Network degradation root cause localization method, system, electronic device, and storage medium

Abstract

The application relates to the technical field of mobile communication, and provides a network degradation root cause positioning method, a network degradation root cause positioning system, electronic equipment and a storage medium. The network degradation root cause positioning method comprises the following steps: according to network degradation data, a plurality of groups of KPIs (Key Performance Indicators) associated with the network degradation data are selected from a KPI set; the total confidence of each group of KPIs is calculated; it is judged whether the total confidence of a target group of KPIs with the maximum total confidence is greater than a threshold value; if yes, according to the corresponding relationship between KPI classes and root causes, the root cause corresponding to the KPI class to which the target group of KPIs belongs is obtained as the target root cause of the network degradation data; and if no, the remaining KPIs except the target group of KPIs in the plurality of groups of KPIs are obtained, and the target root cause is determined according to the deviation degree of the remaining KPIs and the root cause. The application can comprehensively and accurately position the network degradation root cause according to different network conditions.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and more specifically, to a method, system, electronic device, and storage medium for locating the root causes of network degradation. Background Technology

[0002] With the development of communication networks, complex networks where 3G (3rd Generation Mobile Communication Technology), 4G (4th Generation Mobile Communication Technology), 5G NSA (5th Generation Mobile Communication Technology Non-Standalone), and 5G SA (5th Generation Mobile Communication Technology Standalone) coexist are becoming increasingly difficult to maintain manually. However, AI (Artificial Intelligence) technology, utilizing big data, can enable automatic network optimization, comprehensively improving the efficiency of integrated cloud-network operations.

[0003] During network optimization, it is necessary to detect poor-quality networks and pinpoint the root causes of network degradation. Currently, methods for locating the root causes of network degradation include: analyzing individual indicators one by one and combining this with the experience of engineers; or using logical methods to make machine judgments based on the explicit characteristics of the indicators.

[0004] Current methods for locating the root causes of network degradation often identify only one cause, and in many cases, they fail to pinpoint the root cause at all.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method, system, electronic device and storage medium for locating the root causes of network degradation, which can achieve comprehensive and accurate location of the root causes of network degradation according to different network conditions.

[0007] One aspect of the present invention provides a method for locating the root cause of network degradation, comprising: selecting multiple sets of KPIs associated with the network degradation data from a set of key performance indicators (KPIs); calculating the full confidence score of each set of KPIs, and determining whether the full confidence score of the target set of KPIs with the highest full confidence score is greater than a threshold; if so, obtaining the root cause corresponding to the KPI class to which the target set of KPIs belongs, based on the correspondence between KPI class and root cause, and using it as the target root cause of the network degradation data; if not, obtaining the remaining KPIs excluding the target set of KPIs from the multiple sets of KPIs, and determining the target root cause based on the deviation between the remaining KPIs and the root cause.

[0008] In some embodiments, each KPI class has multiple preset interval values, and the correspondence is the correspondence between the interval value combination of the KPI class and the root cause combination of the root cause; when using network degradation data, the index value of each selected KPI is also obtained; when obtaining the root cause corresponding to the KPI class to which the target group KPI belongs, the corresponding interval value combination is determined from the correspondence and the corresponding root cause combination is obtained based on the target KPI and its index value in the target group KPI.

[0009] In some embodiments, the correspondence includes a multi-level transition matrix; in the multi-level transition matrix, the first-level matrix uses the range values ​​of two first-level KPI classes in the KPI class as row and column identifiers, the Nth-level matrix uses the range value of one Nth-level KPI class in the KPI class and the element of the (N-1)th-level matrix as row and column identifiers, N≥2; each element is a root cause combination consisting of one or more root causes, and each element can be associated with the KPI class corresponding to its row identifier and / or column identifier.

[0010] In some embodiments, obtaining the root cause corresponding to the KPI class to which the target group KPIs belong includes: obtaining the target KPI class to which each target KPI in the target group KPIs belongs; obtaining an initial matrix with at least one current target KPI class interval value as row identifier and / or column identifier according to the transition order of the multi-level transition matrix; pruning the initial matrix by removing row and column identifiers and elements related to other KPI classes besides the current target KPI class from the initial matrix; determining whether there are any remaining target KPI classes not included in the initial matrix; if so, obtaining a target matrix with the interval value of the remaining target KPI class and the elements in the pruned initial matrix as row and column identifiers according to the transition order; if not, using the pruned initial matrix as the target matrix; and determining a target interval value combination based on the interval value of the target KPI class to which the indicator value of each target KPI falls, and obtaining the root cause corresponding to the target interval value combination from the target matrix.

[0011] In some embodiments, when obtaining network degradation data, the index value of each selected KPI is also obtained; the step of determining the deviation between the remaining KPI and the root cause includes: fitting the remaining KPI to the root cause to obtain a fitting value of each remaining KPI relative to each root cause; obtaining the deviation between each remaining KPI and each root cause according to the deviation calculation formula d = (yr) / r; in the deviation calculation formula, d is the deviation between a remaining KPI and a root cause, y is the fitting value of the remaining KPI relative to the root cause, and r is the index value of the remaining KPI.

[0012] In some embodiments, obtaining the fitted value of each remaining KPI relative to each root cause includes: obtaining the root cause value of each root cause, where the root cause value of a root cause is the index value corresponding to the KPI with the highest confidence in the KPI class matching the root cause; calculating the fitted value of each remaining KPI relative to each root cause according to the fitting formula y=(k+α(x)δ)·x+(m+α(x)γ); wherein the fitting formula is a piecewise linear function, where x is the root cause value of a root cause, k is the slope of the current piecewise linear part corresponding to the root cause value of the root cause, δ is the change in the slope of the current piecewise linear part, α(x) is the coefficient of the current piecewise linear part, m is the bias value of the current piecewise linear part, γ is the boundary value of the current piecewise linear part, and y is the fitted value of a remaining KPI relative to the root cause.

[0013] In some embodiments, determining the target root cause includes: for the remaining KPIs whose index values ​​are inversely proportional to network performance, selecting the root cause corresponding to the maximum deviation; for the remaining KPIs whose index values ​​are directly proportional to network performance, selecting the root cause corresponding to the minimum deviation; and using the selected root cause as the target root cause.

[0014] In some embodiments, the step of selecting multiple sets of KPIs associated with the network degradation data from the set of Key Performance Indicators (KPIs) includes: obtaining the current degradation KPI of the network degradation data, target sample data containing the current KPI corresponding to the current degradation KPI, and the KPI set, wherein the target sample data includes target negative samples containing the current degradation KPI; combining the KPI set to obtain multiple itemsets; calculating the support of each itemset, wherein the support of an itemset is the probability of the occurrence of the target negative sample containing that itemset in the target sample data; and based on the support of each itemset, mining frequent itemsets from the multiple itemsets by constructing a frequentity tree, as multiple sets of KPIs associated with the network degradation data.

