A message error identification code delimiting method and device

Abstract

The present invention provides a message error identification code delimitation method and device. The method includes: using an anomaly detection algorithm to process the total frequency sequence to obtain k abnormal moments corresponding to the k abnormal total frequencies; then, extracting data from the error code data table based on the k abnormal moments to obtain a real data table of the abnormal moments; thereafter, linearly fitting the frequency sequences corresponding to different error codes in the error code data table to obtain a corresponding prediction data table. Based on the k abnormal moments, data is extracted from the prediction data table to determine the abnormal error code type and the corresponding occurrence time to obtain a root cause feature list; based on the business indicator data, the correlation between the error code type and the business indicator data is determined for analyzing the cause of the abnormality. In this way, the rapid identification and delimitation of abnormal data can be achieved, which is suitable for multi-dimensional feature root cause analysis scenarios and has good universality.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular to a method and device for delimiting message error identification codes. Background Art

[0002] In the communications sector, with the large-scale rollout of mobile networks in recent years, 5G mobile network coverage has become increasingly complete. However, daily network maintenance has become increasingly burdensome, and network optimization issues have become increasingly prominent. Traditional methods rely on frontline engineers to collect network information, analyze the data, and then manually troubleshoot various network issues such as coverage, interference, and capacity. Based on their experience, they adjust base station parameters to optimize network issues. This optimization method is inefficient and inaccurate, requiring high-level personnel skills, but may not significantly improve user experience.

[0003] Existing techniques use generated gene signature combinations to match existing gene signature libraries to obtain error delimitation information. This method's accuracy depends on the quality and richness of the signature gene library, and has significant limitations on transferability, making it not universally applicable. Alternatively, a rough delimitation method is used to filter data indicators based on the rough delimitation results to determine the type of problem in the wireless cell. However, this method cannot pinpoint the root cause of this problem type. Summary of the Invention

[0004] In order to solve the above problems, embodiments of the present application provide a message error identification code delimiting method and apparatus.

[0005] In a first aspect, the present application provides a message error identification code delimiting method, which is applied to a mobile network and is characterized in that a total table of error code data and a total frequency sequence determined based on the total table are obtained, wherein the total table of error code data includes the real frequencies of N types of error codes appearing at M moments, and the total frequency sequence includes M total frequencies calculated for the N types of error codes arranged in order according to the M moments; the total frequency sequence is processed by an anomaly detection algorithm to obtain k abnormal moments corresponding to k abnormal total frequencies; data is extracted from the total table of error code data based on the k abnormal moments to obtain a real data table of the abnormal moments; and linear fitting is performed on the frequency sequences corresponding to different error codes in the total table of error code data. Obtain a corresponding prediction data summary table; extract data from the prediction data summary table based on the k abnormal moments to obtain a prediction data table for the abnormal moments, wherein the prediction data table for the abnormal moments indicates the predicted frequencies of N types of error codes at the k abnormal moments; based on the real data table and the prediction data table at the abnormal moments, determine the type of error code that has occurred and the corresponding time of occurrence to obtain a root cause feature list; based on the root cause feature list, extract the real frequencies of occurrence of the error code types that have occurred at the M moments from the error code data summary table; according to the business indicator data at the M moments, determine the correlation between the error code type and the business indicator data for analyzing the cause of the abnormality.

[0006] In one embodiment, obtaining the total table of original error code data includes: obtaining the error code report data at the M moments, and preprocessing the data after parsing, the preprocessing including: processing the null values and Nan values after the data parsing, replacing the Nan values with 0, and filling the null values with the actual frequency of the same type of error code at the previous adjacent moment.

[0007] In one embodiment, the process of processing the total frequency sequence using an anomaly detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies includes: setting a confidence interval, calculating an upper threshold of the total frequency sequence, and obtaining k abnormal moments corresponding to the k abnormal total frequencies based on the upper threshold.

[0008] In one embodiment, the method of determining the type of error code that occurs abnormally and the corresponding time of occurrence based on the real data table and the predicted data table of the abnormal moment to obtain a root cause feature list includes: calculating the explanatory power EP value and the unexpectedness S value of the real frequency and predicted frequency of N types of error codes corresponding to k abnormal moments in the real data table and the predicted data table of the abnormal moment; retaining r types of error codes whose explanatory power EP value is greater than a preset threshold, and sorting them in order according to the unexpectedness S value to obtain the root cause feature list, which includes the type of error code that occurs abnormally and the corresponding time of occurrence, where r is less than or equal to N.

