CN112929363B

CN112929363B - Root cause analysis method and equipment for video field performance parameter abnormity

Info

Publication number: CN112929363B
Application number: CN202110155725.8A
Authority: CN
Inventors: 邓伟辉
Original assignee: Beijing Zitiao Network Technology Co Ltd
Current assignee: Beijing Zitiao Network Technology Co Ltd
Priority date: 2021-02-04
Filing date: 2021-02-04
Publication date: 2022-05-17
Anticipated expiration: 2041-02-04
Also published as: CN112929363A

Abstract

The present disclosure provides a root cause analysis method and device for performance parameter abnormality in the video field, the method includes: when the KPI is abnormal at the first time, determining the abnormal contribution degree of the leaf attribute combination according to the difference between the predicted value and the true value of the KPI corresponding to the leaf attribute combination at the first time for at least two leaf attribute combinations if the KPI is a basic KPI; if the KPI is the ratio of the first basic KPI and the second basic KPI, determining the abnormal contribution of the leaf attribute combination according to the difference between the predicted value and the real value of the first basic KPI and the difference between the predicted value and the real value of the second basic KPI at the first time of the leaf attribute combination; and determining the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination so as to determine the abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination. The method and the device can accurately determine the reason of the performance parameter abnormity in the video field.

Description

Root cause analysis method and equipment for video field performance parameter abnormity

Technical Field

The embodiment of the disclosure relates to the technical field of video streaming media, in particular to a root cause analysis method and equipment for performance parameter abnormity in the video field.

Background

In the field of video streaming media technology, video service quality can be expressed by various KPIs (key performance indicators). The KPI may be a pause rate, a play failure rate, a first frame duration, a pause duration, and the like. The KPIs are determined according to attribute combinations, where the attribute combinations include one or more attributes, and the attributes may be video resolution, network type of the terminal device playing the video, and an operator to which a network used by the terminal device playing the video belongs. For example, the katton rate when the video resolution is the first resolution, the network type is 4G (fourth generation mobile communication technology), and the carrier is carrier a may be determined. In this way, video quality of service can be analyzed by KPI under different combinations of attributes.

When the KPIs are abnormal, how to accurately determine the reason for the abnormality according to the KPIs in the video field becomes an urgent problem to be solved.

Disclosure of Invention

The embodiment of the disclosure provides a root cause analysis method and equipment for performance parameter abnormity in the video field, so as to accurately determine the reason of the abnormity according to KPI in the video field.

In a first aspect, an embodiment of the present disclosure provides a root cause analysis method for performance parameter abnormality in a video field, including:

when a key performance index KPI of video playing quality is abnormal at a first time, for at least two leaf attribute combinations, if the KPI is a basic KPI, determining the abnormal contribution degree of each leaf attribute combination according to the difference value between the predicted value and the true value of the KPI corresponding to each leaf attribute combination at the first time;

if the KPI is obtained by calculating the ratio of a first basic KPI to a second basic KPI, determining the abnormal contribution degree of each leaf attribute combination according to the difference between the predicted value and the real value of the first basic KPI of each leaf attribute combination at the first time and the difference between the predicted value and the real value of the second basic KPI of each leaf attribute combination at the first time;

determining the abnormal contribution degree of a non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination;

and determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination, wherein the abnormal attribute combination is used for representing the attribute combination causing the KPI abnormality, and the attribute combination comprises at least one attribute.

In a second aspect, an embodiment of the present disclosure provides a root cause analysis device for performance parameter abnormality in a video domain, including:

a first abnormal contribution degree determining module, configured to, when a key performance indicator KPI of video playing quality is abnormal at a first time, determine, for at least two leaf attribute combinations, an abnormal contribution degree of each leaf attribute combination according to a difference between a predicted value and a true value of the KPI corresponding to the first time for each leaf attribute combination if the KPI is a basic KPI;

a second abnormal contribution determining module, configured to determine, if the KPI is calculated from a ratio of a first base KPI to a second base KPI, an abnormal contribution of each of the leaf attribute combinations according to a difference between a predicted value and a true value of the first base KPI at the first time for each of the leaf attribute combinations, and a difference between a predicted value and a true value of the second base KPI at the first time for each of the leaf attribute combinations;

a third abnormal contribution degree determining module, configured to determine an abnormal contribution degree of a non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination;

and the abnormal attribute combination determining module is used for determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination, wherein the abnormal attribute combination is used for representing the attribute combination causing the KPI abnormality, and the attribute combination comprises at least one attribute.

In a third aspect, an embodiment of the present disclosure provides an electronic device, including: at least one processor and memory;

the memory stores computer-executable instructions;

the at least one processor executes computer-executable instructions stored by the memory to cause the electronic device to implement the method of the first aspect.

In a fourth aspect, the embodiments of the present disclosure provide a computer-readable storage medium, in which computer-executable instructions are stored, and when executed by a processor, cause a computing device to implement the method according to the first aspect.

In a fifth aspect, the present disclosure provides a computer program for implementing the method according to the first aspect.

The root cause analysis method and device for video domain performance parameter abnormality provided by this embodiment may determine the abnormal contribution degree of the leaf attribute combination through a difference between a predicted value and a true value of the KPI, and determine the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination. In this way, the abnormal degree of each attribute combination can be accurately represented by the abnormal contribution degree of each attribute combination, and the abnormal degree between different attribute combinations can be distinguished by the abnormal contribution degree. And comparing the attribute combinations based on the abnormal contribution degree, and selecting a plurality of attribute combinations with the maximum abnormal contribution degree from the leaf attribute combinations and the non-leaf attribute combinations as abnormal attribute combinations. Therefore, the accuracy of the determined abnormal attribute combination is high, and the accuracy of the abnormal reason is improved.

Drawings

In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present disclosure, and for those skilled in the art, other drawings can be obtained according to the drawings without inventive exercise.

Fig. 1 is a schematic diagram illustrating an online video playing scene provided by an embodiment of the present disclosure;

fig. 2 is a flowchart illustrating steps of a method for root cause analysis of video domain performance parameter anomaly according to an embodiment of the present disclosure;

FIG. 3 is a schematic diagram illustrating a hierarchical structure of a combination of attributes provided by an embodiment of the present disclosure;

fig. 4 is a block diagram of a root cause analysis device for performance parameter anomaly in the video domain according to an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure.

Detailed Description

To make the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, the technical solutions in the embodiments of the present disclosure will be described clearly and completely with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are some, but not all embodiments of the present disclosure. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.