[0015] In some embodiments, calculating the full confidence of each group of KPIs includes: calculating the confidence associated with each group of KPIs, wherein the confidence associated with a group of KPIs includes a first confidence (B→A) and a second confidence (A→B), wherein event A is the occurrence of the target negative sample in the target sample data, and event B is the occurrence of the group of KPIs in the target sample data; and calculating the full confidence of each group of KPIs according to the full confidence calculation formula all_confidence(A, B) = min{confidence(A→B), confidence(B→A)}.

[0016] In some embodiments, before calculating the confidence level associated with each group of KPIs, the method further includes: calculating the lift of each group of KPIs, wherein the lift of a group of KPIs is the ratio of the probability of the group of KPIs occurring simultaneously under the condition that the target negative sample occurs in the target sample data to the probability of the group of KPIs occurring; and filtering out several groups of KPIs whose lift results are not positively correlated from multiple groups of KPIs.

[0017] In some embodiments, before determining the network degradation data, the method further includes: obtaining the network degradation data from the alarm data of the network degradation alarm in response to a network degradation alarm; after determining the target root cause of the network degradation data, the method further includes: outputting the network degradation alarm and the target root cause.

[0018] Another aspect of the present invention provides a network degradation root cause localization system, comprising: an associated KPI screening module configured to screen multiple sets of KPIs associated with the network degradation data from a set of key performance indicators (KPIs); a full confidence calculation module configured to calculate the full confidence of each set of KPIs and determine whether the full confidence of a target group of KPIs with the highest full confidence is greater than a threshold; a first root cause determination module configured to, when the full confidence of the target group of KPIs is greater than the threshold, obtain the root cause corresponding to the KPI class to which the target group of KPIs belongs, based on the correspondence between KPI classes and root causes, and use it as the target root cause of the network degradation data; and a second root cause determination module configured to, when the full confidence of the target group of KPIs is not greater than the threshold, obtain the remaining KPIs excluding the target group of KPIs from the multiple sets of KPIs, and determine the target root cause based on the deviation between the remaining KPIs and the root cause.

[0019] Another aspect of the present invention provides an electronic device, comprising: a processor; a memory storing executable instructions; wherein, when the executable instructions are executed by the processor, they implement the network degradation root cause localization method described in any of the above embodiments.

[0020] Another aspect of the present invention provides a computer-readable storage medium for storing a program that, when executed by a processor, implements the network degradation root cause localization method described in any of the above embodiments.

[0021] The beneficial effects of this invention compared to the prior art include at least the following:

[0022] This invention mines associated KPIs from network degradation data, determines whether a target KPI has explicit degradation characteristics based on maximum full confidence, and obtains the target root cause corresponding to the target KPI based on the correspondence between the KPI set and the root cause set when the target KPI has explicit degradation characteristics. When the target KPI does not have explicit degradation characteristics, the target root cause is determined based on the deviation between the remaining KPIs and the root cause. Thus, it achieves comprehensive and accurate localization of network degradation root causes according to different network conditions.

[0023] This invention enables automatic detection of network problems, automatic location of the root cause of network degradation, and automatic output of solutions in network optimization. It can also be evolved into a general microservice to achieve automatic detection and automatic root cause location of network problems.

[0024] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0026] Figure 1 This is a schematic diagram of the steps of a network degradation root cause localization method in one embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of a multi-level transition matrix in one embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram illustrating the acquisition of the initial matrix and the target matrix from a multi-level transition matrix in one embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram illustrating the determination of the root cause of a target KPI based on the relationship between the indicator value and network performance in one embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of the implementation process of the network degradation root cause localization method in one embodiment of the present invention;

[0031] Figure 6 This is a data flow architecture diagram of a network degradation root cause localization method in one embodiment of the present invention;

[0032] Figure 7 This is a schematic diagram of the experimental process for a network degradation root cause localization method in one embodiment of the present invention;

[0033] Figure 8 This is a schematic diagram of the modules of a network degradation root cause localization system in one embodiment of the present invention;

[0034] Figure 9 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0035] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to fully and completely convey the concept of the exemplary embodiments to those skilled in the art.

[0036] The accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0037] Furthermore, the processes shown in the accompanying drawings are merely illustrative and do not necessarily include all steps. For example, some steps can be broken down, some steps can be combined or partially combined, and the actual execution order may change depending on the actual situation. The terms "first," "second," and similar terms used in the specific description do not indicate any order, quantity, or importance, but are only used to distinguish different components. It should be noted that, unless otherwise specified, embodiments of the present invention and features in different embodiments can be combined with each other.

[0038] Figure 1 This illustrates the main steps of a network degradation root cause localization method in one embodiment, referring to... Figure 1 As shown, the network degradation root cause localization method in this embodiment includes the following steps.

[0039] Step S110: Based on the network degradation data, select multiple sets of KPIs that are associated with the network degradation data from the key performance indicators (KPIs).

[0040] Network degradation data can be obtained from network degradation alarms. Specifically, in one embodiment, before obtaining network degradation data, the method further includes: in response to a network degradation alarm, obtaining network degradation data from the alarm data of the network degradation alarm. Key Performance Indicators (KPIs) of networks in various areas, such as communities or parks, can be monitored in real time by machines. When one or more KPIs in a certain area become abnormal, i.e., degradation occurs, a network degradation alarm is automatically issued. The alarm data of the network degradation alarm carries the degraded KPI and its index value, as well as other KPIs and their index values ​​that characterize the network condition of that area.

[0041] The process involves selecting multiple sets of KPIs associated with network degradation data from a set of Key Performance Indicators (KPIs). Specifically, this includes: obtaining the current degradation KPIs of the network degradation data, target sample data containing the current degradation KPIs, and a set of KPIs, where the target sample data includes target negative samples containing the current degradation KPIs; combining the KPI sets to obtain multiple itemsets; calculating the support of each itemset, where the support of an itemset is the probability of the occurrence of a target negative sample containing that itemset in the target sample data; and based on the support of each itemset, constructing a frequent itemset tree to mine frequent itemsets from the multiple itemsets, which serve as the multiple sets of KPIs associated with the network degradation data.

[0042] Taking the current degraded KPI as the degraded wireless connection success rate as an example, this indicates that the wireless connection success rate in the corresponding area is poor. The standard for degradation can be set as needed, for example, when it falls below a certain set value, the wireless connection success rate is judged to be degraded, triggering an alarm. The corresponding current KPI is the wireless connection success rate. The target sample data includes, for example, 1000 records, of which 400 are target positive samples (i.e., samples whose wireless connection success rate reaches its set value) and 600 are target negative samples (i.e., samples whose wireless connection success rate is lower than its set value). The target negative samples are denoted as event A. Each target sample data also contains other KPIs and their indicator values ​​that characterize the network condition.