[0009] In one embodiment, the correlation between the error code type and the business indicator data is determined based on the business indicator data at the M moments, and before analyzing the cause of the exception, it also includes: preprocessing the business indicator data, the preprocessing is: obtaining the business indicator data at the M moments, and preprocessing the data after parsing, the preprocessing includes: processing the null values and Nan values after the data is parsed, replacing the Nan values with 0, and filling the null values with the same type of business indicator values of the previous adjacent moment.

[0010] In one embodiment, determining the correlation between the error code type and the business indicator data based on the business indicator data at the M moments for analyzing the cause of the anomaly also includes: outputting an abnormal error code type greater than a second preset screening threshold.

[0011] In one embodiment, the method further includes: outputting a correlation list between the business indicators and abnormal error code types, and defining the abnormal error codes.

[0012] In one embodiment, the calculation formulas for the explanatory power EP value and the unexpectedness S value are:

[0013] Explanation ability EP value calculation formula:

[0014] EP=(A ij (m)-F ij ) / (A(m)-F(m))

[0015] Where A is the true value of the fault, F is the normal prediction value of the autoregressive moving average model ARMA, the subscript i is the dimension, j is the element, and m is the abnormal indicator.

[0016] The calculation formula of accidental S value is:

[0017]

[0018] The p-value is the predicted probability, and the formula is as follows:

[0019] p ij (m) = F ij (m) / F(m)

[0020] q is the true probability, and the formula is as follows:

[0021] q ij (m) = A ij (m) / A(m)

[0022] According to a second aspect, the present invention provides a message error identification code delimiting device, characterized in that the device comprises:

[0023] An abnormal moment acquisition unit is configured to obtain a total table of error code data and a total frequency sequence determined based thereon, wherein the total table of error code data includes the actual frequencies of occurrence of N types of error codes at M moments, and the total frequency sequence includes M total frequencies calculated for the N types of error codes arranged in order according to the M moments; the total frequency sequence is processed using an abnormality detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies; a real data table acquisition unit is configured to extract data from the total table of error code data based on the k abnormal moments to obtain a real data table of the abnormal moments; and a predicted data table acquisition unit is configured to obtain a corresponding predicted data table by performing linear fitting on the frequency sequences corresponding to different error codes in the total table of error code data; Based on the k abnormal moments, data is extracted from the predicted data summary table to obtain a predicted data table for the abnormal moments, which indicates the predicted frequency of N types of error codes at the k abnormal moments; a root cause feature list acquisition unit is configured to determine the type of error code that has occurred and the corresponding time of occurrence based on the real data table and the predicted data table at the abnormal moment to obtain a root cause feature list; a correlation acquisition unit extracts the real frequency of occurrence of the error code type that has occurred at the M moments from the error code data summary table based on the root cause feature list; and according to the business indicator data at the M moments, determines the correlation between the error code type and the business indicator data for analyzing the cause of the abnormality.

[0024] In one embodiment, the apparatus further comprises: a result output unit configured to output a correlation list between the business indicators and abnormal error code types, and to define the abnormal error code. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The following is a brief introduction to the drawings required for describing the embodiments or prior art.

[0026] Figure 1 A schematic diagram illustrating a flow chart of a message error identification code delimiting method provided in an embodiment of the present application is shown;

[0027] Figure 2 The calculation process of the interpretation capability EP value of N error code types provided by the embodiment of the present application is shown;

[0028] Figure 3 A schematic diagram illustrating a flow chart of a method for identifying and delimiting 5G message error codes provided in an embodiment of the present application is shown;

[0029] Figure 4 A schematic structural diagram of a message error identification code delimiting device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0031] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship. For example, A / B means either A or B.

[0032] The terms "first" and "second" in this specification and claims are used to distinguish different objects rather than to describe a specific order of objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of response messages.

[0033] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0034] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, for example, multiple processing units means two or more processing units, etc.; multiple elements means two or more elements, etc.

