Broadband network performance data analysis method and device, equipment and storage medium

Abstract

The invention discloses a broadband network performance data analysis method and device, equipment and a storage medium, relates to the technical field of household broadband operation and maintenance, and can improve the analysis efficiency and accuracy of broadband network performance data so as to improve the satisfaction degree of service. According to the scheme, the method comprises the following steps: acquiring network performance information, OLT equipment performance information and label information of a plurality of users at each time point of a target time period in a target area, and converting the network performance information, the OLT equipment performance information and the label information into a feature matrix; inputting the feature matrix into a target decision model to obtain an importance sequence of the plurality of pieces of first feature information and the plurality of pieces of second feature information; selecting multiple pieces of target feature information according to the importance sequence, and determining correlation among the target feature information; obtaining complaint log information of the user in the target area in the target time period; and inputting the multiple pieces of target feature information, the correlation and the complaint log information into a large language model to obtain a basic result that the network performance of the target area is reduced in the target time period.

Description

Technical Field

[0001] The present application relates to the field of broadband network operation and maintenance technology, and in particular to a method, device, equipment and storage medium for analyzing broadband network performance data. Background Art

[0002] As technology develops, data-driven decision making becomes increasingly important. With the growth of data and the advancement of technology, network operators need to effectively process and analyze large amounts of network performance data and optical line terminal (OLT) performance data to optimize and improve customer satisfaction. The key to improving customer satisfaction is to quickly and accurately find the root cause of customer complaints about network use and issue corresponding work orders to resolve them.

[0003] Traditional broadband network performance data analysis methods often rely on manual intervention and simple statistical methods, which are incapable of coping with complex and large-scale broadband network performance data, resulting in low data analysis efficiency and accuracy. Summary of the invention

[0004] The present application provides a method, apparatus, device and storage medium for analyzing broadband network performance data, which can improve the analysis efficiency and accuracy of broadband network performance data, thereby improving service satisfaction.

[0005] In order to achieve the above objectives, this application adopts the following technical solutions:

[0006] In a first aspect of an embodiment of the present application, a method for analyzing broadband network performance data is provided, the method comprising:

[0007] Obtain network performance information, OLT device performance information, and label information of multiple users in the target area at each time point in the target time period;

[0008] The target area and the target time period are areas and time periods with high user complaint rates, the tag information indicates satisfaction or dissatisfaction with the network performance information and the OLT device performance information at the current time point, the network performance information includes a plurality of first feature information, and the OLT device performance information includes a plurality of second feature information;

[0009] Converting the network performance information, OLT device performance information and label information of the plurality of users at each time point into a feature matrix;

[0010] Inputting the feature matrix into a preset target decision model to obtain importance rankings of the plurality of first feature information and the plurality of second feature information, wherein the importance rankings indicate importance rankings that affect network satisfaction;

[0011] Selecting a plurality of target feature information according to the importance ranking, and determining the correlation between each of the target feature information, wherein the target feature information is used to indicate feature information having a greater impact on network satisfaction;

[0012] Obtain complaint log information of users in the target area during the target time period;

[0013] The multiple target feature information, the correlation between each target feature information and the complaint log information are input into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

[0014] As a possible implementation, the target decision model is a decision model based on random forest, and the plurality of feature matrices are input into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, including:

[0015] The plurality of feature matrices are input into the random forest-based decision model, and the importance ranking of the plurality of first feature information and the plurality of second feature information is determined according to the reduction of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix.

[0016] As a possible implementation manner, determining the importance ranking of the plurality of first feature information and the plurality of second feature information according to the reduction amount of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix includes:

[0017] Obtaining a reduction amount of the Gini impurity of each feature information before and after splitting on each decision tree, to obtain a plurality of reduction amounts of the Gini impurity, wherein each feature information includes the first feature information and the second feature information;

[0018] Calculating an average of the reduction amounts of the plurality of Gini impurities to obtain a target reduction amount;

[0019] The importance of each feature information is sorted according to the target reduction amount.

[0020] As a possible implementation manner, determining the correlation between each of the target feature information includes:

[0021] Calculate the Pearson correlation coefficient between any two target feature information;

[0022] The correlation between each target feature information is determined according to the Pearson correlation coefficient.

[0023] As a possible implementation manner, before obtaining the network performance information of multiple users in the target area at each time point in the target time period, the method further includes:

[0024] Obtain the geographic location and complaint time of each complaining user;

[0025] Performing a first clustering process on the geographical location of each of the complaining users to obtain the target area;

[0026] A second clustering process is performed on the complaint times of the complaining users in the target area to obtain a target time period for the target area.