[0043] A KPI set is a collection of all KPIs. Each KPI belongs to a KPI class; for example, the KPI "coverage" belongs to the coverage class. Each KPI class contains one or more KPIs; for example, the handover class contains multiple KPIs such as "eNodeB handover" and "X2 handover." Some KPI classes have clearly defined root causes, while others do not. Furthermore, due to the diversity of degradation KPIs causing network deterioration, traditional methods for locating the root cause of network deterioration suffer from problems such as identifying a single root cause and, in some cases, failing to locate the root cause altogether.

[0044] In this embodiment, the advantages of AI machine learning and big data computing are utilized to calculate the support of all KPI itemsets by combining network degradation data, thereby mining frequent itemsets related to network degradation data and providing a data foundation for subsequent root cause identification.

[0045] Specifically, when calculating support, the KPIs in the KPI set are first combined to obtain multiple itemsets. Each itemset can include one or more KPIs. For example, all KPIs can be combined pairwise to obtain multiple 2-itemsets. The support calculation formula is support(B→A)=P(A∩B), where event A is the occurrence of a target negative sample in the target sample data, and event B is the occurrence of a certain itemset in the target sample data. Taking the current degraded KPI as the degraded wireless connection success rate as an example, the target sample data consists of 1000 items. A certain itemset appears 500 times in the target sample data, of which it appears 300 times in the target negative samples. The support of this itemset is the probability that events A and B occur simultaneously: support(B→A)=P(A∩B)=300 / 1000=0.3.

[0046] When calculating the support of all itemsets, frequent itemsets are mined by constructing a frequentity tree. The process of constructing the frequentity tree includes: traversing all itemsets and calculating the support of each; discarding infrequent items; sorting all itemsets in descending order based on their support; rearranging all itemsets according to the obtained order; and discarding infrequent itemsets at the end of each set. The process of mining frequent itemsets from the frequentity tree includes: sorting the head list in descending order; traversing the nodes of the head list from smallest to largest to obtain the conditional pattern base, and simultaneously obtaining a frequent itemset. Finally, the mined frequent itemsets are those with a support greater than or equal to a certain threshold. The construction process of the frequentity tree and the mining of frequent itemsets are existing techniques and will not be elaborated further.

[0047] Through the above steps, we identified various KPIs related to the current alarm transaction (such as poor wireless connection success rate).

[0048] Step S120: Calculate the full confidence score of each KPI group and determine whether the full confidence score of the target group KPI with the highest full confidence score is greater than the threshold.

[0049] The process of calculating the full confidence level for each KPI group specifically includes: calculating the confidence level associated with each KPI group, which includes the first confidence level confidence(B→A) and the second confidence level confidence(A→B), where event A is the occurrence of the target negative sample in the target sample data, and event B is the occurrence of the KPI group in the target sample data; and calculating the full confidence level for each KPI group according to the full confidence level calculation formula all_confidence(A, B)=min{confidence(A→B), confidence(B→A)}.

[0050] Taking the degraded wireless connection success rate as an example, the target sample data consists of 1000 records. A certain KPI appears 500 times in the target sample data, and the probability of event B is P(B) = 500 / 1000 = 0.5. Therefore, the first confidence level of this KPI is confidence(B→A) = P(A∩B) / P(B) = 0.3 / 0.5 = 0.6. The target negative sample appears 600 times in the target sample data, and the probability of event A is P(A) = 600 / 1000 = 0.6. Therefore, the second confidence level of this KPI is confidence(A→B) = P(A∩B) / P(A) = 0.3 / 0.6 = 0.5.

[0051] When calculating full confidence, The full confidence level of the set of KPIs in the example above is 0.5.

[0052] Furthermore, since the confidence index may identify misleading strong correlations, this embodiment also introduces lift, which is compared with the individual support to remove several groups of KPIs that are independent of or negatively correlated with the target negative sample from each group of KPIs. Specifically, before calculating the confidence related to each group of KPIs, the following steps are also included: calculating the lift of each group of KPIs, where the lift of a group of KPIs is the ratio of the probability of that group of KPIs occurring simultaneously with the target negative sample in the target sample data to the probability of that group of KPIs occurring; and filtering out several groups of KPIs whose lift results are not positively correlated from multiple groups of KPIs.

[0053] The formula for calculating lift is: If lift(A→B)>1, the lift result is positively correlated, indicating that the rule "A→B" is a valid strong association rule; if lift(A→B)<1, the rule "A→B" is an invalid strong association rule; if lift(A→B)=1, it indicates that events A and B are independent of each other.

[0054] Through the above steps, the KPI of the target group with the highest full confidence was obtained by automatically mining association rules based on full confidence, and used as the basis for root cause localization.

[0055] Step S130: When the full confidence of the target group KPI is greater than the threshold, the root cause corresponding to the KPI class to which the target group KPI belongs is obtained according to the correspondence between KPI class and root cause, and is used as the target root cause of network degradation data.

[0056] A target group KPI with a full confidence score greater than a threshold (which can be set as needed) indicates that the target KPIs in that group possess explicit characteristics of degradation, allowing the identification of the root causes of network degradation based on these target KPIs. Furthermore, since a target group KPI is a KPI itemset composed of one or more target KPIs, the root causes may be multidimensional. In this embodiment, a multi-level transition matrix based on expert prior knowledge is used for root cause localization.

[0057] Specifically, each KPI class has multiple preset interval values, and the correspondence is between the combination of interval values ​​of the KPI class and the combination of root causes. The correspondence may include a multi-level transition matrix, where the first-level matrix uses the interval values ​​of the two first-level KPI classes in the KPI class as row and column identifiers, the Nth-level matrix uses the interval value of one N-level KPI class in the KPI class and the elements of the (N-1)th-level matrix as row and column identifiers, and N≥2; each element is a combination of root causes consisting of one or more root causes, and each element can be associated with the KPI class corresponding to its row identifier and / or column identifier.

[0058] When analyzing network degradation data, the indicator value of each selected KPI is also obtained; when obtaining the root cause corresponding to the KPI class to which the target group KPI belongs, the corresponding interval value combination is determined from the correspondence relationship based on the target KPI and its indicator value in the target group KPI and the corresponding root cause combination is obtained.