[0035] To quickly identify the root cause of error codes in mobile communication networks, existing techniques use generated gene signature combinations to match existing gene signature libraries to obtain error delimitation information. This method's accuracy depends on the quality and richness of the signature gene library, and its portability is severely limited, making it not universally applicable. Alternatively, a rough delimitation approach is used to filter data indicators based on the rough delimitation results to determine the type of problem in the wireless cell. However, this method cannot pinpoint the root cause of this problem type.

[0036] To address the problems in the above solutions, this application proposes a solution. This invention focuses on identifying and delimiting message error codes. The main strategy is to use algorithms such as time series prediction prophet to analyze abnormal data of message service error codes, and to use a multi-dimensional intelligent drill-down analysis method, Adtr ibutor, to determine the dimensional root cause of the abnormal data. By analyzing the correlation between dimensional root causes and business indicators, the root cause of 5G message service error codes can be delimited. This application uses 5G networks as an example for illustration.

[0037] The 5G network includes the core network, switching network, support network, and mobile terminals. Errors may occur within the network, at network interfaces, and at the interfaces between terminals and the network. Errors are generally reflected in the form of error codes and collected by the backend. For example, an error code of 0 indicates an unconfigured error. Based on the error code type, the link where the network error occurred can be preliminarily determined. The backend will execute a scheduled task to collect error codes and generate error code report data for a period of time. The error codes described below are error identification codes.

[0038] Figure 1 A flow chart of a message error identification code delimiting method provided by an embodiment of the present application is shown as follows: Figure 1 As shown:

[0039] Step S110: Obtain a summary table of error code data and a total frequency sequence determined based thereon, wherein the summary table of error code data includes the actual frequencies of occurrence of N types of error codes at M moments, and the total frequency sequence includes M total frequencies calculated for the N types of error codes in sequence according to the M moments;

[0040] Step S120, processing the total frequency sequence using an anomaly detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies;

[0041] Step S130, extracting data from the error code data table based on the k abnormal moments to obtain a real data table of the abnormal moments;

[0042] Step S140, performing linear fitting on the frequency sequences corresponding to different error codes in the error code data table to obtain a corresponding prediction data table;

[0043] Step S150: extracting data from the prediction data table based on the k abnormal moments to obtain a prediction data table for the abnormal moments, wherein the prediction data table for the abnormal moments indicates the predicted frequencies of N types of error codes at the k abnormal moments;

[0044] Step S160, based on the actual data table and the predicted data table at the abnormal moment, determining the error code type and the corresponding occurrence time of the abnormality, and obtaining a root cause feature list;

[0045] Step S170: extracting the actual frequencies of occurrence of the abnormal error code types at the M moments from the error code data summary table based on the root cause feature list;

[0046] Step S180: Determine the correlation between the error code type and the business indicator data based on the business indicator data at the M moments, so as to analyze the cause of the anomaly.

[0047] In step S110, obtaining the error code data summary table includes: obtaining error code report data at M moments, and preprocessing the data after parsing. The preprocessing includes: processing the null values and Nan values after the data parsing, replacing the Nan values with 0, and filling the null values with the actual frequency of the same type of error code at the previous adjacent moment.

[0048] After the backend collects error code report data for a certain period, it parses the data and counts various error code types within that period on a daily granularity. The format is: All time | Actual frequency of error code type 1 | Actual frequency of error code type 2 | ... | Actual frequency of error code type N. For example, 2023-06-01 | 20-Calling party shutdown | 0-Unconfigured error | 180-Call error ... . If the daily statistical value of an error code type is a NaN value, the NaN value is directly replaced with 0. If the daily statistical value of an error code type is null, the statistical value of the previous day for that error code type is used as the statistical value for that day.

[0049] In one embodiment, the error code data dimensions parsed from the error code report include: time | 1-unknown user | 10-SM sending failure | 11-message waiting queue full | 13-data loss | 14-unexpected data | 15-protocol error (MS end error) | 16-MS not equipped | 17-mobile phone memory full | 194-SMC system error | 2-undefined user | 20-incorrect SME address | 21-unknown SMC | 22-illegal device | 23-user busy | 24-user powered off | 255-undetermined error cause | 3-illegal user | 33-SMC did not obtain sufficient routing information | 34-from Unexpected data from HLR | 35 - Unexpected data from MSC | 4 - Telecommunications service not supported | 40 - No HLR response after sending a routing request | 41 - No HLR response after sending a set message | 48 - No MSC response | 5 - Call prohibited | 51 - MSC rejected | 52 - HLR rejected | 53 - GIW module rejected | 56 - HLR system error | 57 - MSC system error | 6 - Closed user group rejected | 63 - Destination signaling point or signaling transfer point cannot deliver the message | 7 - Device not supported | 75 - Not configured error | 8 - User out of service area | 18 - SC congestion | 83 - Not configured error | Total number of failures.

