Analysis Method, Device and Computer Readable Storage Medium for Network Fault

Abstract

The present invention discloses an analysis method, device and computer-readable storage medium for network faults, relating to the field of network technologies. The analysis method for network faults includes: determining poor-quality call records in a call record set; determining a first poor-quality network segment according to the proportion of poor-quality call records in each network segment; for each dimension, using a pre-trained model corresponding to the dimension and key features to process the poor-quality call records in each network segment to determine a second poor-quality network segment under the dimension; determining the network segments that belong to both the first poor-quality network segment and the second poor-quality network segment as the poor-quality network segments under the corresponding dimension; and determining the root cause of the fault in the poor-quality network segments. Through the analysis of delimiting and segmenting the fault causes, the specific network segment where the cause lies is finally located, so that the fault causes can be analyzed more accurately, effectively guiding the operation and maintenance of the live network and ensuring the user experience.

Description

Technical Field

[0001] The present invention relates to the field of network technologies, and in particular, to a method and apparatus for analyzing network faults and a computer-readable storage medium. Background Art

[0002] The iterative development of networks has promoted the popular application of mobile Internet services. Correspondingly, mobile Internet services have put forward higher requirements for network quality. The user experience of services directly reflects the network quality. How to quickly and accurately locate the reasons for poor quality and troubleshoot network faults is the key for operators to improve operation efficiency, reduce operation and maintenance costs, and ensure customer perception. Summary of the Invention

[0003] After analysis by the inventors, it is found that there is a problem of inaccurate fault analysis in the related technologies.

[0004] One technical problem to be solved by the embodiments of the present invention is: how to improve the accuracy of network fault analysis.

[0005] According to the first aspect of some embodiments of the present invention, there is provided a method for analyzing network faults, including: determining poor-quality call records in a call record set; determining a first poor-quality network segment according to the proportion of poor-quality call records in each network segment; for each dimension, using a pre-trained model and key features corresponding to the dimension to process the poor-quality call records in each network segment, and determining a second poor-quality network segment under the dimension; determining the network segments that belong to both the first poor-quality network segment and the second poor-quality network segment as the poor-quality network segments under the corresponding dimension; and determining the root cause of the faults in the poor-quality network segments.

[0006] In some embodiments, determining a first poor-quality network segment according to the proportion of poor-quality call records in each network segment includes: calculating the proportion of poor-quality call records in each network segment in the call record set; for each network segment, when the proportion of poor-quality call records in the network segment is greater than the poor-quality threshold corresponding to the network segment, determining the network segment as the first poor-quality network segment.

[0007] In some embodiments, for each dimension, using a pre-trained model and key features corresponding to the dimension to process the poor-quality call records in each network segment, and determining a second poor-quality network segment under the dimension includes: for the poor-quality call records of each network segment under each dimension: generating a corresponding subset of poor-quality call records; using a pre-trained random forest model corresponding to the dimension to process the subset of poor-quality call records, and generating a poor-quality classification result of the corresponding network segment under the dimension, where the poor-quality classification result indicates whether the corresponding network segment is a second poor-quality network segment.

[0008] In some embodiments, processing the poor-quality call record subset by using a pre-trained random forest model with corresponding dimensions includes: generating input features corresponding to the poor-quality call record subset, where the input features include the feature values of the key features of each call record in the poor-quality call record subset; and inputting the input features into the pre-trained random forest model with corresponding dimensions.

[0009] In some embodiments, the method for analyzing network faults further includes: calculating the information gain of each feature of a service according to the training data set corresponding to each service; for each service, determining a preset number of features with the largest information gain as the key features of the service.

[0010] In some embodiments, determining the poor-quality call records in the call record set includes: preliminarily determining the poor-quality call records in the call record set according to the preliminary screening features and corresponding preliminary screening thresholds of each type of service; and further screening the poor-quality call records in the call record set by using the pre-determined key features and corresponding thresholds.