The embodiment of the disclosure can be applied to online video playing scenes. The videos are stored in the server, one or more clients can access the server to obtain the videos from the server, and the obtained videos can be played on the clients. Fig. 1 is a schematic diagram of an online video playing scene, where fig. 1 exemplarily shows a server and four clients: CT1(client1), CT2, CT3 and CT4, and CT1, CT2, CT3 and CT4 are application programs installed in different terminal devices. When a video identifier on the client is clicked, the client acquires the video from the server to play. For example, when a video tag on CT1 is clicked, CT1 retrieves the video from the server. Of course, at least two of the CT1, CT2, CT3, and CT4 in fig. 1 may obtain the same video or different videos from the server at the same time, or may obtain the same video or different videos from the server at different times.

In order to ensure the video playing quality, the server can monitor whether the video is abnormal through the KPI. The KPI can be a pause rate, a play failure rate, a first frame time length and the like, and when the pause rate, the play failure rate or the first frame time length is greater than or equal to a preset threshold value, the KPI is determined to be abnormal, and the video play quality is poor.

When a KPI is abnormal, it is necessary to determine which videos of attributes cause the KPI abnormality. The attribute may be a network type, a video resolution, a network provider, etc. It is to be understood that attributes are not differentiated when the KPIs are counted. For example, when the number of times of hitching is large, it is necessary to determine which network type, which video resolution, which video corresponding to the network provider causes the number of times of hitching to be large.

In the prior art, when a KPI is abnormal, it may be obtained whether the true value of the KPI of each attribute is consistent with the predicted value of the KPI of the attribute, and if so, it is determined that the video corresponding to the attribute is normal; and if the attribute is inconsistent, determining that the video corresponding to the attribute causes KPI abnormity.

However, the above scheme cannot accurately distinguish the abnormal degree between different attributes, resulting in poor accuracy of the abnormal cause.

The embodiment of the disclosure may determine the abnormal contribution degree of the leaf attribute combination through a difference value between the predicted value and the true value of the KPI, and determine the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination. In this way, the abnormal degree of each attribute combination can be accurately represented by the abnormal contribution degree of each attribute combination, and the abnormal degree between different attribute combinations can be distinguished by the abnormal contribution degree. And comparing the attribute combinations based on the abnormal contribution degree, and selecting a plurality of attribute combinations with the maximum abnormal contribution degree from the leaf attribute combinations and the non-leaf attribute combinations as abnormal attribute combinations. Therefore, the accuracy of the determined abnormal attribute combination is high, and the accuracy of the abnormal reason is improved.

The following describes in detail the technical solutions of the embodiments of the present disclosure and how to solve the above technical problems with specific embodiments. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments. Embodiments of the present disclosure will be described below with reference to the accompanying drawings.

Referring to fig. 2, fig. 2 is a flowchart illustrating steps of a method for root cause analysis of performance parameter anomaly in the video domain according to an embodiment of the present disclosure, which may be applied to the server in fig. 1, where the method includes:

s101: when one KPI of video playing quality is abnormal at a first time, for at least two leaf attribute combinations, if the KPI is a basic KPI, determining the abnormal contribution degree of each leaf attribute combination according to the difference value between the predicted value and the true value of the KPI corresponding to each leaf attribute combination at the first time.

In the embodiment of the present disclosure, different attributes may be arbitrarily combined into attribute combinations, and each attribute combination may include at least one attribute. The inclusion of only one attribute in the combination of attributes may be a special case. The attribute combinations may form a hierarchical structure according to a combination relationship, a root node of the hierarchical structure is each attribute, a leaf node of the hierarchical structure is an attribute combination of all attribute combinations, each attribute combination is formed by combining sub-attributes thereof, and a leaf node (hereinafter referred to as a leaf attribute combination) does not have a sub-attribute combination.

As shown in fig. 3, there are 3 attributes A, B and C, a and B combined as attribute combination AB, B and C combined as attribute combination BC, a and C combined as attribute combination AC, AB, AC and BC combined as attribute combination ABC. It can be seen that A, B and C in FIG. 3 are the three root nodes of the hierarchy, and ABC is the leaf attribute combination. For example, a in fig. 3 may be a network type Wi-Fi, B is a video resolution 720P (Progressive scan), i.e., 1280x720, and C is a network provider ISP 1.

When a KPI is abnormal at the first time, it is necessary to perform abnormal cause analysis based on the abnormal contribution degree of the leaf attribute combination to obtain an abnormal attribute combination. Wherein, the first time is the time when the KPI is abnormal. According to the embodiment of the disclosure, different calculation formulas are adopted to calculate the abnormal contribution degree of the leaf attribute combination aiming at different types of KPIs. The abnormality contribution degree is used to indicate the degree of abnormality of the attribute combination. It can be understood that the greater the abnormal contribution degree, the greater the representing abnormal degree, and the greater the probability that the attribute combination is an abnormal attribute combination; the smaller the degree of contribution of the anomaly is, the smaller the degree of the anomaly is represented, and the smaller the probability that the attribute combination is an abnormal attribute combination is.

KPIs can be divided into two types: a base KPI and a derivative KPI. The base KPI has a characteristic that can be summed, for example, the duration of a stuck-at. Derived KPIs are calculated from the base KPIs, and common derived KPIs are ratioed from two base KPIs, e.g., play failure rate. The calculation of the abnormal contribution of the base KPI and the derived KPI will be described separately below.

For the basic KPI, the abnormal contribution of the leaf attribute combination may be a difference between a true value and a predicted value of the KPI at the first time of the leaf attribute combination, a ratio of the difference to a total difference between the true values and the predicted values of all the leaf attribute combinations, or a linear transformation of the difference or the ratio.

Alternatively, when the KPI with the abnormality is the base KPI, the abnormality contribution degree of the above leaf attribute combination may be calculated by:

wherein, AC is the abnormal contribution of the leaf attribute combination, a (e), f (e) are the actual value and the predicted value of the KPI of the leaf attribute combination at the first time, respectively, and a (leaf), f (leaf) are the sum of the actual value and the predicted value of the KPI of all the leaf attribute combinations at the first time, respectively.

The real value in the formula is obtained in practical application, the predicted value is obtained by predicting according to the real values of a plurality of second times before the first time, and the prediction can be realized by a machine model.

Optionally, the machine model is obtained by pre-training through the following steps:

clustering at least two leaf attribute combinations according to a sequence formed by actual values of KPIs respectively corresponding to each leaf attribute combination at least one second time to obtain at least one first cluster; a machine model is trained separately for each first class of clusters.

It can be understood that a leaf attribute combination corresponds to one real value of the KPI at a second time, so that the leaf attribute combination corresponds to a plurality of real values of the KPI at a plurality of second times, respectively, to form a sequence according to the sequence of the second times.