[0059] Figure 2 The structure of a multi-level transition matrix in one embodiment is shown, with reference to Figure 2As shown, in this embodiment, the KPI categories include five types: quality, coverage, interference, handover, and capacity. The root causes also include five types: quality, coverage, interference, handover, and capacity. It should be noted that while the classification names of the KPI categories match the root cause names, the root causes causing the degradation of one or more KPIs are not unique. Therefore, the correspondence is not a one-to-one relationship between KPI categories and root causes; instead, a multi-level transition matrix is ​​used. The interval values ​​of each KPI category are subdivided into three categories: good, medium, and poor, but this is not a limitation. Based on historical network data and expert experience, the importance coefficient of each KPI category and the root causes associated with each KPI category are determined, constructing a four-level transition matrix. The first-level matrix 210 uses the interval values ​​of two first-level KPI categories (quality and coverage) as row and column identifiers. Each element of the first-level matrix 210 consists of one or more root causes, and each element may not be related to its corresponding row and column identifier, or may only be related to its corresponding row identifier, or only to its corresponding column identifier, or both. For example, the element "coverage," determined by the row identifier "poor quality" and the column identifier "poor coverage," is only related to its corresponding column identifier; the element "quality," determined by the row identifier "poor quality" and the column identifier "good coverage," is only related to its corresponding row identifier; and the element "coverage and quality," determined by the row identifier "poor quality" and the column identifier "relatively poor coverage," is related to both its corresponding row and column identifiers. However, the element "good coverage quality," determined by the row identifier "good quality" and the column identifier "good coverage," indicates a good network condition with no root causes of network degradation; therefore, this element is not related to the KPI category corresponding to its row and column identifiers.

[0060] In the second-level matrix 220, a new range value for a second-level KPI category (interference) is added as the row identifier, and the column identifiers are derived from the elements of the first-level matrix after deduplication. Specifically, the elements of the first-level matrix, after deduplication, include four values: "coverage," "quality," "coverage and quality," and "good coverage and quality." These are then transferred to serve as the column identifiers of the second-level matrix, forming the row and column identifiers of the second-level matrix together with the three range values ​​of the second-level KPI category (interference). The composition of each element in the second-level matrix is ​​similar to that of the first-level matrix. For example, the row identifier "severe interference" and the column identifier "coverage" determine the element "coverage and interference," which is related to the KPI category corresponding to both the row and column identifiers. The row identifier "no interference" and the column identifier "good coverage and quality" determine the element "good wireless environment," which is unrelated to either the KPI category corresponding to its row or column identifier.

[0061] In the third-level matrix 230, a new range value for a third-level KPI category (switching) is added as the row identifier, and the column identifiers are derived from the elements of the second-level matrix after deduplication. In the fourth-level matrix 240, a new range value for a fourth-level KPI category (capacity) is added as the row identifier, and the column identifiers are derived from the elements of the third-level matrix after deduplication. The specific matrix composition is the same as above and will not be repeated.

[0062] Based on the aforementioned multi-level transition matrix, the root causes corresponding to the KPI classes to which the target group KPIs belong are obtained. Specifically, this includes: obtaining the target KPI class to which each target KPI in the target group KPIs belongs; obtaining an initial matrix with at least one current target KPI class's interval value as the row identifier and / or column identifier according to the transition order of the multi-level transition matrix; pruning the initial matrix by removing row and column identifiers and elements related to the remaining KPI classes excluding the current target KPI class; determining whether there are any remaining target KPI classes not included in the initial matrix; if so, obtaining a target matrix with the remaining target KPI class's interval value and the elements in the pruned initial matrix as row and column identifiers according to the transition order; if not, using the pruned initial matrix as the target matrix; and determining the target interval value combination based on the interval value of the target KPI class to which the indicator value of each target KPI falls, and obtaining the root cause corresponding to the target interval value combination from the target matrix.

[0063] Figure 3 This illustrates a process for obtaining the initial matrix and the target matrix from a multi-level transition matrix in one embodiment, combined with... Figure 2 and Figure 3 As shown, taking the degraded wireless connection success rate as an example of the current degraded KPI, after full confidence calculation, target KPIs that meet the full confidence threshold are identified, including two target KPIs: RSSI (Radio Frequency Indicator) and maximum number of users. RSSI belongs to the interference category, while maximum number of users belongs to the capacity category. Following the transition order, a second-level matrix 220, labeled with "interference" as the row identifier, is obtained as the initial matrix. This initial matrix is ​​then pruned to remove irrelevant KPI classes. Specifically, elements in the second-level matrix 220 identified by columns unrelated to "interference," such as "coverage," "quality," and "coverage, quality," are pruned, resulting in a pruned initial matrix 310. The target KPI group also includes the remaining target KPI class "capacity" not included in the initial matrix. Following the transition order, a fourth-level matrix 240, labeled with "capacity" as the row identifier, is obtained. The column identifiers and corresponding elements formed by the elements in the pruned initial matrix 310 are retained in the fourth-level matrix 240, ultimately forming the target matrix 320.

[0064] If the column identifiers in the fourth-level matrix 240 do not match the elements in the pruned initial matrix 310, the third-level matrix 230 can be pruned first according to the transfer order, removing elements related to other KPI categories except "interference". Then, the elements retained in the pruned third-level matrix 230 are transferred to the column identifiers of the fourth-level matrix 240, and together with the row identifier "capacity" and the corresponding elements, they form the target matrix 320.

[0065] After obtaining the target matrix 320, the target interval values ​​are determined based on the range of the target KPI class into which the indicator values ​​of the two target KPIs fall. For example, in the network degradation data, the actual indicator value of the target KPI "RSSI (Rated Noise Level Indicator)" falls into the "Interference" range of the interference KPI, and the actual indicator value of the target KPI "Maximum Number of Users" falls into the "High Load" range of the capacity KPI. Then, the combination of target range values ​​is determined from the target matrix 320, that is, the combination corresponding to the row label "High Load" and the column label "Interference". Thus, the root cause "Capacity" is obtained as the target root cause of the network degradation data.

[0066] Step S140: When the full confidence level of the target group KPI does not reach the threshold, obtain the remaining KPIs excluding the target group KPI from multiple groups of KPIs, and determine the target root cause based on the deviation of the remaining KPIs from the root cause.

[0067] When the most frequent itemset with the highest full confidence does not meet the threshold requirement, it indicates that the primary KPI, i.e. the target KPI, does not have obvious deterioration characteristics and the root cause cannot be directly determined. In this case, the root cause can be indirectly determined by comparing the deviation and using the secondary KPI, i.e. the remaining KPI.

[0068] Specifically, based on the deviation between the remaining KPIs and the root cause, the process includes: fitting the remaining KPIs to the root cause to obtain the fitted value of each remaining KPI relative to each root cause; and obtaining the deviation between each remaining KPI and each root cause according to the deviation calculation formula d=(yr) / r. In the deviation calculation formula, d is the deviation between a remaining KPI and a root cause, y is the fitted value of the remaining KPI relative to the root cause, and r is the index value of the remaining KPI.