[0011] In some embodiments, for call records of browsing services, the preset preliminary screening features include at least one of the first-screen latency of the page, the page opening latency, the RTT (Round Trip Time) uplink latency, the RTT downlink latency, or the HyperText Transfer Protocol (HTTP) download rate; for call records of video services, the preset preliminary screening features include at least one of the video download rate, the video stuttering frequency, and the rate-bitrate ratio; for call records of game services, the preset preliminary screening features include the game interaction latency; for call records of instant messaging services, the preset preliminary screening features include the message sending success rate.

[0012] In some embodiments, determining the root cause of the fault in the poor-quality network segment includes: determining the root cause of the fault according to the statistical proportion of each poor-quality index of the poor-quality network segment.

[0013] In some embodiments, the method for analyzing network faults further includes: for a network segment belonging to one of the first poor-quality network segment and the second poor-quality network segment, determining whether the network segment is a poor-quality network segment according to the alarm information.

[0014] In some embodiments, each network segment includes at least one of a core network, an SP (Service Provider), an IPRAN (IP Radio Access Network), a radio, or a terminal.

[0015] In some embodiments, each dimension includes at least one of a whole-network dimension, a cell dimension, a slice dimension, or a user dimension.

[0016] According to a second aspect of some embodiments of the present invention, there is provided an analysis device for network faults, including: a determination module configured to determine poor-quality call records in a call record set; a first poor-quality network segment determination module configured to determine a first poor-quality network segment according to the proportion of poor-quality call records in each network segment; a second poor-quality network segment determination module configured to, for each dimension, process the poor-quality call records in each network segment by using a pre-trained model corresponding to the dimension and key features to determine a second poor-quality network segment in the dimension; a poor-quality network segment determination module configured to determine a network segment that belongs to both the first poor-quality network segment and the second poor-quality network segment as the poor-quality network segment in the corresponding dimension; and a fault root cause determination module configured to determine the fault root cause of the poor-quality network segment.

[0017] According to a third aspect of some embodiments of the present invention, there is provided an analysis device for network faults, including: a memory; and a processor coupled to the memory, the processor being configured to execute any one of the foregoing network fault analysis methods based on instructions stored in the memory.

[0018] According to a fourth aspect of some embodiments of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements any one of the foregoing network fault analysis methods.

[0019] Some embodiments of the above invention have the following advantages or beneficial effects. The embodiments of the present invention perform a preliminary poor-quality judgment on each network segment in two ways, namely, based on statistics and based on a model, and then further confirm the poor-quality network segment through a cross-validation method, so that the poor-quality network segment can be accurately identified from the end-to-end communication full link. By analyzing and delimiting the fault cause and segment, the specific network segment where the cause lies is finally located, so that the fault cause can be analyzed more accurately, effectively guiding the operation and maintenance of the existing network and ensuring the user experience.

[0020] Other features and advantages of the present invention will become clear from the following detailed description of the exemplary embodiments of the present invention with reference to the accompanying drawings. Description of the Drawings

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0022] Figure 1 A flowchart showing an analysis method for network faults according to some embodiments of the present invention is shown.

[0023] Figure 2 The flowchart shows the network fault analysis method according to other embodiments of the present invention.

[0024] Figure 3 The structural diagram shows the analysis device for network faults according to some embodiments of the present invention.

[0025] Figure 4 The structural diagram shows the analysis device for network faults according to other embodiments of the present invention.

[0026] Figure 5 The structural diagram shows the analysis device for network faults according to still other embodiments of the present invention. Detailed implementation manners

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and in no way limits the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0028] Unless otherwise specifically stated, the relative arrangements, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0029] At the same time, it should be understood that, for the sake of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationships.

[0030] For technologies, methods, and devices known to those of ordinary skill in the relevant field, detailed discussions may not be made, but where appropriate, the technologies, methods, and devices should be regarded as part of the specification.

