1402 is used to store program codes and data of the electronic device 1400 . The communication module 1403 is used to communicate with other devices. For example, the communication module is used to receive or send data to other devices.

The processing module 1401 may be a processor or a controller, for example, a CPU, a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application-Specific Integrated Circuit, ASIC), a field programmable Gate Array (Field Programmable Gate Array, FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute the various exemplary logical blocks, modules and circuits described in connection with this disclosure. The processor may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The communication module 1403 may be a transceiver, a transceiver circuit, a communication interface, or the like. The storage module 1402 may be a memory.

Embodiments of the present application also provide a computer-readable storage medium, where computer program codes are stored in the computer-readable storage medium. When the above-mentioned electronic device executes the computer program code, the electronic device realizes the execution of FIG. 1 , FIG. 3 , and FIG. 4. The relevant method steps in any one of FIG. 6 , FIG. 7 , FIG. 8 , FIG. 9 , FIG. 11 and FIG. 12 implement the method in the above embodiment.

Embodiments of the present application also provide a computer program product, which, when the computer program product runs on an electronic device, enables the electronic device to execute FIG. 1 , FIG. 3 , FIG. 4 , FIG. 6 , FIG. 7 , FIG. 8 , FIG. 9 , The relevant method steps in any of Figures 11 and 12 implement the methods in the above-described embodiments.

Wherein, the root cause locating device 1300, the electronic device 1400, the computer-readable storage medium or the computer program product provided in this application are all used to execute the corresponding method provided above, therefore, the beneficial effects that can be achieved can refer to the above The beneficial effects in the corresponding method provided in this article will not be repeated here.

From the description of the above embodiments, those skilled in the art can clearly understand that for the convenience and brevity of the description, only the division of the above functional modules is used as an example for illustration. In practical applications, the above functions can be allocated as required. It is completed by different functional modules, that is, the internal structure of the device is divided into different functional modules, so as to complete all or part of the functions described above. For the specific working process of the system, apparatus and unit described above, reference may be made to the corresponding process in the foregoing method embodiments, and details are not described herein again.

In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the device embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be Incorporation may either be integrated into another system, or some features may be omitted, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.

The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, and the computer software products are stored in a storage medium , which includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: flash memory, removable hard disk, read-only memory, random access memory, magnetic disk or optical disk and other media that can store program codes.

The above are only specific embodiments of the present application, but the protection scope of the present application is not limited to this. should be covered within the scope of protection of this application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

A root cause locating method, characterized in that the method comprises:

Obtaining a tree-shaped network of first performance indicators; wherein, the tree-shaped network includes the first performance indicators and a plurality of second performance indicators that affect the first performance indicators, and the first performance indicators correspond to the tree The root node of the tree-shaped network, the plurality of second performance indicators correspond to other nodes in the tree-shaped network except the root node;

acquiring the sampled value of the first performance index and the sampled values of the plurality of second performance indicators obtained by sampling within a preset positioning period;

According to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, the influence factor of each child node in the tree network on the parent node is calculated; wherein, the influence of each child node on the parent node The factor is used to characterize the degree of influence of the performance index of each child node on the performance index of the parent node;

According to the influence factor of each child node in the tree-shaped network on the parent node, the second performance index in the tree-shaped network that causes the abnormality of the first performance index is determined.
The method according to claim 1, wherein the preset positioning period includes Q sampling moments, where Q≥2, and Q is an integer;

The acquiring the sampled value of the first performance indicator and the sampled values of the plurality of second performance indicators sampled within the preset positioning period includes:

At the qth sampling time of the preset positioning period, the qth sampling value of the first performance index and the qth sampling value of the plurality of second performance indicators are obtained by sampling; wherein, q is in {1 , ..., Q} and take values in turn.
The method according to claim 2, wherein, according to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, calculating the relative value of each child node in the tree network to the parent The impact factor of the node, including:

Execute for each parent node in the tree network: take the qth sampled value of the performance index of all child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, run The impact factor estimation model of the first parent node, outputting the qth impact factor of each child node of the first parent node on the first parent node; wherein, the first parent node is the tree Any parent node in the shape network; the impact factor estimation models of different parent nodes are different; the impact factor estimation model of the first parent node has sampling values from the performance indicators of all child nodes of the first parent node , extracting the ability of each child node of the first parent node to influence factors of the first parent node;

Perform for each child node of the first parent node: calculate the average value of Q influence factors of the first child node on the first parent node, and obtain the impact of the first child node on the first parent node factor; wherein, the first child node is any child node of the first parent node.
The method according to claim 2, wherein, according to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, calculating the relative value of each child node in the tree network to the parent The impact factor of the node, including:

Execute for each parent node in the tree network: take the qth sampled value of the performance index of all child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, run The impact factor estimation model of the first parent node, outputting the qth impact factor of each child node of the first parent node on the first parent node; wherein, the first parent node is the tree Any parent node in the shape network; the impact factor estimation models of different parent nodes are different; the impact factor estimation model of the first parent node has sampling values from the performance indicators of all child nodes of the first parent node , extracting the ability of each child node of the first parent node to influence factors of the first parent node;

If the p-th sampling time in the Q sampling times is the sudden change of the performance index, perform for each child node of the first parent node: calculate the first p- of the first child node to the first parent node. The average value of 1 impact factor, to obtain the impact factor of the first child node on the first parent node before mutation; calculate the p-th -Qth impact factor of the first child node on the first parent node to obtain the influence factor of the first child node on the first parent node after mutation; calculate the influence factor of the first child node on the first parent node before mutation and the The difference or ratio of the influence factor of the child node to the first parent node, and the difference or ratio is used as the influence factor of the first child node to the first parent node;

Wherein, p is a positive integer less than or equal to Q; the first child node is any child node of the first parent node.
The method according to claim 3 or 4, characterized in that, in the calculation according to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, each child in the tree network is calculated Before the influence factor of the node on the parent node, the method further includes:

The h-th sample value of the performance index of the first parent node and the h-th sample value of the performance index of all child nodes of the first parent node are used as training samples to train the preset attention model to obtain the The impact factor prediction model of the first parent node;

Among them, h takes values in sequence in {1,...,H}, H≥100, and H is an integer.
The method according to claim 2, wherein, according to the sampled value of the first performance index and the sampled value of the plurality of second performance indicators, calculating the relative value of each child node in the tree network to the parent The impact factor of the node, including:

Perform the following steps for each child node in the tree network to obtain the influence factor of each child node on the parent node:

Calculate the first residual of the performance index of the second child node according to the Q sampled values of the performance index of the second child node; wherein the second child node is any child node of the tree network; The first residual is used to represent the fluctuation of the Q sampling values of the performance index of the second child node;

Calculate a second residual of the performance index of the parent node of the second child node according to the Q sampled values of the performance index of the parent node of the second child node, where the second residual is used to characterize the first The fluctuation of the Q sampling values of the performance index of the parent node of the two child nodes;

Calculate the correlation between the first residual and the second residual, and according to the correlation between the first residual and the second residual, obtain the relationship between the second child node and the parent The influence factor of the node.
The method according to claim 6, wherein the calculating the first residual of the performance index of the second child node according to the Q sampling values of the performance index of the second child node comprises:

Using a time series decomposition algorithm, extract the trend component of the performance index of the second child node and the periodic component of the performance index of the second child node from the Q sampled values of the performance index of the second child node; Wherein, the trend component is used to represent the change trend of the Q sampled values of the performance index of the second child node, and the period component is used to represent the period of the Q sampled values of the performance index of the second child node sexual change;

The trend component and the period component are subtracted from the Q sampled values of the performance index of the second child node to obtain a first residual of the performance index of the second child node.
The method according to any one of claims 1-7, characterized in that, according to the influence factor of each child node in the tree network on the parent node, determining that the tree network causes the first A second performance index with abnormal performance index, including:

According to the influence factor of each child node in the tree network on the parent node, calculate the influence factor of the leaf node in the tree network on the root node;

Determine the performance indicators of the N leaf nodes in the tree-shaped network as the second performance indicators that cause the first performance indicator to be abnormal; wherein, the impact factor of the N leaf nodes on the root node is greater than all Influence factor of other leaf nodes in the tree network on the root node, N≥1, N is an integer.
The method according to any one of claims 1-7, wherein the tree network includes a first preset node, the first preset node is not a leaf node, and the first preset node The influence factor of the child node on the first preset node is less than or equal to the preset threshold;

Wherein, according to the influence factor of each child node in the tree network on the parent node, determining the second performance index in the tree network that causes the first performance index to be abnormal, including:

According to the influence factor of each child node in the tree network on the parent node, calculate the influence factor of the leaf node in the tree network on the root node, and calculate the influence factor of the first preset node on the root node. Impact factor;

Determining the performance indicators of M nodes in the leaf node and the first preset node as the second performance indicator that causes the first performance indicator to be abnormal; wherein the M nodes are responsible for the root node The influence factor of is greater than the influence factor of other nodes in the tree network on the root node, M≥1, and M is an integer.
A root cause locating device, characterized in that the device comprises:

an obtaining module, configured to obtain a tree-shaped network of the first performance index, and obtain the sampling value of the first performance index and the sampling value of the plurality of second performance indexes obtained by sampling within a preset positioning period; wherein, the The tree network includes the first performance indicator and a plurality of second performance indicators that affect the first performance indicator, the first performance indicator corresponds to the root node of the tree network, and the plurality of second performance indicators corresponding to other nodes in the tree network except the root node;

a calculation module, configured to calculate the influence factor of each child node in the tree network on the parent node according to the sampling value of the first performance index and the sampling value of the plurality of second performance indices; wherein, each child node The influence factor on the parent node is used to characterize the degree of influence of the performance index of each child node on the performance index of the parent node;

The positioning module is configured to determine the second performance index in the tree network that causes the abnormality of the first performance index according to the influence factor of each child node in the tree network on the parent node.
The device according to claim 10, wherein the preset positioning period includes Q sampling moments, where Q≥2, and Q is an integer;

The acquisition module is specifically configured to obtain the qth sampling value of the first performance index and the qth sampling value of the plurality of second performance indicators by sampling at the qth sampling time of the preset positioning period value; among them, q takes values in sequence in {1,...,Q}.
The apparatus of claim 11, wherein:

The computing module is specifically used for:

Execute for each parent node in the tree network: take the qth sampled value of the performance index of all child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, run The impact factor estimation model of the first parent node, outputting the qth impact factor of each child node of the first parent node on the first parent node; wherein, the first parent node is the tree Any parent node in the shape network; the impact factor estimation models of different parent nodes are different; the impact factor estimation model of the first parent node has sampling values from the performance indicators of all child nodes of the first parent node , extracting the ability of each child node of the first parent node to influence factors of the first parent node;

Perform for each child node of the first parent node: calculate the average value of Q influence factors of the first child node on the first parent node, and obtain the impact of the first child node on the first parent node factor; wherein, the first child node is any child node of the first parent node.
The apparatus of claim 11, wherein:

The computing module is specifically used for:

Execute for each parent node in the tree network: take the qth sampled value of the performance index of all child nodes of the first parent node as the input of the impact factor estimation model of the first parent node, run The impact factor estimation model of the first parent node, outputting the qth impact factor of each child node of the first parent node on the first parent node; wherein, the first parent node is the tree Any parent node in the shape network; the impact factor estimation models of different parent nodes are different; the impact factor estimation model of the first parent node has sampling values from the performance indicators of all child nodes of the first parent node , extracting the ability of each child node of the first parent node to influence factors of the first parent node;