[0050] In step S120, the anomaly detection algorithm may be the time series prediction algorithm Prophet. The embodiment of the present application takes the time series prediction algorithm Prophet as an example for illustration, and other anomaly detection algorithms also fall within the scope of application of the present invention.

[0051] The total number of error codes of each type is counted at a daily granularity. As mentioned above, the error code data table for 30 moments from 2023-06-01 to 2023-06-30 is taken, and the error codes counted are 1, 10, 11, 13, 14, 15, 16, 17, 194, 2, 20, 21, 22, 23, 24, 255, 3, 33, 34, 35, 4, 40, 41, 48, 5, 51, 52, and 53. The number of error codes in the 37 error code dimensions, namely 56, 57, 6, 63, 7, 75, 8, 18, and 83, from 2023-06-01 to 2023-06-30 is calculated, and the total number of error codes corresponding to the 30 moments is obtained. The column names of the time series data table constructed in this way are ds and y, where ds represents time and y represents the total number of error codes at each moment. This is used as the input data of the time series prediction algorithm Prophet.

[0052] Input the time series data table into the Prophet time series prediction algorithm and set the confidence interval.

[0053] Preferably, the confidence interval is set to 0.9.

[0054] After setting the confidence interval, the time series prediction algorithm calculates the upper threshold of the input data and filters out abnormal data. The abnormal data are k abnormal moments corresponding to the total frequency of k abnormalities. In this embodiment, the abnormal moments are several moments under 30 moments.

[0055] In step S130, the error code type corresponding to the abnormal moment is selected from the error code data table using the abnormal moment as an index to obtain a real data table for the abnormal moment. The table format is: abnormal moment | real frequency value of error code type 1 | real frequency value of error code type 2 | real frequency value of error code type 3 | ... | real frequency value of error code type N.

[0056] In step S140 , a summary table of error code data is selected, and a linear fit is performed on the N error code types at M time points according to the frequency sequence of the N error codes to obtain the predicted frequency values of each error code type at M time points. The format of the summary table of predicted data is: all time points | predicted frequency of error code type 1 | predicted frequency of error code type 2 | predicted frequency of error code type 3 ... | predicted frequency of error code type N.

[0057] In step S150 , the prediction data table format of the abnormal time is: abnormal time|predicted frequency of error code type 1|predicted frequency of error code type 2 . . . |predicted frequency of error code type N.

[0058] In step S160, the explanatory power (EP) value and the unexpectedness (S) value of the actual frequency and the predicted frequency of N types of error codes corresponding to k abnormal moments in the actual data table and the predicted data table of the abnormal moments are calculated;

[0059] Retain r types of error codes whose explanation capability EP value is greater than a preset threshold, and sort them in order according to the unexpectedness S value to obtain the root cause feature list, which includes the abnormal error code type and the corresponding occurrence time, where r is less than or equal to N.

[0060] The explanatory power (EP) value for each dimension, namely the actual and predicted frequencies of error code types at each anomaly moment, is calculated. The greater the contribution of the element's fluctuation to the fluctuation of the abnormal Key Performance Indicator (KPI), the more likely the element is to explain the KPI anomaly. The EP value measures the element's ability to explain the anomaly.

[0061] The calculation formula of the explanatory power EP value is as follows (1):

[0062] EP = (A ij (m)-Fij ) / (A(m)-F(m)) (1)

[0063] Where A is the true value of the fault, F is the normal prediction value of the autoregressive moving average model ARMA, the subscript i is the dimension, j is the element, and m is the abnormal indicator.