[0069] Obtaining the fitted value of each remaining KPI relative to each root cause specifically includes: obtaining the root cause value of each root cause, where the root cause value of a root cause is the index value corresponding to the KPI with the highest confidence among the KPI classes matching that root cause; calculating the fitted value of each remaining KPI relative to each root cause according to the fitting formula y=(k+α(x)δ)·x+(m+α(x)γ); the fitting formula is a piecewise linear function, where x is the root cause value of a root cause, k is the slope of the current piecewise linear part corresponding to the root cause value of that root cause, δ is the change in the slope of the current piecewise linear part, α(x) is the coefficient of the current piecewise linear part, m is the bias value of the current piecewise linear part, γ is the boundary value of the current piecewise linear part, and y is the fitted value of a remaining KPI relative to that root cause.

[0070] By fitting KPIs to root causes, a piecewise linear fitting function is constructed. Piecewise linear regression is a regression estimation method where the regression of y on x follows a certain linear relationship within a certain range of x, and follows other linear relationships with different slopes in other ranges. Using the piecewise linear fitting function is equivalent to constructing a fitting function between KPIs and each type of root cause.

[0071] Before obtaining the fitting formula, it is necessary to determine the root cause value for each type of root cause. Specifically, the root causes and KPI classes include five categories: quality, coverage, interference, handover, and capacity. When determining the root cause value for quality-related root causes, the confidence level of each quality-related KPI in the quality KPI class that matches the quality-related root cause is calculated. The confidence level of a quality-related KPI is the probability that the KPI of that quality-related KPI will also appear when a negative sample appears in all sample data. Negative samples are sample data where network performance degradation occurs. The confidence level of each quality-related KPI can be obtained by machine simulation to determine the indicator value at which its confidence level reaches its maximum. Finally, the quality-related KPI with the highest confidence level is selected, and the indicator value corresponding to its highest confidence level is taken as the root cause value of the quality-related root cause. After calculation, the quality-related KPI with the highest confidence level in the quality KPI class is CQI (Channel Quality Indicator). The confidence level reaches its maximum value when the proportion of CQI>=7 is less than 74%. Therefore, the root cause value of the quality-related root cause is 7. The root cause values ​​for coverage, interference, handover, and capacity categories are obtained in the same way. Calculations show that the coverage KPI with the highest confidence is RSRP (Reference Signal Receiving). Power (reference signal received power), the confidence level reaches its highest value when the proportion of RSRP > -105dBm is less than 74%, therefore the root cause value for the coverage category is -105; In the interference KPI category, the interference KPI with the highest confidence level is interference noise, the confidence level reaches its highest value when the average interference noise in the 1800M / 800M band is greater than -95dBm, therefore the root cause value for the interference category is -95; In the handover KPI category, the handover KPI with the highest confidence level is eNodeB intra-handover or X2 handover, the confidence level reaches its highest value when the eNodeB intra-handover success rate in the 1800M / 800M band is less than 98%, and the confidence level reaches its highest value when the X2 interface handover success rate in the 1800M / 800M band is less than 98%, therefore the root cause value for the handover category is 98%; In the capacity KPI category, the capacity KPI with the highest confidence level is maximum RRC (Radio Resource ...). The confidence level for the maximum number of connected users (RRC) in the 1800M band reaches 175, and the confidence level for the maximum number of connected users in the 800M band reaches 31. Therefore, the root cause values ​​for capacity-related root causes are 175 in the 1800M band and 31 in the 800M band.

[0072] The process of determining the root cause value for each type of root cause and constructing the piecewise linear fitting function can be obtained during machine training. In practical applications, the pre-set fitting formula and root cause value can be used directly.

[0073] The above fitting formula is a piecewise linear function, meaning that there is a different fitting function for each type of root cause. Here, α(x) is the coefficient of a piecewise linear part, taking the value 0 or 1 depending on the range of root cause values ​​corresponding to each piecewise linear part; k is the slope of each piecewise linear part; δ is the change in the slope of each piecewise linear part, used to control the fitting accuracy and avoid overfitting, taking a value between 0 and 1; m is the bias value of each piecewise linear part, its value related to the range of root cause values ​​corresponding to each piecewise linear part; γ is the boundary value of each piecewise linear part, for example, taking the lower limit of the root cause value corresponding to each piecewise linear part.

[0074] After obtaining the fitted value of a remaining KPI relative to a root cause according to the fitting formula, the fitted value is substituted into the deviation calculation formula and calculated with the actual index value of the remaining KPI to obtain the deviation between the remaining KPI and the root cause.

[0075] Finally, different root cause determination methods are adopted according to the different characteristics of the remaining KPIs: for the remaining KPIs whose index values ​​are inversely proportional to network performance, the root cause corresponding to the largest deviation is selected; for the remaining KPIs whose index values ​​are directly proportional to network performance, the root cause corresponding to the smallest deviation is selected; and the selected root cause is used as the target root cause.

[0076] Figure 4 This illustrates an embodiment of the process for determining target root causes based on the relationship between indicator values ​​and network performance, with reference to... Figure 4 As shown, firstly, in step S140-2, the deviation between each remaining KPI and coverage / quality / capacity / interference / handover is calculated. Then, in step S140-4, for indicators representing poor quality (i.e., larger values ​​indicate worse network performance, such as latency indicators and call drop rate indicators), the root cause corresponding to the result with the largest deviation is selected. For indicators representing good performance (i.e., larger values ​​indicate better network performance, such as success rate indicators and MR coverage indicators), the root cause corresponding to the result with the smallest deviation is selected. Finally, the root cause corresponding to the result with the largest absolute value of deviation is taken as the target root cause.

[0077] Furthermore, after obtaining the root cause, it also includes: outputting network degradation alarms and the root cause for manual follow-up and handling.

[0078] The aforementioned method for locating the root causes of network degradation leverages the advantages of AI machine learning and big data computing. It calculates the support of all KPI itemsets, identifies frequent itemsets, and determines the target group KPI with the highest full confidence. For root cause location of explicit indicators: Traditional methods locate a single root cause through a single indicator, but in reality, network degradation may be caused by a combination of factors. This method combines the target KPIs obtained through machine learning with a multi-level transition matrix to achieve multi-dimensional root cause analysis and location output. For root cause location of inexplicable indicators: When a network has a deteriorating KPI, but its related primary KPIs lack obvious characteristics and cannot be used as a basis for determining the root cause, this method establishes a fitting function between secondary KPIs and the root cause, and uses a deviation comparison method to locate the root cause of the deteriorating KPI.