[0031] In all the examples shown and discussed here, any specific value should be construed as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values.

[0032] It should be noted that: like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0033] Users can perceive whether the communication service has poor quality. However, the poor quality perceived by users may be caused by various reasons, and the occurrence of these reasons cannot be limited to the radio side only. It may also be problems with the core network, IPRAN, SP, or the terminal. Therefore, the root cause analysis of poor quality should be end-to-end and cover all links involved in the communication process. However, related technologies focus more on classifying and analyzing various faults in a single network segment, lacking an end-to-end comprehensive fault judgment mechanism for the network. When multiple faults occur simultaneously in the network, the problem points cannot be locked in a timely and comprehensive manner. In addition, related technologies judge the result of poor quality through a single dimension, which is rather one-sided and there is a risk of inaccurate positioning. The present invention provides an analysis method, device, and computer-readable storage medium for network faults to solve the above problems.

[0034] Figure 1 shows a schematic flowchart of an analysis method for network faults according to some embodiments of the present invention. As Figure 1 shown, the analysis method for network faults in this embodiment includes steps S102 to S110.

[0035] In step S102, determine the poor-quality call records in the call record set.

[0036] The poor-quality call records can be determined by means of threshold screening or using a pre-trained call record classification model.

[0037] In some embodiments, according to the preset preliminary screening features and corresponding preliminary screening thresholds for each type of service, preliminarily determine the poor-quality call records in the call record set; use the pre-determined key features and corresponding thresholds to further screen the poor-quality call records in the call record set. The preliminary screening thresholds can be determined based on network service historical data and expert experience, and the thresholds of the key features used in the further screening can be re-determined, so as to more accurately determine the poor-quality call records through multi-layer screening. If necessary, it is also possible to perform only one screening based on the preliminary screening features and preliminary screening thresholds, or perform one screening using the key features and the preliminary screening thresholds of the key features, etc., which will not be elaborated here.

[0038] In some embodiments, for the call records of browsing services, the preset preliminary screening features include at least one of the first-screen latency of a page, the page opening latency, the RTT uplink latency, the RTT downlink latency, or the HTTP download rate; for the call records of video services, the preset preliminary screening features include at least one of the video download rate, the video stuttering frequency, and the rate-bitrate ratio; for the call records of game services, the preset preliminary screening features include the game interaction latency; for the call records of instant messaging services, the preset preliminary screening features include the message sending success rate. The key features can also be determined from these preliminary screening features. Of course, the service types involved in the embodiments of the present invention are not limited to the above several types, and the features in each type of service are not limited to the features listed above. Those skilled in the art can process other types of services as needed or use other features, which will not be elaborated here.

[0039] In some embodiments, the key features are determined according to the information gain: according to the training data set corresponding to each service, calculate the information gain of each feature of the service; for each service, determine the preset number of features with the largest information gain as the key features of the service.

[0040] The following exemplarily describes a method for determining key features.

[0041] 1. Obtain the training data set D and multiple alternative features, such as the preliminary screening features. For each service, select one of the multiple alternative features for training separately. The service types include, for example, browsing services, game services, video services, instant messaging services, and so on.

[0042] 2. Calculate the empirical entropy H(D) of the data set D.

[0043]

[0044] Among them, |D| represents the sample size, that is, the number of samples in the data set D. Let C k represent different categories, k = 1, 2,..., K, and |C k | represents the sample size of category k, then

[0045] 3. Calculate the empirical conditional entropy H(D|A) of the feature A with respect to the data set D. Let A i represent different metric features, where i = 1, 2,..., n. According to the different values of A i , divide D into n subsets D1, D2,..., D n , |D i | is the number of samples in D i , and there is Let D ik represent the subset D i that belongs to the class C kFor the sample set, there is D ik = D i ∩ C k , |D ik | is the number of samples in D ik .