If the p-th sampling time in the Q sampling times is the sudden change of the performance index, perform for each child node of the first parent node: calculate the first p- of the first child node to the first parent node. The average value of 1 impact factor, to obtain the impact factor of the first child node on the first parent node before mutation; calculate the p-th -Qth impact factor of the first child node on the first parent node to obtain the influence factor of the first child node on the first parent node after mutation; calculate the influence factor of the first child node on the first parent node before mutation and the The difference or ratio of the influence factor of the child node to the first parent node, and the difference or ratio is used as the influence factor of the first child node to the first parent node;

Wherein, p is a positive integer less than or equal to Q; the first child node is any child node of the first parent node.
The device according to claim 12 or 13, wherein the device further comprises:

A model training module, configured to calculate, according to the sampling value of the first performance index and the sampling value of the plurality of second performance indices, before calculating the influence factor of each child node in the tree network on the parent node, The h-th sample value of the performance index of the first parent node and the h-th sample value of the performance index of all child nodes of the first parent node are used as training samples to train the preset attention model to obtain the The impact factor prediction model of the first parent node;

Among them, h takes values in sequence in {1,...,H}, H≥100, and H is an integer.
The apparatus of claim 11, wherein:

The computing module is specifically configured to perform the following steps for each child node in the tree network to obtain the influence factor of each child node on the parent node:

Calculate the first residual of the performance index of the second child node according to the Q sampled values of the performance index of the second child node; wherein the second child node is any child node of the tree network; The first residual is used to represent the fluctuation of the Q sampling values of the performance index of the second child node;

Calculate a second residual of the performance index of the parent node of the second child node according to the Q sampled values of the performance index of the parent node of the second child node, where the second residual is used to characterize the first The fluctuation of the Q sampling values of the performance index of the parent node of the two child nodes;

Calculate the correlation between the first residual and the second residual, and according to the correlation between the first residual and the second residual, obtain the relationship between the second child node and the parent The influence factor of the node.
The apparatus of claim 15, wherein:

The computing module is specifically used for:

Using a time series decomposition algorithm, extract the trend component of the performance index of the second child node and the periodic component of the performance index of the second child node from the Q sampled values of the performance index of the second child node; Wherein, the trend component is used to represent the change trend of the Q sampled values of the performance index of the second child node, and the period component is used to represent the period of the Q sampled values of the performance index of the second child node sexual change;

The trend component and the period component are subtracted from the Q sampled values of the performance index of the second child node to obtain a first residual of the performance index of the second child node.
The device according to any one of claims 10-16, characterized in that,

The positioning module is specifically used for:

According to the influence factor of each child node in the tree network on the parent node, calculate the influence factor of the leaf node in the tree network on the root node;

Determine the performance indicators of the N leaf nodes in the tree-shaped network as the second performance indicators that cause the first performance indicator to be abnormal; wherein, the impact factor of the N leaf nodes on the root node is greater than all Influence factor of other leaf nodes in the tree network on the root node, N≥1, N is an integer.
The device according to any one of claims 10-16, wherein the tree network includes a first preset node, the first preset node is not a leaf node, and the first preset node The influence factor of the child node on the first preset node is less than or equal to the preset threshold;

Wherein, the positioning module is specifically used for:

According to the influence factor of each child node in the tree network on the parent node, calculate the influence factor of the leaf node in the tree network on the root node, and calculate the influence factor of the first preset node on the root node. Impact factor;

Determining the performance indicators of M nodes in the leaf node and the first preset node as the second performance indicator that causes the first performance indicator to be abnormal; wherein the M nodes are responsible for the root node The influence factor of is greater than the influence factor of other nodes in the tree network on the root node, M≥1, and M is an integer.
An electronic device, characterized in that the electronic device comprises: a processor and a memory for storing executable instructions of the processor;

Wherein, the processor is configured to execute the instructions, so that the electronic device executes the root cause locating method according to any one of claims 1-9.
A computer-readable storage medium, characterized in that the computer-readable storage medium has computer instructions stored thereon, and when the computer instructions are executed on an electronic device, the electronic device is made to perform the execution as in claims 1-9. The root cause localization method of any one.