[0079] Figure 5 This illustrates the implementation flow of a network degradation root cause localization method in one embodiment, with reference to... Figure 5 As shown, the process begins with data input and cleaning in step S510, which includes data collection, data extraction, transformation and loading, feature extraction, model training, feature cross-validation, model output, and model parameter parsing, to obtain all sample data. During data collection, various network data sources can be used, including PM (stored network performance data), MR (stored network coverage data), FM (stored network device alarm data), and CM (stored network configuration data). Then, step S520 performs preliminary root cause identification. This step is based on machine learning and obtains root cause values ​​for five categories: coverage, interference, handover, capacity, and quality, as well as the basic algorithmic logic for frequent itemset mining and full confidence calculation before specific root cause localization. Next, step S530 performs enhanced root cause localization, corresponding to the root cause localization process for both explicit and implicit indicators. Finally, in step S540, the root cause localization scheme is output. That is, when the target KPI has obvious deterioration characteristics, the target root cause is determined based on the target KPI and the multi-level transition matrix. When the target KPI does not have obvious deterioration characteristics, the target root cause is determined based on the deviation between the remaining KPI and the root cause. The specific process can be referred to the above embodiment and will not be repeated.

[0080] Figure 6 This illustrates the data flow architecture of a network degradation root cause localization method in one embodiment, with reference to... Figure 6As shown, in the data preparation stage 610, cleaned and processed data selected from the data source 6101 flows into the HIVE database 6102; in the machine learning stage 620, rule learning is performed based on random positive and negative samples, and the results are stored in the rule base 6201. Root cause localization is performed using the learned rules and the identified problem network, and the root cause localization logic is stored in the localization database 6202; in the root cause enhancement stage 630, root cause localization is enhanced by combining alarm data and root cause localization logic through the SPARK computing engine 6301, and the localization results can be stored in the summary database 6302; the summary can be performed according to configuration, by hourly or other standards. In the database service stage 640, relevant application tables 6401 can be generated based on the summary results; in the Web service stage 650, the network degradation root cause localization logic can be abstracted into a general microservice to achieve automatic perception and automatic root cause localization of network problems.

[0081] Compared to traditional methods that rely on a single indicator and apply a rigid threshold-based approach to network degradation root cause localization, the aforementioned method utilizes machine learning to uncover associated KPIs and calculate full confidence scores. Different methods are employed for root cause localization depending on whether the degradation network exhibits explicit or implicit characteristics. For degradation networks with explicit characteristics, a multi-level transition matrix approach ensures more comprehensive localization results. For degradation networks with implicit characteristics, a deviation analysis method can also pinpoint the target root cause. This method is timely and continuous, improving the efficiency, standardization, accuracy, and operability of network problem handling. It also shortens the time required for problem discovery, resolution, and closure, enhancing customer network awareness and achieving cost reduction and efficiency improvement.

[0082] Figure 7 This illustrates the experimental procedure for a network degradation root cause localization method in one embodiment, with reference to... Figure 7 As shown, the verification was carried out at a base station in a certain area of ​​Shanghai. The main process includes the following steps.

[0083] S710, Confidence-Based Machine Learning: This method selects seven consecutive days of data from the entire network as samples for machine learning. Specifically, it uses seven consecutive days of base station data from the entire network as training data. After software performs calculations such as data consistency checks, missing value handling, data association, and outlier handling, machine learning is performed to obtain a rule base for relevant indicators. This includes identifying the KPIs related to wireless connection success rate, intra-system handover success rate, and the 4G to 3G transition ratio. It also identifies the main KPIs included in each group with the highest full confidence score and their full confidence values. The method for root cause analysis is determined based on the full confidence score. If a multi-level transition matrix method is used, the target matrix is ​​determined based on the main KPIs.

[0084] S720 Automatic Output of Poor-Quality Cells: This process selects two consecutive days of data from a specific area as input and uses machine learning to automatically locate poor-quality cells. After training on a rule base, a subset of sample data is selected for result verification. The method involves reselecting two consecutive days of data from a specific area as the data source for the pilot region, and then calculating and outputting a list of problematic cells. This step applies the AI ​​training results to real-world data applications to proactively discover poor-quality wireless cells.

[0085] S730, Root Cause Analysis of Poor-Quality Community KPIs: Combining expert-provided multi-level transition matrices and contrast deviation, the root causes of KPIs are output. When calculating the list of problematic communities, the root causes of their poor quality are also located, mainly including the following two steps: First, for primary KPIs with explicit characteristics, the multi-level transition matrix method is used to locate the root causes; second, for secondary KPIs with inexplicable characteristics, the contrast deviation method is used to locate the root causes. After learning and calculating through the above process, all poor-quality communities in the Shanghai area for two days were output and their root causes were located.

[0086] S740, Root Cause Analysis of Poor-Quality Cell KPIs: Verifying the Location Results. Taking wireless access indicators as an example; in one sample data set, the root cause was capacity. Looking at the specific KPIs of this sample data set, the maximum number of RRC users was 796 and the average number of RRC users was 767, both very clearly high load indicators, indicating a capacity problem in this sample data set, and the root cause was accurately located. In another sample data set, the root cause was interference. Looking at the interference-related KPIs in this sample data set, the port noise floor reached -60dB, indicating a very high interference problem, further indicating accurate root cause location.

[0087] In summary, the aforementioned methods for locating the root causes of network degradation involve traversing all KPI itemsets, calculating support to mine frequent itemsets, and then further determining the root cause based on the target group KPI with the highest total confidence within the frequent itemsets. The multi-level transition matrix method is used to locate the root causes of indicators with explicit characteristics: when the maximum total confidence exceeds a threshold, the multi-level transition matrix method quickly locates multi-dimensional KPI root causes. This method allows target KPIs with explicit characteristics to comprehensively consider root causes from multiple dimensions, resulting in a more comprehensive location result, rather than relying solely on a single indicator. The contrast deviation method is used to locate the root causes of indicators with non-explicit characteristics: when a network degradation indicator exists but the related primary KPIs lack obvious characteristics, the primary KPI cannot directly identify the root cause. By establishing a model between other secondary KPIs and the root cause, the contrast deviation method is used to locate the root cause of the degradation indicator.

[0088] This invention also provides a network degradation root cause localization system, which can be used to implement the network degradation root cause localization method described in any of the above embodiments. The features and principles of the network degradation root cause localization method described in any of the above embodiments can be applied to the following network degradation root cause localization system embodiments. In the following network degradation root cause localization system embodiments, the features and principles already explained regarding network degradation root cause localization will not be repeated.

[0089] Figure 8 This diagram illustrates the main modules of a network degradation root cause localization system in one embodiment, with reference to... Figure 8 As shown, the network degradation root cause localization system 800 in this embodiment includes: an associated KPI screening module 810, configured to screen multiple sets of KPIs associated with network degradation data from a set of key performance indicators (KPIs); a full confidence calculation module 820, configured to calculate the full confidence of each set of KPIs and determine whether the full confidence of the target set of KPIs with the highest full confidence is greater than a threshold; a first root cause determination module 830, configured to, when the full confidence of the target set of KPIs is greater than the threshold, obtain the root cause corresponding to the KPI class to which the target set of KPIs belongs, based on the correspondence between KPI class and root cause, and use it as the target root cause of the network degradation data; and a second root cause determination module 840, configured to, when the full confidence of the target set of KPIs is not greater than the threshold, obtain the remaining KPIs excluding the target set of KPIs from the multiple sets of KPIs, and determine the target root cause based on the deviation between the remaining KPIs and the root cause.