[0046]

[0047] 4. Calculate the information gain g(D, A i ) of feature A i :

[0048] g(D, A i ) = H(D) - H(D|A i ) (3)

[0049] 5. Select the preset number of A i with the largest g(D, A i ) as the key features of this service.

[0050] By determining the key features, the dimensionality of the call record features can be reduced, reducing the complexity of later processing and improving the processing efficiency without affecting the judgment accuracy.

[0051] In step S104, according to the proportion of poor-quality call records in each network segment, the first poor-quality network segment is determined.

[0052] In some embodiments, calculate the proportion of poor-quality call records in each network segment in the call record set, for example, the ratio of the number of poor-quality call records to the total number of call records; for each network segment, when the proportion of poor-quality call records in the network segment is greater than the corresponding poor-quality threshold of the network segment, determine the network segment as the first poor-quality network segment. The poor-quality threshold can be a preset value.

[0053] In step S106, for each dimension, use the pre-trained model corresponding to the dimension and the key features to process the poor-quality call records of each network segment, and determine the second poor-quality network segment under the dimension.

[0054] In some embodiments, each dimension includes a whole-network dimension, a cell dimension, a slice dimension, a user dimension, etc. As needed, those skilled in the art can also process based on other dimensions, which will not be elaborated here.

[0055] For example, determine the poor-quality call records of the whole-network dimension, cell dimension, slice dimension, or user dimension respectively. Taking the user dimension as an example, obtain the poor-quality call records of each network segment of the core network, SP, IPRAN, wireless, and terminal of a certain user for a period of time. Then, use the pre-trained model to process the poor-quality call records of each network segment of the user respectively to obtain the classification result of whether each network segment is a second poor-quality network segment.

[0056] In some embodiments, the model is a random forest model. For the poor-quality call records of each network segment in each dimension: generate corresponding subsets of poor-quality call records; use the pre-trained random forest model corresponding to this dimension to process the subsets of poor-quality call records, and generate the poor-quality classification results of the corresponding network segments in this dimension, where the poor-quality classification results indicate whether the corresponding network segments are the second poor-quality network segments. As needed, it can also be other types of models, which will not be elaborated here.

[0057] In some embodiments, the random forest model can be used to process in the following manner: generate input features corresponding to the subsets of poor-quality call records, where the input features include the feature values of the key features of each call record in the subsets of poor-quality call records; input the input features into the pre-trained random forest model. Since each subset of poor-quality call records usually includes multiple call records, when determining the input features, for example, the feature value of a certain key feature can be set as the set of feature values of the corresponding key feature in each call record, or as the statistical information of these feature values (such as mean, median, variance, standard deviation, quantity, etc.). Thus, the model can determine whether the cause of the poor quality is in this network segment according to the characteristics of the poor-quality call records of a certain network segment in a certain dimension, and then obtain the judgment result on whether this network segment is the second poor-quality network segment.

[0058] In some embodiments, the random forest model can also be pre-trained according to the poor-quality data of each network segment in each dimension. For example, obtain the call record data corresponding to the network segments that have been marked as whether they are poor-quality network segments, and divide them into a training set and a test set; then, use the training set to train the model to determine the key features corresponding to each node in the random forest to construct the random forest; then input the data of the test set into the trained model to judge the usability of the model.

[0059] In step S108, the network segments that belong to both the first poor-quality network segments and the second poor-quality network segments are determined as the poor-quality network segments in the corresponding dimension.

[0060] For example, by analyzing the poor-quality call records in the cell dimension, it is judged that the wireless network segment is the second poor-quality network segment. At the same time, the wireless network segment is also determined as the first poor-quality network segment. At this time, this network segment can be determined as the poor-quality network segment in the cell dimension.

[0061] Steps S104 and S106 respectively use two methods to judge the poor-quality network segments. If a certain network segment is a poor-quality network segment under the judgment of both methods, it can be determined as the poor-quality network segment. Through this cross-validation method, the accuracy of the judgment is improved.