Clustering leaf attribute combinations based on the sequences can be clustering the leaf attribute combinations according to the similarity between the sequences, and clustering at least two leaf attribute combinations with larger sequence similarity into a first cluster. In one example, the similarity between sequences of any two leaf attribute combinations may be calculated, and two leaf attribute combinations having a similarity greater than or equal to a preset similarity threshold may be classified into the same first cluster. In another example, the clustering may also be implemented by using an existing time series clustering method, for example, UPGMA (unweighted pair group method using arithmetric clusters, group average clustering algorithm), DTW (dynamic time warping) similarity measure method.

After clustering to obtain one or more first-class clusters, combining a plurality of actual values of the KPIs respectively corresponding to at least two second times by using leaf attributes in the first-class clusters, and training to obtain a machine model of the first-class clusters. The training may be achieved through multiple iterations. In each iteration, firstly, the actual values of the KPIs corresponding to the last second time of the leaf attribute combination in the first cluster are used as sample values, the actual values of the KPIs corresponding to the leaf attribute combination in the first cluster at the rest second time are input into a machine model, and the training values of the KPIs corresponding to the last second time of the leaf attribute combination are obtained through prediction; then, based on a large number of sample values and training values obtained by a large number of sequences, calculating a loss value of the iteration of the current round through a loss function; finally, judging whether the loss value meets a preset condition or not, and if the loss value does not meet the preset condition, adjusting parameters of the machine model to perform the next iteration; and if the loss value meets the preset condition, ending the training, wherein the machine model at the moment is the machine model obtained by the training.

The above loss function may adopt a mean square error loss function, a cross entropy loss function, and the like commonly used in the prior art, which is not limited by the embodiments of the present disclosure. The loss value satisfying the preset condition includes at least one of: the loss value after the multiple iterations is not continuously smaller, and the loss value is less than or equal to a preset loss value function.

It is understood that the above training process is a training process for one machine model, and in practical applications, if multiple machine models are trained for a first cluster, different types of machine models may be trained using the same sequence.

After the training of the machine model is finished, the corresponding relation among the leaf attribute combinations, the first class clusters and the machine model can be recorded, so that the machine model of each leaf attribute combination can be obtained according to the corresponding relation in the subsequent KPI prediction.

For each of the first clusters, a machine model may be obtained by training the first cluster, or at least two machine models may be obtained by training the first cluster.

When the first cluster corresponds to a machine model, after the training of the machine model is completed, the KPI can be predicted by the following steps:

aiming at each leaf attribute combination, obtaining a machine model corresponding to a first cluster corresponding to the leaf attribute combination;

and inputting the actual values of the KPIs corresponding to the leaf attribute combinations at least one second time into the machine model, and predicting to obtain the predicted values of the KPIs corresponding to the leaf attribute combinations at the first time.

When the first cluster corresponds to at least two machine models, after the machine model training is completed, the KPI can be predicted by the following steps:

aiming at each leaf attribute combination, acquiring at least two machine models corresponding to a first cluster corresponding to the leaf attribute combination;

if the number of the machine models is at least two, inputting actual values of KPIs corresponding to the leaf attribute combinations at least one second time into the at least two machine models, and predicting to obtain at least two candidate predicted values of the KPIs corresponding to the leaf attribute combinations at the first time;

and determining the candidate predicted value closest to the actual value of the KPI of the leaf attribute combination at the first time as the predicted value of the KPI corresponding to the leaf attribute combination at the first time.

It will be appreciated that the accuracy of predictions using at least two machine models is higher than the accuracy of predictions using one machine model.

Alternatively, when one machine model is adopted, the machine model may be one of an ARIMA (integrated moving average autoregressive) model, an XGBOOST (extreme gradient boosting) model, and an LSTM (long short-term memory network) model; when at least two machine models are employed, the at least two machine models include: at least two of an ARIMA model, an XGBOOST model, and an LSTM model.

S102: and if the KPI is obtained by calculating the ratio of the first basic KPI to the second basic KPI, determining the abnormal contribution of each leaf attribute combination according to the difference between the predicted value and the real value of the first basic KPI of each leaf attribute combination at the first time and the difference between the predicted value and the real value of the second basic KPI of each leaf attribute combination at the first time.

Wherein, the first and second basic KPIs are both basic KPIs. When the KPI in which the abnormality occurs is a ratio of the first base KPI and the second base KPI, the abnormality contribution degree of the leaf attribute combination may be a difference between a first difference value and a second difference value, where the first difference value is a difference value between a predicted value and a true value of the first base KPI of the leaf attribute combination at the first time, and the second difference value is a difference value between a predicted value and a true value of the second base KPI of the leaf attribute combination at the first time. . The abnormal contribution of the leaf attribute combination may be a difference between a predicted value and a true value of a first base KPI of the leaf attribute combination at the first time and a second difference between a predicted value and a true value of a second base KPI of the leaf attribute combination at the first time.

Alternatively, when the KPI in which an abnormality occurs is the ratio of the first base KPI and the second base KPI, the abnormality contribution degree of the leaf attribute combination may be calculated by the following formula:

wherein AC is the abnormal contribution degree of the leaf attribute combination, a_M1(e)、f_M1(e) Respectively combining actual values, predicted values, f of the first base KPI at the first time for the leaf attributes_M1(leaf) is the sum of the predicted values of the first base KPI at the first time for all leaf attribute combinations, f_M2(leaf) is the sum of the predicted values of the second base KPI at the first time for all leaf attribute combinations, a_M2(e)、f_M2(e) And combining the real value and the predicted value of the second basic KPI at the first time for the leaf attribute respectively.

S103: and determining the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination.

Specifically, the abnormal contribution degrees of the non-leaf attribute combinations may be calculated layer by layer upward from the leaf attribute combinations, and the abnormal contribution degree of each non-leaf attribute combination is the sum of the abnormal contribution degrees of the attribute combinations of its child nodes. For example, as shown in fig. 3, when a is a network type Wi-Fi, B is a video resolution 720P, and C is a network provider ISP1, the anomaly contribution degree of the network type Wi-Fi is the sum of the anomaly contribution degree of a first attribute combination and the anomaly contribution degree of a second attribute combination, the first attribute combination is an attribute combination formed by the network type Wi-Fi and the video resolution 720P, and the second attribute combination is an attribute combination formed by the network type Wi-Fi and the network provider ISP 1.

Optionally, before determining the abnormal attribute combination according to the abnormal contribution degree, a leaf attribute combination which does not satisfy a preset condition may be deleted, where the leaf attribute combination which satisfies the preset condition includes: and at least one leaf attribute combination which is positioned at the front position and has the total abnormal contribution degree larger than or equal to and closest to the abnormal contribution degree threshold value is selected from at least two leaf attribute combinations which are arranged in descending order according to the abnormal contribution degree.