[0090] Furthermore, the network degradation root cause localization system 800 may also include modules that implement other process steps of the above-described embodiments of the network degradation root cause localization methods. For example, see [reference needed]. Figure 6 As shown, a complete system architecture is formed, encompassing data preparation, machine learning, root cause enhancement, and the creation of database services. The specific principles of each module can be found in the descriptions of the above embodiments of the network degradation root cause localization methods, and will not be repeated here.

[0091] As described above, the network degradation root cause localization system of the present invention can mine related KPIs based on network degradation data, determine whether the target KPI has obvious degradation characteristics based on the maximum full confidence, locate the root cause based on the target KPI and the multi-level transition matrix when the target KPI has obvious degradation characteristics, and determine the target root cause based on the deviation between the remaining KPIs and the root cause when the target KPI does not have obvious degradation characteristics. Thus, it can achieve comprehensive and accurate localization of network degradation root causes according to different network conditions. Furthermore, it can automatically detect network problems, automatically locate network degradation root causes, and automatically output solutions in network optimization, and can evolve into a general microservice to achieve automatic detection and automatic root cause localization of network problems.

[0092] This invention also provides an electronic device, including a processor and a memory, wherein the memory stores executable instructions, and when the executable instructions are executed by the processor, the network degradation root cause localization method described in any of the above embodiments is implemented.

[0093] As described above, the electronic device of the present invention can mine related KPIs based on network degradation data, determine whether the target KPI has obvious degradation characteristics based on the maximum full confidence, and perform root cause localization based on the target KPI and a multi-level transition matrix when the target KPI has obvious degradation characteristics. When the target KPI does not have obvious degradation characteristics, the target root cause is determined based on the deviation between the remaining KPIs and the root cause. Thus, it can achieve comprehensive and accurate localization of network degradation root causes according to different network conditions. Furthermore, it can automatically detect network problems, automatically locate network degradation root causes, and automatically output solutions during network optimization. It can also evolve into a general microservice to achieve automatic detection and automatic root cause localization of network problems.

[0094] Figure 9 This is a schematic diagram of the structure of the electronic device in an embodiment of the present invention. It should be understood that... Figure 9 The modules are merely shown schematically. These modules can be virtual software modules or actual hardware modules. The merging, splitting, and addition of other modules are all within the scope of protection of this invention.

[0095] like Figure 9 As shown, the electronic device 900 is presented in the form of a general-purpose computing device. The components of the electronic device 900 include, but are not limited to: at least one processing unit 910, at least one storage unit 920, a bus 930 connecting different platform components (including storage unit 920 and processing unit 910), a display unit 940, etc.

[0096] The storage unit stores program code, which can be executed by the processing unit 910, causing the processing unit 910 to perform the steps of the network degradation root cause localization method described in any of the above embodiments. For example, the processing unit 910 can perform, as follows: Figure 1 The steps are shown.

[0097] Storage unit 920 may include readable media in the form of volatile storage units, such as random access memory (RAM) 9201 and / or cache memory 9202, and may further include read-only memory (ROM) 9203.

[0098] The storage unit 920 may also include a program / utility 9204 having one or more program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0099] Bus 930 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0100] Electronic device 900 can also communicate with one or more external devices 9000, which may be one or more of the following: keyboard, pointing device, Bluetooth device, etc. These external devices 9000 enable users to interact and communicate with electronic device 900. Electronic device 900 can also communicate with one or more other computing devices, including routers and modems. This communication can be performed via input / output (I / O) interface 950. Furthermore, electronic device 900 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 960. Network adapter 960 can communicate with other modules of electronic device 900 via bus 930. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage platforms.

[0101] This invention also provides a computer-readable storage medium for storing a program that, when executed, implements the network degradation root cause localization method described in any of the above embodiments. In some possible implementations, various aspects of this invention can also be implemented as a program product comprising program code, which, when run on a terminal device, causes the terminal device to execute the network degradation root cause localization method described in any of the above embodiments.

[0102] As described above, the computer-readable storage medium of the present invention can mine relevant KPIs from network degradation data, determine whether the target KPI has explicit degradation characteristics based on the maximum full confidence level, perform root cause localization based on the target KPI and a multi-level transition matrix when the target KPI has explicit degradation characteristics, and determine the target root cause based on the deviation between the remaining KPIs and the root cause when the target KPI does not have explicit degradation characteristics. Thus, it can achieve comprehensive and accurate localization of network degradation root causes according to different network conditions. Furthermore, it can automatically detect network problems, automatically locate network degradation root causes, and automatically output solutions during network optimization, and can evolve into a general microservice to achieve automatic detection and automatic root cause localization of network problems.

[0103] The program product may be a portable compact disc read-only memory (CD-ROM) containing program code and may run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto; it may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0104] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include, but are not limited to: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0105] A readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0106] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device, for example, via the Internet using an Internet service provider.

[0107] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for locating the root causes of network degradation, characterized in that, include: Based on network degradation data, multiple sets of KPIs associated with the network degradation data are selected from the key performance indicators (KPIs). The step of selecting multiple sets of KPIs associated with the network degradation data from the set of Key Performance Indicators (KPIs) includes: obtaining the current degradation KPI of the network degradation data, target sample data containing the current KPI corresponding to the current degradation KPI, and the KPI set, wherein the target sample data includes target negative samples containing the current degradation KPI; combining the KPI set to obtain multiple itemsets; calculating the support of each itemset, wherein the support of an itemset is the probability of the occurrence of the target negative sample containing that itemset in the target sample data; and based on the support of each itemset, constructing a frequent itemset tree to mine frequent itemsets from the multiple itemsets, which serve as multiple sets of KPIs associated with the network degradation data. Calculate the full confidence score for each group of KPIs, and determine whether the full confidence score of the target group KPI with the highest full confidence score is greater than the threshold. The calculation of the full confidence score for each group of KPIs includes: calculating the confidence score associated with each group of KPIs, where the confidence score associated with a group of KPIs includes a first confidence score (B→A) and a second confidence score (A→B), wherein event A is the occurrence of the target negative sample in the target sample data, and event B is the occurrence of that group of KPIs in the target sample data; according to the full confidence score calculation formula... Calculate the full confidence level for each group of KPIs; If so, based on the correspondence between KPI class and root cause, obtain the root cause corresponding to the KPI class to which the target group KPI belongs, and use it as the target root cause of the network degradation data; If not, obtain the remaining KPIs excluding the target group KPI from the multiple sets of KPIs, and determine the target root cause based on the deviation of the remaining KPIs from the root cause.