[0062] In step S110, determine the root cause of the failure of the poor-quality network segment. For example, determine the root cause of the failure according to the statistical proportion of each poor-quality index of the poor-quality network segment.

[0063] For a network segment that belongs to one of the first poor-quality network segment and the second poor-quality network segment, it is determined whether the network segment is a poor-quality network segment according to the alarm information.

[0064] The above embodiment makes a preliminary quality judgment on each network segment through two methods, one based on statistics and the other based on models, and then further confirms the poor quality network segment through cross-validation, so that the poor quality network segment can be accurately identified from the end-to-end communication link. By analyzing the delimitation and segmentation of the fault cause, the specific network segment where the cause is located can be finally located, so that the fault cause can be analyzed more accurately, effectively guiding the operation and maintenance of the existing network and ensuring the user's perception.

[0065] The output results of the present invention can be applied to the end-to-end perception monitoring and analysis system of mobile services, and alarm dispatch can be generated for the determined poor quality network segments to achieve closed-loop full-process automation. In addition, the solution of the present invention can also be applied to 4G and 5G networks. For example, root cause analysis of faults can be performed based on the user experience of network slices.

[0066] Reference below Figure 2 An application example of the network fault analysis method of the present invention is described.

[0067] Figure 2 FIG. 2 shows a flow chart of a network fault analysis method according to some other embodiments of the present invention. Figure 2 As shown, the network fault analysis method of this embodiment includes steps S202 to S214.

[0068] In step S202, a detailed list of DPI data of user services of the entire network is obtained, including user information, service type, service name, service index, network element information, wireless cell or base station information, etc.

[0069] In step S204, the DPI details are marked as poor quality based on the empirical value of the service indicator threshold.

[0070] Taking browsing business as an example, business indicators include page first screen delay, page opening delay, RTT uplink delay, RTT downlink delay, HTTP download rate, etc.

[0071] In step S206, these detailed lists are collected from the dimensions of the entire network and the cell to form a quality difference list.

[0072] In step S208, statistical classification is performed based on the data of the entire network to determine the first poor quality network segment.

[0073] For example, obtain data for three consecutive days from the quality - poor detail list of the entire network. Among these data, it is necessary to meet the conditions that the MME or SGW - U / PGW - U is continuously greater than or equal to two 3 - hour periods, the SP is continuously greater than or equal to two 6 - hour periods, and the IPRAN / wireless / terminal is continuously greater than or equal to 2 days. Then, according to the quality - poor labels of the call records, respectively count the proportion of the number of quality - poor records in paragraphs such as the core network, SP, IPRAN, wireless, and terminal. If the proportion of quality - poor call records in network segment i exceeds the threshold, then determine that network segment i is the first quality - poor network segment, denoted as Si_statistical_poor = 1; otherwise, Si_statistical_poor = 0.

[0074] Table 1 exemplarily shows the quality - poor determination results based on statistical classification.

[0075] Table 1

[0076] Value of i Network segment Proportion of poor quality Threshold value <![CDATA[S i _statistical_poor]]> 1 Core network 12％ 30％ 0 2 SP 8％ 35％ 0 3 IPRAN 18％ 35％ 0 4 Wireless 43％ 40％ 1 5 Terminal 22％ 40％ 0

[0077] In the example of Table 1, the wireless network segment is determined to be the first quality - poor network segment.

[0078] In step S210, determine the key features.

[0079] Taking the browsing service as an example, let the number of training set samples be |D|, the number of quality - poor samples be |C1|, and the number of non - quality - poor samples be |C2|. Take the empirical index (i.e., the preliminary screening feature) as feature A i , where A1 represents the first - screen delay of the page, A2 represents the page - opening delay, A3 represents the RTT upstream delay, A4 represents the RTT downstream delay, A5 represents the HTTP download rate, and so on.