[0064] The calculation formula of the accidental S value is as follows (2):

[0065]

[0066] The p value is the predicted probability, and the formula is as follows (3):

[0067] p ij (m) = F ij (m) / F(m) (3)

[0068] q is the true probability, and the formula is as follows (4):

[0069] q ij (m) = A ij (m) / A(m) (4)

[0070] Calculate the EP value and S value for N error code types. The N error code types include:

[0071] 1-Unknown user | 10-SM send failure | 11-Message queue full | 13-Data loss | 14-Unexpected data | 15-Protocol error (MS side error) | 16-MS not equipped | 17-Mobile phone memory full | 194-SMC system error | 2-Undefined user | 20-Incorrect SME address | 21-Unknown SMC | 22-Illegal device | 23-User busy | 24-User powered off | 255-Undetermined error cause | 3-Illegal user | 33-SMC did not obtain sufficient routing information | 34-Unexpected data from HLR | 35- Unexpected data from MSC | 4 - Telecommunications service not supported | 40 - No HLR response after sending a routing request | 41 - No HLR response after sending a set message | 48 - No MSC response | 5 - Call barred | 51 - MSC rejected | 52 - HLR rejected | 53 - GIW module rejected | 56 - HLR system error | 57 - MSC system error | 6 - Closed user group rejected | 63 - Destination signaling point or signaling transfer point cannot deliver the message | 7 - Device not supported | 75 - Not configured error | 8 - User out of service area | 18 - SC congestion | 83 - Not configured error

[0072] In one embodiment, the predetermined time period is one month, and the historical data of that month is selected. If the values of dimension i: 83-unconfigured errors for that month are all 0, then this error code data will not appear in the data table, and therefore will not affect the final calculation result.

[0073] Figure 2 The calculation process of the interpretation capability EP value of N error code types provided in the embodiment of the present application is shown as follows: Figure 2 As shown, the calculation process of the interpretation capability EP value of each error code type is as follows:

[0074] Step S210: Obtain the real frequency A of each element in the real data table at the abnormal moment ij (m).

[0075] Step S220: Obtain the predicted frequency F of each element in the prediction data table at the abnormal moment. ij .

[0076] Step S230: Calculate the total real frequencies (A(m)) of N error code types in the real data table at the abnormal moment.

[0077] Step S240 , calculating the total predicted frequency F(m) of N error code types in the prediction data table at the abnormal moment.

[0078] Step S250, using the real frequency A ij (m) and predicted frequency F ij The difference between the total true frequency and the total predicted frequency F(m) is divided by the difference between the total true frequency and the total predicted frequency F(m) to obtain the interpretation ability EP value of N error code types.

[0079] After obtaining the explanatory power EP value and unexpectedness S value of each element, the elements are screened according to the explanatory power EP value, and elements with an explanatory power greater than a second preset threshold are retained. Preferably, the second preset threshold is 0.1. Error code types with an explanatory power EP value greater than 0.1 are retained.

[0080] The retained error code types are sorted by the size of the S value. The error code type with a larger S value is closer to the root cause feature. The format of the root cause feature list is: abnormal time | filtered error code type. As shown in Table 1 below:

[0081] Table 1: List of root cause characteristics

[0082] Abnormal moments Error code type June 1, 2023 0 - Not configured error June 1, 2023 180-ring June 20, 2023 181-ring June 21, 2023 0 - Not configured error June 22, 2023 20-Calling party turns off the phone June 27, 2023 0 - Not configured error June 27, 2023 20-Calling party turns off the phone

[0083] In step S170 , F error code types are obtained according to the extracted root cause feature list, where the F error code types are part of the N error code types mentioned above.

[0084] Based on the F error code types obtained, select M time points in the error code data table to obtain the actual frequencies of the error code types corresponding to the M time points. The format is: M time points | actual frequency of error code type 1 | ... | actual frequency of error code type F.

[0085] In one embodiment, the F error code types are all of the above-mentioned N error code types.

[0086] In step S180, data is parsed from the daily traffic report to obtain business indicator data. The business indicator data is then preprocessed to obtain a business data table. The preprocessing step involves processing null and Nan values in the business indicator data, replacing Nan values with 0 and filling null values with the business indicator data value at the previous moment.

[0087] In one embodiment, the business indicator data includes: a sending success rate, a receiving success rate, etc.

[0088] The business indicator data table includes M moments and S business indicators corresponding to each moment. The format of the business indicator data table is: M moments | business indicator 1 | business indicator 2 ... | business indicator S.

[0089] In one embodiment, a correlation analysis is performed between the business indicator data table and the actual frequencies of the F error code types corresponding to the M moments obtained in step S170, focusing on business indicators with a high correlation with the root cause characteristics.

[0090] In one embodiment, the Pearson algorithm is used to perform correlation analysis to calculate the correlation between the root cause characteristics and the business indicators.