2. The network degradation root cause localization method as described in claim 1, characterized in that, Each KPI category has multiple preset interval values, and the correspondence is the correspondence between the interval value combination of the KPI category and the root cause combination of the root cause. When processing network degradation data, the index value of each selected KPI is also obtained; When obtaining the root cause corresponding to the KPI class to which the target group KPI belongs, the corresponding interval value combination is determined from the correspondence relationship based on the target KPI and its indicator value in the target group KPI and obtain the corresponding root cause combination.

3. The network degradation root cause localization method as described in claim 2, characterized in that, The correspondence includes a multi-level transition matrix; In the multi-level transition matrix, the first-level matrix is ​​identified by the range values ​​of the two first-level KPI classes in the KPI class, and the Nth-level matrix is ​​identified by the range values ​​of one Nth-level KPI class in the KPI class and the elements of the (N-1)th-level matrix, where N≥2. Each element is a combination of one or more root causes, and each element may be associated with a KPI class corresponding to its row identifier and / or column identifier.

4. The network degradation root cause localization method as described in claim 3, characterized in that, The step of obtaining the root cause corresponding to the KPI class to which the target group KPI belongs includes: Obtain the target KPI class to which each target KPI in the target group belongs; According to the transition order of the multi-level transition matrix, an initial matrix is ​​obtained with at least one current target KPI class interval value as row identifier and / or column identifier; The initial matrix is ​​pruned by removing row and column identifiers and elements related to all KPI classes except the current target KPI class. Determine if there are any remaining target KPI classes not included in the initial matrix; If so, according to the transfer order, a target matrix is ​​obtained with the interval values ​​of the remaining target KPI class and the elements in the pruned initial matrix as row and column identifiers, respectively; If not, the pruned initial matrix shall be used as the target matrix; and The target interval value combination is determined based on the interval value of the target KPI class into which the indicator value of each target KPI falls, and the root cause corresponding to the target interval value combination is obtained from the target matrix.

5. The network degradation root cause localization method as described in claim 1, characterized in that, When processing network degradation data, the index value of each selected KPI is also obtained; The step of determining the deviation between the remaining KPIs and the root cause includes: By fitting the remaining KPIs with the root cause, a fitting value for each remaining KPI relative to each root cause is obtained; The deviation between each remaining KPI and each root cause is obtained according to the deviation calculation formula d=(yr) / r; In the deviation calculation formula, d is the deviation between a residual KPI and a root cause, y is the fitted value of the residual KPI relative to the root cause, and r is the index value of the residual KPI.

6. The network degradation root cause localization method as described in claim 5, characterized in that, Obtaining the fitted value of each remaining KPI relative to each root cause includes: Obtain the root cause value for each root cause, where the root cause value is the indicator value corresponding to the KPI with the highest confidence among the KPI classes that match that root cause. According to the fitting formula Calculate the fitted value of each of the remaining KPIs relative to each of the root causes; The fitting formula is a piecewise linear function, in which x is the root cause value of a root cause, k is the slope of the current piecewise linear part corresponding to the root cause value of the root cause, δ is the change in the slope of the current piecewise linear part, α(x) is the coefficient of the current piecewise linear part, m is the bias value of the current piecewise linear part, γ is the boundary value of the current piecewise linear part, and y is the fitted value of a residual KPI relative to the root cause.

7. The network degradation root cause localization method as described in claim 5, characterized in that, Determining the target root cause includes: For the remaining KPIs whose index values ​​are inversely proportional to network performance, the root causes corresponding to the maximum deviation are selected. For the remaining KPIs whose values ​​are directly proportional to network performance, the root causes corresponding to the minimum deviation are selected; and The selected root causes are used as the target root causes.

8. The network degradation root cause localization method as described in claim 1, characterized in that, Before calculating the confidence level associated with each group of KPIs, the method further includes: Calculate the improvement of each group of KPIs. The improvement of a group of KPIs is the ratio of the probability of the KPIs occurring simultaneously under the condition that the target negative sample occurs in the target sample data to the probability of the KPIs occurring in that group. We screened out several groups of KPIs whose improvement results were not positively correlated from the multiple groups of KPIs.

9. The network degradation root cause localization method as described in claim 1, characterized in that, Before using network degradation data, the following are also included: In response to a network degradation alarm, the network degradation data is obtained from the alarm data of the network degradation alarm; The target root cause of the network degradation data, or the determination of the target root cause, further includes: Output the network degradation alarm and the target root cause.

10. A network degradation root cause localization system, characterized in that, include: The KPI filtering module is configured to filter multiple sets of KPIs associated with the network degradation data from the key performance indicator (KPI) set. The KPI filtering module selects multiple sets of KPIs associated with the network degradation data from the set of key performance indicators (KPIs). This includes: obtaining the current degradation KPI of the network degradation data, target sample data containing the current KPI, and the KPI set, wherein the target sample data includes target negative samples containing the current degradation KPI; combining the KPI set to obtain multiple itemsets; calculating the support of each itemset, where the support of an itemset is the probability of a target negative sample containing that itemset appearing in the target sample data; and based on the support of each itemset, constructing a frequent itemset tree to mine frequent itemsets from the multiple itemsets, which serve as the multiple sets of KPIs associated with the network degradation data. The full confidence calculation module is configured to calculate the full confidence of each group of KPIs and determine whether the full confidence of the target group KPI with the highest full confidence is greater than a threshold. The full confidence calculation module calculates the full confidence of each group of KPIs, including: calculating the confidence associated with each group of KPIs, where the confidence associated with a group of KPIs includes a first confidence (B→A) and a second confidence (A→B), where event A is the occurrence of the target negative sample in the target sample data, and event B is the occurrence of that group of KPIs in the target sample data; according to the full confidence calculation formula... Calculate the full confidence level for each group of KPIs; The first root cause determination module is configured to, when the full confidence of the target group KPI is greater than the threshold, obtain the root cause corresponding to the KPI class to which the target group KPI belongs, based on the correspondence between KPI class and root cause, and use it as the target root cause of the network degradation data. The second root cause determination module is configured to, when the full confidence level of the target group KPI is not greater than the threshold, obtain the remaining KPIs excluding the target group KPIs in multiple groups of KPIs, and determine the target root cause based on the deviation between the remaining KPIs and the root cause.

11. An electronic device, characterized in that, include: One processor; A memory, wherein executable instructions are stored; When the executable instructions are executed by the processor, they implement the network degradation root cause localization method as described in any one of claims 1-9.

12. A computer-readable storage medium for storing a program, characterized in that, When the program is executed by the processor, it implements the network degradation root cause localization method as described in any one of claims 1-9.

Images