[0080] Calculate the information gain of each index feature according to formulas (1)(2)(3) to obtain the information gain G i = g(D,A i ), and select the feature with large G i as the optimal feature of the service index, that is, the key feature.

[0081] Similarly, the key features of various services such as game services, video services, and instant messaging services can be obtained.

[0082] In step S212, based on the cell data, use the random forest model for classification to determine the second quality - poor network segment.

[0083] For example, extract the sample data of a certain cell for three consecutive days, and the time range of these data is the same as the time range of the data selected in step S208.

[0084] These data are processed using a random forest model to obtain classification results. For example, for network segment i, Si_random_poor = 1 indicates that the network segment is the second poorest quality network segment, and Si_random_poor = 0 indicates that the network segment is not the second poorest quality network segment.

[0085] Table 2 exemplarily shows the poor quality determination results based on the random forest. The "Day1", "Day2", and "Day3" columns respectively represent the Si_random_poor values corresponding to the data for each of the 3 days.

[0086] Table 2

[0087] Value of i Network segment Day1 Day2 Day3 1 Core network 0 0 0 2 SP 0 0 0 3 IPRAN 0 0 0 4 Wireless 1 1 1 5 Terminal 0 0 0

[0088] In the example of Table 2, the wireless network segment is determined to be the first poorest quality network segment.

[0089] In step S214, cross-validation is performed on the judgment results in steps S208 and S212.

[0090] Referring to Table 1 and Table 2, cross-validation is performed on the output results of statistical classification and random forest classification. It can be seen that in the analysis conclusions of both analysis methods, the wireless network segment is determined to be the poor quality network segment. Therefore, the root cause of the fault can be further determined based on the wireless network segment. Through further analysis, it is concluded that the root cause of the fault in the wireless network segment is concentrated in weak cell coverage.

[0091] Figure 3 Shows a schematic structural diagram of an analysis device for network faults according to some embodiments of the present invention. As Figure 3 shown, the analysis device 30 for network faults in this embodiment includes: a determination module 310 configured to determine poor quality call records in a call record set; a first poor quality network segment determination module 320 configured to determine a first poor quality network segment according to the proportion of poor quality call records in each network segment; a second poor quality network segment determination module 330 configured to, for each dimension, process the poor quality call records in each network segment using a pre-trained model corresponding to the dimension and key features to determine the second poor quality network segment under the dimension; a poor quality network segment determination module 340 configured to determine the network segments that belong to both the first poor quality network segment and the second poor quality network segment as the poor quality network segments under the corresponding dimension; and a fault root cause determination module 350 configured to determine the root cause of the fault of the poor quality network segment.

[0092] In some embodiments, the first poor quality network segment determination module 320 is further configured to calculate the proportion of poor quality call records in each network segment in the call record set; for each network segment, when the proportion of poor quality call records in the network segment is greater than the poor quality threshold corresponding to the network segment, the network segment is determined to be the first poor quality network segment.

[0093] In some embodiments, the second poor-quality network segment determination module 330 is further configured to, for the poor-quality call records of each network segment in each dimension: generate corresponding subsets of poor-quality call records; use a pre-trained random forest model corresponding to the dimension to process the subsets of poor-quality call records, and generate a poor-quality classification result of the corresponding network segment in the dimension, where the poor-quality classification result indicates whether the corresponding network segment is a second poor-quality network segment.

[0094] In some embodiments, the second poor-quality network segment determination module 330 is further configured to generate input features corresponding to the subsets of poor-quality call records, where the input features include the feature values of the key features of each call record in the subsets of poor-quality call records; and input the input features into a pre-trained random forest model corresponding to the dimension.

[0095] In some embodiments, the network fault analysis device 30 further includes: a key feature determination module 360, configured to calculate the information gain of each feature of a service according to the training data set corresponding to each service; and for each service, determine a preset number of features with the largest information gain as the key features of the service.