[0091] In one embodiment, correlations are calculated between S service indicators and F error code types. Based on the correlation coefficient, the error code type with the highest correlation with the service indicator is obtained. Correlations are calculated between the S service indicators and the actual frequency of each error code type. The correlation is calculated for a data set consisting of service indicator i and error code type j, where 1≤i≤S and 1≤j≤F, resulting in S×F data sets.

[0092] Preferably, the screening threshold is set to 0.4, and the screening results are processed and output. When the absolute value of the correlation coefficient is above 0.4, it is a high correlation, otherwise it is a low correlation.

[0093] In one embodiment, a list of correlations between business indicators and error codes is output to achieve rapid demarcation. Business indicators in a group of data with high correlation coefficients are business indicators that require special attention.

[0094] The following is a flowchart of the 5G message service error code identification and delimitation method. Figure 3 A flow chart of a method for identifying and delimiting 5G message error codes provided by an embodiment of the present invention is shown as follows: Figure 3As shown, 5G message error code data is received, parsed, and preprocessed. Using the Prophet time series prediction algorithm, anomaly time data is detected. Using the Anomalous Attribution Prophet time series prediction algorithm, a linear fit is performed on the 5G error code data to generate a summary table of predicted data. Based on the predicted time table and the actual and predicted data at the anomaly time, the explanatory power (EP) and unexpectedness (S) values are calculated for each error code type. Correlation analysis is then performed with business indicator data to generate a list of correlations between error code types and business indicators, enabling rapid demarcation.

[0095] In one embodiment, each error code type is associated with an explanatory power (EP) value and an unexpectedness (S) value. For example, the EP value for a 0-unconfigured error is 0.3 and the S value is 0.0031, while the EP value for a 180-ring error is 0.64 and the S value is 0.0014. Specifically, the EP values for both the 0-unconfigured error and the 180-ring error are greater than 0.1. Therefore, after error code type screening, these error code types are retained and then sorted by S value. The 0-unconfigured error has a greater S value than the 180-ring error and is ranked higher in the root cause feature list.

[0096] In one embodiment, according to a list of correlations between error code types and service indicators, for example, the correlation between 0-unconfigured error and MO success rate is 0.57, and the correlation between 180-ring and MT success rate is 0.78. The occurrence of the 180-ring error type has the highest correlation with the MT success rate. Based on the list, the root cause of the network problem can be quickly identified.

[0097] The embodiment of the present application is aimed at the identification and delimitation of message service error codes, among which it is very important to realize the identification and delimitation of 5G message service error codes. The main strategy uses algorithms such as the time series prediction algorithm prophet to analyze the abnormal data of 5G message service error codes, and uses the abnormal attribution Adtr ibutor multi-dimensional drill-down to determine the dimensional root cause of the abnormal data. Through the correlation analysis of the dimensional root cause and the business indicators, the root cause of the 5G message service error code is delimited. Based on the 5G message service error code data, an anomaly detection algorithm model and a multi-dimensional root cause drill-down algorithm are constructed to realize the identification and delimitation of 5G message service error codes. It should be noted that the embodiment of the present application can also realize the identification and delimitation of other message error codes.

[0098] The embodiment of the present application is based on algorithms such as the abnormal attribution Adtr ibutor and the time series prediction algorithm Prophet to implement a method for identifying and delimiting 5G message service error codes. First, abnormal data is screened through anomaly detection. Second, by fitting and predicting the data and combining the principles of the abnormal attribution Adtr ibutor algorithm, abnormal features at abnormal moments are screened to achieve multi-dimensional intelligent drill-down analysis. By combining Pearson with business feature data, rapid identification and delimitation of abnormal data are achieved. This method is suitable for multi-dimensional feature root cause analysis scenarios and has good universality.

[0099] According to the second aspect, the embodiment of the present application further provides a message error identification code delimiting device, Figure 4 A schematic diagram showing the structure of a message service error identification code delimiting device provided by an embodiment of the present application is shown. Figure 4 As shown, the apparatus 400 includes:

[0100] The abnormal moment acquisition unit 410 is configured to obtain a total table of error code data and a total frequency sequence determined based thereon, wherein the total table of error code data includes the actual frequencies of occurrence of N types of error codes at M moments, and the total frequency sequence includes M total frequencies calculated for the N types of error codes arranged in order according to the M moments; and process the total frequency sequence using an abnormality detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies.