Specifically, firstly, at least two leaf attribute combinations are sorted in a descending order according to the abnormal contribution degree; then, starting from the first leaf attribute combination, adding each leaf attribute combination into a candidate leaf attribute combination set, and after each leaf attribute combination is added, calculating the sum of the contribution degrees of each leaf attribute combination in the candidate leaf attribute combination set to obtain a total contribution degree; then, when the total contribution degree is larger than or equal to a preset abnormal contribution degree threshold value, stopping adding the leaf attribute combination to the candidate leaf attribute combination set; and finally, deleting the leaf attribute combinations which are not added into the candidate leaf attribute combination set, and reserving the leaf attribute combinations added into the candidate leaf attribute combination set.

S104: and determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination, wherein the abnormal attribute combination is used for representing the attribute combination causing the KPI abnormality, and the attribute combination comprises at least one attribute.

Specifically, one or more attribute combinations with the highest abnormal contribution degree may be selected from the leaf attribute combinations and the non-leaf attribute combinations as the abnormal attribute combinations.

Optionally, determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination includes:

firstly, clustering at least two leaf attribute combinations according to the deviation degree between the predicted value and the true value of the KPI corresponding to each leaf attribute combination at the first time to obtain at least one second cluster, wherein the deviation degree is determined according to the difference between the predicted value and the true value and the ratio of the sum of the predicted value and the true value; then, in each second class cluster, adding attribute combinations of the same type into a set according to the order of the abnormal contribution degrees from large to small, wherein the attribute combinations are leaf attribute combinations in the second class cluster or non-leaf attribute combinations corresponding to the leaf attribute combinations in the second class cluster, and when each attribute combination is added, taking the set as a candidate attribute combination set of the second class cluster, and determining the fluctuation contribution degree of the candidate attribute combination set; and finally, determining an abnormal attribute combination according to the candidate attribute combination set with the maximum fluctuation contribution degree in each second-class cluster. Wherein the degree of deviation may be used to represent a difference between the predicted value and the true value that does not take into account the remaining leaf attribute combinations.

Alternatively, the degree of deviation may be calculated by the following formula:

wherein, DC is the deviation degree between the predicted value f (e) and the real value a (e) of the KPI corresponding to the leaf attribute combination at the first time respectively.

After the leaf attribute combinations are clustered based on the deviation degrees, the leaf attribute combinations with closer deviation degrees can be divided into the same second class cluster.

After the second-class clusters are obtained, at least one candidate attribute combination set of each second-class cluster is generated according to the leaf attribute combination in each second-class cluster or the non-leaf attribute combination corresponding to the leaf attribute combination, and the attribute combination in each candidate attribute combination set is the attribute combination of the same type. The types of attribute combination are single types of video resolution, network type, network provider and the like, or types of combinations of the types, including types of combinations of video resolution and network type, types of combinations of video resolution and network provider, types of combinations of network type and network provider, video resolution, types of networks and types of combinations of network provider. For example, for the video resolution, three attribute combinations corresponding to three video resolutions, such as 576P, 720P, and 1080P, may be added to the set in the order of increasing contribution degree to obtain three candidate attribute combination sets. For another example, for the video resolution and the network type, the attribute combination of 576P and Wi-Fi, the attribute combination of 720P and Wi-Fi, the attribute combination of 1080P and Wi-Fi, the attribute combination of 576P and mobile communication network, the attribute combination of 720P and mobile communication network, and the attribute combination of 1080P and mobile communication network may be added to the set in the order of contribution degree from large to small, to obtain six candidate attribute combination sets.

The fluctuation contribution degree of the candidate attribute combination set in the second-class cluster is used for representing the fluctuation degree between the actual value and the predicted value of the KPI corresponding to the first time of each candidate attribute combination in the candidate attribute combination set. It can be understood that the greater the fluctuation contribution degree, the greater the probability that the candidate attribute combination in the candidate attribute combination set is an abnormal attribute combination; the smaller the fluctuation contribution degree is, the smaller the probability that the candidate attribute combination representing the candidate attribute combination set is an abnormal attribute combination is.

Considering the ripple effect, for any combination of leaf attributes, one can obtain:

wherein, a (S) and f (S) are the sum of actual values and the sum of predicted values of KPIs of all leaf attribute combinations at a first time respectively, and f (e) and a (e) are the predicted values and the actual values of the KPIs of any leaf attribute combination at the first time respectively.

Performing variable substitution on the formula, and taking

The above formula can be converted into the following formula:

thereby obtaining

Ideally, the true value a (e) of the KPI of any leaf attribute combination e at the first time is equal to the expected value v (e) of the KPI of that leaf attribute combination e at the first time. Therefore, for a leaf attribute combination L corresponding to the attribute combination in the candidate attribute combination set₁The method comprises the following steps:

based on the expected value, the fluctuation contribution of the candidate attribute combination set can be calculated by the following formula:

wherein FC is the fluctuation contribution degree, L, of the candidate attribute combination set R₁All leaf attribute combinations corresponding to the attribute combination in R, a (L)₁)、f(L₁)、v(L₁) Are respectively L₁Actual, predicted, expected values, L, of KPIs at a first time₂Is a genus other than RAll combinations of leaf attributes, a (L), corresponding to the sexual combinations₂)、f(L₂) Are respectively L₂The actual value and the predicted value of the KPI at the first time, avg, are averaged, ac (R) is the sum of the abnormal contribution degrees of all attribute combinations in R, that is, the sum of the abnormal contribution degrees of the leaf attribute combinations corresponding to all attribute combinations in R, m is the number of attribute combinations in R, and c is a constant. It will be appreciated that all leaf attributes combine to form the set S, L₁And L₂In S. If the leaf attribute combination which does not meet the preset condition is deleted before S104, the leaf attribute combination remaining after deletion is included in S; if the leaf attribute combinations which do not meet the preset condition are not deleted before the step S104, the step S comprises all the leaf attribute combinations which are not deleted to form a set S.

Based on the fluctuation contribution degrees, one or more candidate attribute combinations with the largest fluctuation contribution degree in all the second-class clusters can be used as the abnormal attribute combination, and the candidate attribute combination with the largest fluctuation contribution degree in each second-class cluster can also be used as the abnormal attribute combination.

Optionally, the anomaly attribute combination may also be determined by combining the anomaly contribution degree and the fluctuation contribution degree through the following process: firstly, weighting the maximum fluctuation contribution degree in the second cluster and the total abnormal contribution degree of all leaf attribute combinations in the second cluster aiming at each second cluster to obtain the comprehensive abnormal contribution degree of the second cluster; and determining an abnormal attribute combination from the candidate attribute combination set with the maximum fluctuation contribution degree of at least one second-class cluster according to the comprehensive abnormal contribution degree.