[0096] In some embodiments, the determination module 310 is further configured to preliminarily determine the poor-quality call records in the call record set according to the preset initial screening features and corresponding initial screening thresholds for each type of service; and further screen the poor-quality call records in the call record set by using the pre-determined key features and corresponding thresholds.

[0097] In some embodiments, for the call records of browsing services, the preset initial screening features include at least one of the page first-screen delay, page opening delay, RTT uplink delay, RTT downlink delay, or HTTP download rate; for the call records of video services, the preset initial screening features include at least one of the video download rate, video stuttering frequency, and rate-bitrate ratio; for the call records of game services, the preset initial screening features include the game interaction delay; for the call records of instant messaging services, the preset initial screening features include the message sending success rate.

[0098] In some embodiments, the fault root cause determination module 350 is further configured to determine the fault root cause according to the statistical proportion of each poor-quality index of the poor-quality network segment.

[0099] In some embodiments, the fault root cause determination module 350 is further configured to, for a network segment belonging to one of the first poor-quality network segment and the second poor-quality network segment, determine whether the network segment is a poor-quality network segment according to the alarm information.

[0100] In some embodiments, each network segment includes at least one of a core network, SP, IPRAN, radio, or terminal.

[0101] In some embodiments, each dimension includes at least one of a network-wide dimension, a cell dimension, a slice dimension, or a user dimension.

[0102] Figure 4 FIG. shows a schematic structural diagram of an analysis device for network faults according to other embodiments of the present invention. As Figure 4 shown, the analysis device 40 for network faults in this embodiment includes: a memory 410 and a processor 420 coupled to the memory 410. The processor 420 is configured to execute the analysis method for network faults in any of the foregoing embodiments based on instructions stored in the memory 410.

[0103] Among them, the memory 410 may include, for example, a system memory, a fixed non-volatile storage medium, etc. The system memory stores, for example, an operating system, application programs, a boot loader, and other programs.

[0104] Figure 5 FIG. shows a schematic structural diagram of an analysis device for network faults according to still other embodiments of the present invention. As Figure 5 shown, the analysis device 50 for network faults in this embodiment includes: a memory 510 and a processor 520, and may further include an input / output interface 530, a network interface 540, a storage interface 550, etc. These interfaces 530, 540, 550 and the memory 510 and the processor 520 may be connected through a bus 560, for example. Among them, the input / output interface 530 provides a connection interface for input / output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 540 provides a connection interface for various networking devices. The storage interface 550 provides a connection interface for external storage devices such as an SD card and a USB flash drive.

[0105] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. The program, when executed by a processor, implements the analysis method for network faults in any of the foregoing.

[0106] Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system, or a computer program product. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate a means for implementing the specified functions in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the specified functions in one or more blocks.

[0108] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the specified functions in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the specified functions in one or more blocks.

[0109] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the specified functions in the Figure 1 one or more flows and / or blocks Figure 1 means for implementing the specified functions in one or more blocks.

[0110] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for analyzing network faults, comprising: Determining poor-quality call records in a call record set; Determining a first poor-quality network segment according to the proportion of poor-quality call records in each network segment, where each network segment includes at least one of a core network, a service provider (SP), an Internet Protocol Radio Access Network (IPRAN) based on a network protocol, a wireless device, or a terminal; For each dimension, using a pre-trained model and key features corresponding to the dimension to process the poor-quality call records in each network segment, and determining a second poor-quality network segment under the dimension, including: for the poor-quality call records of each network segment under each dimension: generating a corresponding subset of poor-quality call records, and using a pre-trained random forest model corresponding to the dimension to process the subset of poor-quality call records to generate a poor-quality classification result of the corresponding network segment under the dimension, where the poor-quality classification result indicates whether the corresponding network segment is a second poor-quality network segment, and each dimension includes at least one of a whole network dimension, a cell dimension, a slice dimension, or a user dimension; Determining the network segments that belong to both the first poor-quality network segment and the second poor-quality network segment as the poor-quality network segments under the corresponding dimension; Determining the root cause of the faults in the poor-quality network segments.