[0101] The real data table acquisition unit 420 is configured to extract data from the error code data table based on the k abnormal moments to obtain a real data table at the abnormal moments; the predicted data table acquisition unit is configured to obtain a corresponding predicted data table by performing linear fitting on the frequency sequences corresponding to different error codes in the error code data table; and extract data from the predicted data table based on the k abnormal moments to obtain a predicted data table at the abnormal moments, wherein the predicted data table at the abnormal moments indicates the predicted frequency of N types of error codes at the k abnormal moments.

[0102] The root cause feature list acquisition unit 430 is configured to determine the error code type and the corresponding occurrence time of the abnormality based on the real data table and the predicted data table at the abnormality time, and obtain the root cause feature list.

[0103] The correlation acquisition unit 440 extracts the actual frequency of occurrence of the error code types that have caused the anomalies at the M moments from the error code data summary table based on the root cause feature list; and determines the correlation between the error code type and the business indicator data based on the business indicator data at the M moments, so as to analyze the cause of the anomaly.

[0104] In one embodiment, the apparatus further includes: a result output unit 450 configured to output a correlation list between the business indicators and abnormal error code types, and to define the abnormal error code.

[0105] In one embodiment, obtaining the original error code data table includes:

[0106] Obtain the error code report data at the M moments, and perform preprocessing on the data after parsing. The preprocessing includes: processing the null values and Nan values after the data parsing, replacing the Nan values with 0, and filling the null values with the actual frequency of the same type of error code at the previous adjacent moment.

[0107] In one embodiment, the process of processing the total frequency sequence using an anomaly detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies includes: setting a confidence interval, calculating an upper threshold of the total frequency sequence, and obtaining k abnormal moments corresponding to the k abnormal total frequencies based on the upper threshold.

[0108] In one embodiment, the method of determining the type of error code that occurs abnormally and the corresponding time of occurrence based on the real data table and the predicted data table of the abnormal moment to obtain a root cause feature list includes: calculating the explanatory power EP value and the unexpectedness S value of the real frequency and predicted frequency of N types of error codes corresponding to k abnormal moments in the real data table and the predicted data table of the abnormal moment; retaining r types of error codes whose explanatory power EP value is greater than a preset threshold, and sorting them in order according to the unexpectedness S value to obtain the root cause feature list, which includes the type of error code that occurs abnormally and the corresponding time of occurrence, where r is less than or equal to N.

[0109] It should be noted that for the description of the device in the figure, reference can also be made to the description of the aforementioned method.

[0110] According to another embodiment, there is also provided a computer readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute a combination of Figure 1 The method described.

[0111] According to another embodiment, a computing device is provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the system realizes the combination of Figure 1 Those skilled in the art will appreciate that in one or more of the above examples, the functions described herein may be implemented using hardware, software, firmware, or any combination thereof. When implemented using software, these functions may be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0112] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A message error identification code delimiting method, applied to a mobile network, characterized in that: Obtaining a summary table of error code data and a total frequency sequence determined based thereon, wherein the summary table of error code data includes the actual frequencies of occurrence of N types of error codes at M moments, and the total frequency sequence includes M total frequencies calculated for the N types of error codes, sequentially arranged according to the M moments; Processing the total frequency sequence using an anomaly detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies; Extracting data from the error code data table based on the k abnormal moments to obtain a real data table of the abnormal moments; By performing linear fitting on the frequency sequences corresponding to different error codes in the error code data table, a corresponding prediction data table is obtained; Extracting data from the prediction data table based on the k abnormal moments to obtain a prediction data table for the abnormal moments, wherein the prediction data table for the abnormal moments indicates the predicted frequencies of N types of error codes at the k abnormal moments; Based on the actual data table and the predicted data table at the abnormal time, determine the error code type and the corresponding occurrence time of the abnormality, and obtain a root cause feature list; Based on the root cause feature list, extracting the actual frequencies of occurrence of the abnormal error code types at the M moments from the error code data summary table; Based on the business indicator data at the M moments, the correlation between the error code type and the business indicator data is determined to analyze the cause of the anomaly.