The total abnormal contribution degree of all leaf attribute combinations in the second type cluster is the sum of the abnormal contribution degrees of all leaf attribute combinations in the second type cluster, and the weighting of the total abnormal contribution degree and the maximum fluctuation contribution degree may be summation according to a preset weight.

It can be understood that the embodiment of the disclosure may determine the abnormal attribute combination by combining the abnormal contribution degree and the fluctuation contribution degree, not only considering the difference between the predicted value and the true value, but also considering the fluctuation degree between the predicted value and the true value, so that the accuracy of the determined abnormal attribute combination is higher.

Optionally, for each second-class cluster, before adding the non-leaf attribute combination and the leaf attribute combination to the candidate attribute combination set of the second-class cluster in the order from the largest abnormality contribution degree to the smallest abnormality contribution degree, part of the second-class clusters may be further deleted through the following process to determine an abnormality attribute combination from the remaining second-class clusters: firstly, determining the fluctuation direction of each second cluster, wherein the fluctuation direction is used for indicating that the real KPI of the second cluster is greater than or equal to or less than the predicted KPI of the second cluster, the real KPI of the second cluster is the sum of the real values of the KPIs of all leaf attribute combinations in the second cluster at the first time, and the predicted KPI of the second cluster is the sum of the predicted values of the KPIs of all the leaf attribute combinations in the second cluster at the first time; then, determining the target fluctuation direction of the KPI; and finally, deleting the second cluster with the fluctuation direction inconsistent with the target fluctuation direction.

And when the target fluctuation direction does not distinguish the attributes, the actual value of the KPI is greater than or equal to or less than the predicted value of the KPI.

It is understood that when the fluctuation direction of a second cluster is not consistent with the target fluctuation direction, the attribute combination in the second cluster is unlikely to be an abnormal attribute combination. The second cluster with the inconsistent fluctuation direction and the target direction can be deleted, so that the candidate attribute combination during the determination of the abnormal attribute combination is reduced, the calculation amount of the fluctuation contribution degree is reduced, and the speed of determining the abnormal attribute combination is improved.

Corresponding to the root cause analysis method for video domain performance parameter abnormality in the foregoing embodiment, fig. 4 is a block diagram of a root cause analysis device for video domain performance parameter abnormality according to an embodiment of the present disclosure. For ease of illustration, only portions that are relevant to embodiments of the present disclosure are shown. Referring to fig. 4, the root cause analysis device 200 for video domain performance parameter anomaly includes: a first anomaly contribution determining module 201, a second anomaly contribution determining module 202, a third anomaly contribution determining module 203 and an anomaly attribute combination determining module 204.

The first abnormal contribution degree determining module 201 is configured to, when a key performance indicator KPI of video playing quality is abnormal at a first time, determine, for at least two leaf attribute combinations, an abnormal contribution degree of each leaf attribute combination according to a difference between a predicted value and a true value of the KPI corresponding to each leaf attribute combination at the first time if the KPI is a basic KPI;

a second abnormal contribution determining module 202, configured to determine, if the KPI is calculated from a ratio of the first base KPI to the second base KPI, an abnormal contribution of each leaf attribute combination according to a difference between a predicted value and a true value of the first base KPI of each leaf attribute combination at the first time and a difference between a predicted value and a true value of the second base KPI of each leaf attribute combination at the first time;

a third abnormal contribution degree determining module 203, configured to determine an abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination;

an abnormal attribute combination determining module 204, configured to determine an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination, where the abnormal attribute combination is used to represent an attribute combination causing a KPI abnormality, and the attribute combination includes at least one attribute.

Optionally, the apparatus further comprises:

the first machine model acquisition module is used for acquiring a machine model corresponding to the first cluster corresponding to the leaf attribute combination aiming at each leaf attribute combination;

and the first KPI prediction module is used for inputting the actual values of the KPIs corresponding to the leaf attribute combinations at least one second time into the machine model, and predicting to obtain the predicted values of the KPIs corresponding to the leaf attribute combinations at the first time.

Optionally, the apparatus further comprises:

the second machine model acquisition module is used for acquiring at least two machine models corresponding to the first cluster corresponding to the leaf attribute combination aiming at each leaf attribute combination;

the second KPI prediction module is used for inputting the actual values of the KPIs corresponding to the leaf attribute combinations at least at one second time into at least two machine models if the number of the machine models is at least two, and predicting to obtain at least two candidate predicted values of the KPIs corresponding to the leaf attribute combinations at the first time;

and the KPI predicted value determining module is used for determining the candidate predicted value which is closest to the actual value of the KPI of the leaf attribute combination at the first time as the predicted value of the KPI corresponding to the leaf attribute combination at the first time.

Optionally, the at least two machine models comprise: at least two of a differential integration moving average autoregressive (ARIMA) model, an extreme gradient boost (XGB OST) model and a long-short term memory network (LSTM) model.

Optionally, the machine model is pre-trained by:

the first clustering module is used for clustering at least two leaf attribute combinations according to a sequence formed by actual values of KPIs respectively corresponding to each leaf attribute combination at least one second time to obtain at least one first cluster;

and the machine model training module is used for respectively training the machine model aiming at each first cluster.

Optionally, the abnormal attribute combination determining module 204 is further configured to:

clustering at least two leaf attribute combinations according to the deviation degree between the predicted value and the true value of the KPI corresponding to each leaf attribute combination at the first time to obtain at least one second cluster, wherein the deviation degree is determined according to the difference between the predicted value and the true value and the ratio of the sum of the predicted value and the true value;

in each second type cluster, adding attribute combinations of the same type into a set according to the order of the abnormal contribution degrees from large to small, wherein the attribute combinations are leaf attribute combinations in the second type cluster or non-leaf attribute combinations corresponding to the leaf attribute combinations in the second type cluster;

and determining an abnormal attribute combination according to the candidate attribute combination set with the maximum fluctuation contribution degree in each second-class cluster.

weighting the maximum fluctuation contribution degree in the second cluster and the total abnormal contribution degree of all leaf attribute combinations in the second cluster aiming at each second cluster to obtain the comprehensive abnormal contribution degree of the second cluster;

and determining an abnormal attribute combination from the candidate attribute combination set with the maximum fluctuation contribution degree of at least one second-class cluster according to the comprehensive abnormal contribution degree.