2. The analysis method according to claim 1, wherein The determining the first poor-quality network segment according to the proportion of poor-quality call records in each network segment includes: Calculating the proportion of poor-quality call records in each network segment in the call record set; For each network segment, when the proportion of poor-quality call records in the network segment is greater than the poor-quality threshold corresponding to the network segment, determining the network segment as the first poor-quality network segment.

3. The analysis method according to claim 1, wherein, The using a pre-trained random forest model corresponding to the dimension to process the subset of poor-quality call records includes: Generating input features corresponding to the subset of poor-quality call records, where the input features include the feature values of the key features of each call record in the subset of poor-quality call records; Inputting the input features into a pre-trained random forest model corresponding to the dimension.

4. The analysis method according to claim 1, further comprising: Calculating the information gain of each feature of the service according to the training data set corresponding to each service; For each service, determining the preset number of features with the largest information gain as the key features of the service.

5. The analysis method according to claim 1, wherein, The determining the poor-quality call records in the call record set includes: Preliminarily determining the poor-quality call records in the call record set according to the preset initial screening features and corresponding initial screening thresholds for each type of service; Using the pre-determined key features and corresponding thresholds to further screen the poor-quality call records in the call record set.

6. The analysis method according to claim 5, wherein: For call records of browsing services, the preset initial screening features include at least one of the first-screen page delay, page opening delay, round-trip time (RTT) uplink delay, RTT downlink delay, or HyperText Transfer Protocol (HTTP) download rate; For call records of video services, the preset initial screening features include at least one of the video download rate, video stuttering frequency, or rate-bitrate ratio; For call records of game services, the preset initial screening features include the game interaction delay; For call records of instant messaging services, the preset initial screening features include the message sending success rate.

7. The analysis method according to any one of claims 1 to 6, wherein, Said determining the root cause of the fault of the poor quality network segment includes: Determining the root cause of the fault according to the statistical proportion of each poor quality index of the poor quality network segment.

8. The analysis method according to any one of claims 1 to 6 further includes: For a network segment belonging to one of the first poor quality network segment and the second poor quality network segment, determining whether the network segment is a poor quality network segment according to the alarm information.

9. An analysis device for network faults includes: A determination module configured to determine poor quality call records in a call record set; A first poor quality network segment determination module configured to determine a first poor quality network segment according to the proportion of poor quality call records of each network segment, where each of the network segments includes at least one of a core network, a service provider (SP), an Internet Protocol Radio Access Network (IPRAN) based on a network protocol, a radio or a terminal; A second poor quality network segment determination module configured to, for each dimension, use a pre-trained model and key features corresponding to the dimension to process the poor quality call records of each network segment to determine the second poor quality network segment under the dimension, including: for the poor quality call records of each network segment under each dimension: generating a corresponding subset of poor quality call records, using a pre-trained random forest model corresponding to the dimension to process the subset of poor quality call records, and generating a poor quality classification result of the corresponding network segment under the dimension, where the poor quality classification result indicates whether the corresponding network segment is a second poor quality network segment, and each of the dimensions includes at least one of a whole network dimension, a cell dimension, a slice dimension or a user dimension; A poor quality network segment determination module configured to determine the network segments that belong to both the first poor quality network segment and the second poor quality network segment as the poor quality network segments under the corresponding dimension; A fault root cause determination module configured to determine the root cause of the fault of the poor quality network segment.

10. An analysis device for network faults includes: A memory; And A processor coupled to the memory, the processor being configured to execute the network fault analysis method according to any one of claims 1 to 8 based on instructions stored in the memory.

11. A computer-readable storage medium having a computer program stored thereon, the program, when executed by a processor, implements the network fault analysis method according to any one of claims 1 to 8.

Images