2. The message error identification code delimiting method according to claim 1, characterized in that: The error code data acquisition table includes: Obtain the error code report data at the M moments, and perform preprocessing on the data after parsing. The preprocessing includes: processing the null values and Nan values after the data parsing, replacing the Nan values with 0, and filling the null values with the actual frequency of the same type of error code at the previous adjacent moment.

3. The message error identification code delimiting method according to claim 1, characterized in that: The method of processing the total frequency sequence using an anomaly detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies includes: A confidence interval is set, an upper threshold of the total frequency sequence is calculated, and based on the upper threshold, k abnormal moments corresponding to the k abnormal total frequencies are obtained.

4. The message error identification code delimiting method according to claim 1, wherein: The actual data table and the predicted data table at the abnormal time are used to determine the error code type and the corresponding occurrence time of the abnormality, and obtain a root cause feature list, including: Calculate the explanatory power EP value and unexpectedness S value of the actual frequency and predicted frequency of N types of error codes corresponding to k abnormal moments in the actual data table and the predicted data table of the abnormal moments; Retain r types of error codes whose explanation capability EP value is greater than a preset threshold, and sort them in order according to the unexpectedness S value to obtain the root cause feature list, which includes the abnormal error code type and the corresponding occurrence time, where r is less than or equal to N.

5. The message error identification code delimiting method according to claim 1, characterized in that: Before determining the correlation between the error code type and the business indicator data based on the business indicator data at the M moments for analyzing the cause of the anomaly, the method further includes: The business indicator data is preprocessed, and the preprocessing is: obtaining the business indicator data at the M moments, and preprocessing the data after parsing. The preprocessing includes: processing the null values and Nan values after the data parsing, replacing the Nan values with 0, and filling the null values with the same type of business indicator values at the previous adjacent moment.

6. The message service error identification code delimiting method according to claim 1, characterized in that: The determining, based on the business indicator data at the M moments, the correlation between the error code type and the business indicator data for analyzing the cause of the anomaly, further includes: The output is greater than the second preset screening threshold abnormal error code type.

7. The message error identification code delimiting method according to any one of claims 1 to 6, characterized in that: The method further comprises: Output a correlation list between the business indicators and abnormal error code types, and define the abnormal error codes.

8. The calculation formulas for the explanatory power EP value and the unexpectedness S value according to claim 4 are: Explanation ability EP value calculation formula: EP=(A ij (m)-F ij ) / (A(m)-F(m)) Where A is the true value of the fault, F is the normal prediction value of the autoregressive moving average model ARMA, the subscript i is the dimension, j is the element, and m is the abnormal indicator. The calculation formula of accidental S value is: The p-value is the predicted probability, and the formula is as follows: p ij (m)=F ij (m) / F(m) q is the true probability, and the formula is as follows: q ij (m)=A ij (m) / A(m) 9. A message error identification code delimiting device, characterized in that: The device comprises: an abnormal moment acquisition unit configured to acquire a total table of error code data and a total frequency sequence determined based thereon, the total table of error code data including the actual frequencies of occurrence of N types of error codes at M moments, the total frequency sequence including M total frequencies calculated for the N types of error codes arranged in sequence according to the M moments; and process the total frequency sequence using an abnormality detection algorithm to obtain k abnormal moments corresponding to the k abnormal total frequencies; a real data table acquisition unit configured to extract data from the error code data summary table based on the k abnormal moments to obtain a real data table at the abnormal moments; a prediction data table acquisition unit configured to obtain a corresponding prediction data table by performing linear fitting on frequency sequences corresponding to different error codes in the error code data table; extract data from the prediction data table based on the k abnormal moments to obtain a prediction data table for the abnormal moments, wherein the prediction data table for the abnormal moments indicates predicted frequencies of N types of error codes at the k abnormal moments; a root cause feature list acquisition unit configured to determine the error code type and the corresponding occurrence time of the abnormality based on the real data table and the predicted data table at the abnormality time, and obtain a root cause feature list; The correlation acquisition unit extracts the actual frequency of occurrence of the error code types that have caused the exceptions at the M moments from the error code data summary table based on the root cause feature list; and determines the correlation between the error code type and the business indicator data based on the business indicator data at the M moments, so as to analyze the cause of the exception.

10. The message error identification code delimiting device according to claim 9, characterized in that: The device further comprises: The result output unit is configured to output a correlation list between the business indicators and the abnormal error code types, and define the abnormal error code.

Images