Optionally, the first abnormal contribution determining module 201 is further configured to:

calculating the abnormal contribution degree of the leaf attribute combination by the following formula:

Optionally, the second abnormal contribution determining module 201 is further configured to:

where AC is the abnormal contribution of the leaf attribute combination, a_M1(e)、f_M1(e) Respectively combining the true value and the predicted value f of the first basic KPI at the first time for the leaf attribute_M1(leaf) is the sum of the predicted values of the first base KPI at a first time for all leaf attribute combinations, f_M2(leaf) is allSum of predicted values of second base KPIs at a first time of leaf attribute combination, a_M2(e)、f_M2(e) And respectively combining the true value and the predicted value of the second basic KPI at the first time for the leaf attribute.

calculating the fluctuation contribution degree of the candidate attribute combination set by the following formula:

or

Wherein FC is the fluctuation contribution degree of the candidate attribute combination set, a (L)₁)、f(L₁)、v(L₁) The actual value, the predicted value and the expected value of the KPI at the first time, a (L) are respectively leaf attribute combinations corresponding to the attribute combinations in the candidate attribute combination set₂)、f(L₂) The actual values and the predicted values of the KPIs at the first time for the leaf attribute combinations corresponding to the attribute combinations outside the candidate attribute combination set are respectively, avg is averaging, AC (R) is the sum of abnormal contribution degrees of all the attribute combinations in the candidate attribute combination set, m is the number of the attribute combinations in the candidate attribute combination set, c is a constant, and a, (S) and f (S) are respectively the sum of the actual values and the predicted values of the KPIs at the first time for all the leaf attribute combinations.

Optionally, the apparatus further comprises:

the leaf attribute combination deleting module is used for deleting the leaf attribute combination which does not meet the preset condition, and the leaf attribute combination which meets the preset condition comprises the following steps: and at least one leaf attribute combination which is positioned at the front position and has the total abnormal contribution degree larger than or equal to and closest to the abnormal contribution degree threshold value is selected from at least two leaf attribute combinations which are arranged in descending order according to the abnormal contribution degree.

Optionally, the apparatus further comprises:

a fluctuation direction determining module, configured to determine a fluctuation direction of each second cluster, where the fluctuation direction is used to indicate that a true KPI of the second cluster is greater than or equal to or less than a predicted KPI of the second cluster, where the true KPI of the second cluster is a sum of true values of KPIs of each leaf attribute combination in the second cluster at the first time, and the predicted KPI of the second cluster is a sum of predicted values of KPIs of each leaf attribute combination in the second cluster at the first time;

the target fluctuation direction determining module is used for determining the target fluctuation direction of the KPI;

and the second cluster deleting module is used for deleting the second cluster with the fluctuation direction inconsistent with the target fluctuation direction.

The root cause analysis apparatus for performance parameter abnormality in the video field provided in this embodiment may be used to implement the technical solution of the method embodiment shown in fig. 2, and the implementation principle and the technical effect are similar, which are not described herein again.

Fig. 5 is a block diagram illustrating an electronic device 600 according to an exemplary embodiment of the disclosure. The electronic device 600 comprises a memory 602 and at least one processor 601;

wherein the memory 602 stores computer-executable instructions;

the at least one processor 601 executes computer-executable instructions stored by the memory 602 to cause the electronic device to implement the root cause analysis method for video domain performance parameter anomalies in fig. 2, as described above;

in addition, the electronic device 600 may further include a receiver 603 and a transmitter 604, where the receiver 603 is configured to receive information from the remaining apparatuses or devices and forward the information to the processor 601, and the transmitter 604 is configured to transmit the information to the remaining apparatuses or devices.

The embodiment of the disclosure also provides a computer program, and the computer program is used for implementing the root cause analysis method for the performance parameter abnormity in the video field.

The embodiment of the present disclosure further provides a computer-readable storage medium, in which computer-executable instructions are stored, and when a processor executes the computer-executable instructions, a computing device is enabled to implement the root cause analysis method for performance parameter abnormality in the video field.

In a first example of the first aspect, the present disclosure provides a root cause analysis method for video domain performance parameter abnormality, including:

In a second example of the first aspect, based on the first example of the first aspect, the method further comprises:

aiming at each leaf attribute combination, acquiring a machine model corresponding to a first cluster corresponding to the leaf attribute combination;

In a third example of the first aspect, based on the second example of the first aspect, the method further comprises:

if the number of the machine models is at least two, inputting real values of the KPIs corresponding to the leaf attribute combinations at least at one second time into the at least two machine models, and predicting to obtain at least two candidate predicted values of the KPIs corresponding to the leaf attribute combinations at the first time;

Based on the third example of the first aspect, in a fourth example of the first aspect, the at least two machine models comprise: at least two of a differential integration moving average autoregressive (ARIMA) model, an extreme gradient boost (XGB OST) model and a long-short term memory network (LSTM) model.

In a fifth example of the first aspect, based on any one of the second to fourth examples of the first aspect, the machine model is pre-trained by:

clustering the at least two leaf attribute combinations according to a sequence formed by actual values of the KPIs respectively corresponding to each leaf attribute combination at least one second time to obtain at least one first cluster;

a machine model is trained separately for each of the first clusters.

In a sixth example of the first aspect, based on any one of the first to fourth examples of the first aspect, the determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination includes:

clustering the at least two leaf attribute combinations according to the deviation degree between the predicted value and the true value of the KPI corresponding to each leaf attribute combination at the first time respectively to obtain at least one second cluster, wherein the deviation degree is determined according to the difference between the predicted value and the true value and the ratio of the sum of the predicted value and the true value;

in each second type cluster, adding attribute combinations of the same type into a set according to the sequence of abnormal contribution degrees from large to small, wherein the attribute combinations are leaf attribute combinations in the second type cluster or non-leaf attribute combinations corresponding to the leaf attribute combinations in the second type cluster, and when each attribute combination is added, taking the set as a candidate attribute combination set of the second type cluster, and determining the fluctuation contribution degree of the candidate attribute combination set;

Based on the sixth example of the first aspect, in a seventh example of the first aspect, the determining an abnormal attribute combination according to the candidate attribute combination set with the largest fluctuation contribution degree in each of the second-class clusters includes:

for each second cluster, weighting the maximum fluctuation contribution degree in the second cluster and the total abnormal contribution degree of all leaf attribute combinations in the second cluster to obtain the comprehensive abnormal contribution degree of the second cluster;

and determining an abnormal attribute combination from the candidate attribute combination set with the maximum fluctuation contribution degree of the at least one second-class cluster according to the comprehensive abnormal contribution degree.

Based on the first example of the first aspect, in an eighth example of the first aspect, the determining, according to a difference between a predicted value and a true value of the KPI corresponding to each leaf attribute combination at the first time, an abnormal contribution degree of each leaf attribute combination includes:

wherein AC is the abnormal contribution of the leaf attribute combination, a (e), f (e) are the true value and predicted value of the KPI at the first time of the leaf attribute combination, respectively, and a (leaf), f (leaf) are the sum of the true value and predicted value of the KPI at the first time of all leaf attribute combinations, respectively.

In a ninth example of the first aspect, based on the first example of the first aspect, the determining the abnormal contribution degree of each leaf attribute combination according to a difference between a predicted value and a true value of the first base KPI at the first time of each leaf attribute combination and a difference between a predicted value and a true value of the second base KPI at the first time of each leaf attribute combination includes:

wherein AC is the abnormal contribution degree of the leaf attribute combination, a_M1(e)、f_M1(e) Combining the true value, predicted value, f of the first base KPI at the first time for the leaf attribute, respectively_M1(leaf) is the sum of the predicted values of the first base KPI at the first time for all leaf attribute combinations, f_M2(leaf) is the sum of the predicted values of the second base KPI at the first time for all leaf attribute combinations, a_M2(e)、f_M2(e) Combining the leaf attributes at the first time respectivelyThe true value and the predicted value of the second base KPI.

In a tenth example of the first aspect, based on the sixth example of the first aspect, the determining the fluctuation contribution of the candidate property combination set includes:

or

Wherein FC is the fluctuation contribution of the candidate attribute combination set, a (L)₁)、f(L₁)、v(L₁) Respectively corresponding to the leaf attribute combination in the candidate attribute combination set, and the actual value, the predicted value and the expected value of the KPI at the first time, a (L)₂)、f(L₂) The actual values and the predicted values of the KPIs of the leaf attribute combinations corresponding to the attribute combinations outside the candidate attribute combination set at the first time respectively, avg is averaging, ac (r) is the sum of abnormal contribution degrees of all the attribute combinations in the candidate attribute combination set, m is the number of the attribute combinations in the candidate attribute combination set, c is a constant, and a,(s) and f(s) are the sum of the actual values and the sum of the predicted values of the KPIs of all the leaf attribute combinations at the first time respectively.

In an eleventh example of the first aspect, before determining the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination, the method further includes:

deleting leaf attribute combinations which do not meet preset conditions, wherein the leaf attribute combinations which meet the preset conditions comprise: and at least one leaf attribute combination which is positioned at the front position and has the total abnormal contribution degree larger than or equal to and closest to the abnormal contribution degree threshold value is selected from the at least two leaf attribute combinations which are arranged in descending order according to the abnormal contribution degree.

In a twelfth example of the first aspect, before, for each of the second-class clusters, adding the non-leaf attribute combination and the leaf attribute combination to the candidate attribute combination set of the second-class cluster in an order from a large abnormality contribution degree to a small abnormality contribution degree, the method further includes:

determining a fluctuation direction of each second cluster, wherein the fluctuation direction is used for indicating that a real KPI of the second cluster is greater than or equal to or less than a predicted KPI of the second cluster, the real KPI of the second cluster is the sum of real values of the KPIs of each leaf attribute combination in the second cluster at the first time, and the predicted KPI of the second cluster is the sum of predicted values of the KPIs of each leaf attribute combination in the second cluster at the first time;

determining a target fluctuation direction of the KPI;

and deleting the second cluster with the fluctuation direction inconsistent with the target fluctuation direction.

In a first example of the second aspect, there is provided a root cause analysis apparatus for video domain performance parameter abnormality, including:

In a second example of the second aspect, based on the first example of the second aspect, the apparatus further comprises:

a first machine model obtaining module, configured to obtain, for each leaf attribute combination, one machine model corresponding to a first cluster corresponding to the leaf attribute combination;

and the first KPI prediction module is used for inputting the actual values of the KPIs corresponding to the leaf attribute combination at least one second time into the machine model, and predicting to obtain the predicted values of the KPIs corresponding to the leaf attribute combination at the first time.

In a third example of the second aspect, based on the second example of the second aspect, the apparatus further comprises:

a second machine model obtaining module, configured to obtain, for each leaf attribute combination, at least two machine models corresponding to the first cluster corresponding to the leaf attribute combination;

a second KPI prediction module, configured to, if the number of the machine models is at least two, input real values of the KPIs corresponding to the leaf attribute combination at least one second time into the at least two machine models, and predict to obtain at least two candidate prediction values of the KPIs corresponding to the leaf attribute combination at the first time;

and a KPI predicted value determining module, configured to determine a candidate predicted value closest to a true value of the KPI of the leaf attribute combination at the first time as a predicted value of the KPI corresponding to the leaf attribute combination at the first time.

In a fourth example of the second aspect, based on the third example of the second aspect, the at least two machine models comprise: at least two of a differential integration moving average autoregressive (ARIMA) model, an extreme gradient boost (XGB OST) model and a long-short term memory network (LSTM) model.

In a fifth example of the second aspect, based on any one of the second to fourth examples of the second aspect, the machine model is pre-trained by:

the first clustering module is used for clustering the at least two leaf attribute combinations according to a sequence formed by actual values of the KPIs respectively corresponding to the leaf attribute combinations at least at one second time to obtain at least one first cluster;

and the machine model training module is used for respectively training a machine model for each first cluster.

In a sixth example of the second aspect, based on any one of the first to fourth examples of the second aspect, the abnormal attribute combination determination module is further configured to:

In a seventh example of the second aspect, based on the sixth example of the second aspect, the abnormal attribute combination determination module is further configured to:

In an eighth example of the second aspect, based on the first example of the second aspect, the first abnormality contribution degree determination module is further configured to:

wherein AC is the abnormal contribution of the leaf attribute combination, a (e), f (e) are the actual value and the predicted value of the KPI of the leaf attribute combination at the first time, respectively, and a (leaf), f (leaf) are the sum of the actual value and the predicted value of the KPI of all the leaf attribute combinations at the first time, respectively.

In a ninth example of the second aspect, based on the first example of the second aspect, the second abnormality contribution degree determination module is further configured to:

wherein AC is the abnormal contribution degree of the leaf attribute combination, a_M1(e)、f_M1(e) Combining the true value, predicted value, f of the first base KPI at the first time for the leaf attribute, respectively_M1(leaf) is the sum of the predicted values of the first base KPI at the first time for all leaf attribute combinations, f_M2(leaf) is the sum of the predicted values of the second base KPI at the first time for all leaf attribute combinations, a_M2(e)、f_M2(e) And combining the real value and the predicted value of the second basic KPI at the first time for the leaf attribute respectively.

In a tenth example of the second aspect, based on the sixth example of the second aspect, the abnormal attribute combination determination module is further configured to:

or

Wherein FC is the fluctuation contribution of the candidate attribute combination set, a (L)₁)、f(L₁)、v(L₁) The actual values of the KPIs at the first time, the leaf attribute combinations corresponding to the attribute combinations in the candidate attribute combination set,Predicted value, expected value, a (L)₂)、f(L₂) The actual values and the predicted values of the KPIs of the leaf attribute combinations corresponding to the attribute combinations outside the candidate attribute combination set at the first time respectively, avg is averaging, ac (r) is the sum of abnormal contribution degrees of all the attribute combinations in the candidate attribute combination set, m is the number of the attribute combinations in the candidate attribute combination set, c is a constant, and a,(s) and f(s) are the sum of the actual values and the sum of the predicted values of the KPIs of all the leaf attribute combinations at the first time respectively.

In an eleventh example of the second aspect, based on any one of the first to fourth examples of the second aspect, the apparatus further includes:

a leaf attribute combination deleting module, configured to delete a leaf attribute combination that does not satisfy a preset condition, where the leaf attribute combination that satisfies the preset condition includes: and at least one leaf attribute combination which is positioned at the front position and has the total abnormal contribution degree larger than or equal to and closest to the abnormal contribution degree threshold value is selected from the at least two leaf attribute combinations which are arranged in descending order according to the abnormal contribution degree.

In a twelfth example of the second aspect, based on the sixth example of the second aspect, the apparatus further comprises:

a fluctuation direction determining module, configured to determine a fluctuation direction of each of the second clusters, where the fluctuation direction is used to indicate that a true KPI of each of the second clusters is greater than or equal to or less than a predicted KPI of each of the second clusters, where the true KPI of each of the second clusters is a sum of true values of the KPIs of each of the combinations of leaf attributes in the second clusters at the first time, and the predicted KPI of each of the second clusters is a sum of predicted values of the KPIs of each of the combinations of leaf attributes in the second clusters at the first time;

determining a target fluctuation direction of the KPI;

In a third aspect, according to one or more embodiments of the present disclosure, there is provided an electronic device including: at least one processor and memory;

the memory stores computer-executable instructions;

the at least one processor executes computer-executable instructions stored by the memory to cause the electronic device to implement the method of the first aspect as described above.

In a fourth aspect, according to one or more embodiments of the present disclosure, there is provided a computer-readable storage medium having stored therein computer-executable instructions that, when executed by a processor, cause a computing device to implement the method of the first aspect as described above.

In a fifth aspect, according to one or more embodiments of the present disclosure, there is provided a computer program for implementing the method of the first aspect as described above.

The foregoing description is only exemplary of the preferred embodiments of the disclosure and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the disclosure herein is not limited to the particular combination of features described above, but also encompasses other embodiments in which any combination of the features described above or their equivalents does not depart from the spirit of the disclosure. For example, the above features and (but not limited to) the features disclosed in this disclosure having similar functions are replaced with each other to form the technical solution.

Further, while operations are depicted in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Likewise, while several specific implementation details are included in the above discussion, these should not be construed as limitations on the scope of the disclosure. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination.

Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.

Claims

1. A root cause analysis method for performance parameter abnormity in the video field is characterized by comprising the following steps:

2. The method of claim 1, further comprising:

3. The method of claim 2, further comprising:

4. A method according to claim 2 or 3, characterized in that the machine model is pre-trained by the following steps:

a machine model is trained separately for each of the first clusters.

5. The method according to any one of claims 1 to 3, wherein determining an abnormal attribute combination from the leaf attribute combination and the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination and the abnormal contribution degree of the non-leaf attribute combination comprises:

clustering the at least two leaf attribute combinations to obtain at least one second cluster according to the deviation degree between the predicted value and the real value of the KPI corresponding to each leaf attribute combination at the first time, wherein the deviation degree is determined according to the difference between the predicted value and the real value and the ratio of the sum of the predicted value and the real value;

6. The method according to claim 5, wherein the determining an abnormal attribute combination according to the candidate attribute combination set with the largest fluctuation contribution degree in each of the second-class clusters comprises:

7. The method according to claim 1, wherein the determining the abnormal contribution of each of the leaf attribute combinations according to a difference between a predicted value and a true value of the KPI corresponding to each of the leaf attribute combinations at the first time comprises:

8. The method of claim 1, wherein determining the anomalous contribution for each of the combinations of leaf attributes based on a difference between a predicted value and a true value for the first base KPI for each of the combinations of leaf attributes at the first time and a difference between a predicted value and a true value for the second base KPI for each of the combinations of leaf attributes at the first time comprises:

9. The method of claim 5, wherein determining the fluctuation contribution of the candidate set of attribute combinations comprises:

wherein FC is the fluctuation contribution of the candidate attribute combination set, a (L)₁)、f(L₁)、v(L₁) Respectively corresponding to the attribute combinations in the candidate attribute combination set, and the actual value, the predicted value and the expected value of the KPI at the first time, a (L)₂)、f(L₂) The actual values and the predicted values of the KPIs of the leaf attribute combinations corresponding to the attribute combinations outside the candidate attribute combination set at the first time respectively, avg is averaging, ac (r) is the sum of abnormal contribution degrees of all the attribute combinations in the candidate attribute combination set, m is the number of the attribute combinations in the candidate attribute combination set, c is a constant, and a,(s) and f(s) are the sum of the actual values and the sum of the predicted values of the KPIs of all the leaf attribute combinations at the first time respectively.

10. The method according to any one of claims 1 to 3, wherein before determining the abnormal contribution degree of the non-leaf attribute combination according to the abnormal contribution degree of the leaf attribute combination, the method further comprises:

11. The method according to claim 5, wherein before adding the non-leaf attribute combination and leaf attribute combination to the candidate attribute combination set of the second class cluster in order of the abnormal contribution degree from large to small for each of the second class clusters, further comprising:

determining a fluctuation direction of each second cluster, wherein the fluctuation direction is used for indicating that a real KPI of the second cluster is greater than or equal to or less than a predicted KPI of the second cluster, the real KPI of the second cluster is a sum of real values of the KPIs of each leaf attribute combination in the second cluster at the first time, and the predicted KPI of the second cluster is a sum of predicted values of the KPIs of each leaf attribute combination in the second cluster at the first time;

determining a target fluctuation direction of the KPI;

12. A root cause analysis device for video domain performance parameter abnormity is characterized by comprising:

13. An electronic device, comprising: at least one processor and a memory;

the memory stores computer-executable instructions;

the at least one processor executing the computer-executable instructions stored by the memory causes the electronic device to implement the method of any of claims 1-11.

14. A computer-readable storage medium having computer-executable instructions stored thereon, which, when executed by a processor, cause a computing device to implement the method of any one of claims 1 to 11.

15. A computer program for implementing the method according to any one of claims 1 to 11.