[0027] As a possible implementation manner, the first clustering process of the geographical location of each of the complaining users includes:

[0028] The geographic location of each of the complaining users is input into a preset clustering model, and a first clustering process is performed on the geographic location of each of the complaining users. The clustering model is a model based on a K-means clustering algorithm.

[0029] As a possible implementation manner, before inputting the plurality of target feature information, the correlation between each of the target feature information, and the complaint log information into a preset large language model, the method further includes:

[0030] Acquire a training set, wherein the training set includes a plurality of training samples and label information of each training sample, wherein one training sample includes a plurality of target feature information, correlations between each target feature information, and the complaint log information, wherein the label information is used to indicate a real text processing result of the training sample, and the real text processing result indicates a fundamental result of the degradation of network performance of the target area in the target time period;

[0031] The training set is input into a preset initial large language model, and based on the difference between the predicted text processing result and the real text processing result corresponding to the training sample, the training loss corresponding to the large language model is obtained. The large language model is iteratively updated multiple times according to the training loss until the training loss meets the preset conditions, so as to obtain the target large language model, and the prompt rules of the text processing results are pre-configured in the initial large language model.

[0032] According to a second aspect of an embodiment of the present application, a device for analyzing broadband network performance data is provided, the device comprising:

[0033] A first acquisition module is used to acquire network performance information, OLT device performance information and label information of multiple users in a target area at each time point in a target time period;

[0034] The target area and the target time period are areas and time periods with high user complaint rates, the tag information indicates satisfaction or dissatisfaction with the network performance information and the OLT device performance information at the current time point, the network performance information includes a plurality of first feature information, and the OLT device performance information includes a plurality of second feature information;

[0035] A first processing module, configured to convert the network performance information, OLT device performance information and label information of the plurality of users at each time point into a feature matrix;

[0036] A second processing module is used to input the feature matrix into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, wherein the importance ranking indicates the importance ranking affecting network satisfaction;

[0037] A determination module, used for selecting a plurality of target feature information according to the importance ranking, and determining the correlation between each of the target feature information, wherein the target feature information is used for indicating feature information having a greater impact on network satisfaction;

[0038] A second acquisition module is used to acquire complaint log information of users in the target area during the target time period;

[0039] The third processing module is used to input the multiple target feature information, the correlation between each target feature information and the complaint log information into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

[0040] According to a third aspect of an embodiment of the present application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for analyzing broadband network performance data in the first aspect of an embodiment of the present application is implemented.

[0041] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for analyzing broadband network performance data described in the first aspect of the embodiments of the present application is implemented.

[0042] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0043] The analysis method of broadband network performance data provided by the embodiment of the present application obtains network performance information, OLT equipment performance information and label information of multiple users in a target area at each time point in a target time period; wherein the target area and the target time period are areas and time periods with high user complaint rates, and the label information indicates satisfaction or dissatisfaction with the network performance information and OLT equipment performance information at the current time point, the network performance information includes multiple first feature information, and the OLT equipment performance information includes multiple second feature information; the network performance information, OLT equipment performance information and label information of the multiple users at each time point are converted into a feature matrix; the feature matrix is ​​input into a preset target decision model to obtain multiple first feature information and multiple second feature information. The importance ranking of information is as follows: the importance ranking indicates the importance ranking that affects network satisfaction; a plurality of target feature information is selected according to the importance ranking, and the correlation between each of the target feature information is determined, the target feature information is used to indicate the feature information that has a greater impact on network satisfaction; complaint log information of users in the target area in the target time period is obtained; the plurality of target feature information, the correlation between each of the target feature information and the complaint log information are input into a preset large language model to obtain the fundamental result of the degradation of network performance in the target area in the target time period, which can improve the analysis efficiency and accuracy of broadband network performance data, and because the fundamental result of the degradation of network performance in the target area in the target time period can be analyzed, the service satisfaction can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a method for analyzing broadband network performance data provided in an embodiment of the present application;

[0045] Figure 2 A structural diagram of a broadband network performance data analysis device provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of the internal structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "plurality" means two or more.

[0049] Additionally, the use of “based on” or “according to” is meant to be open and inclusive, as a process, step, calculation, or other action “based on” or “according to” one or more conditions or values ​​may, in practice, be based on additional conditions or beyond values.

[0050] As technology develops, data-driven decision making becomes increasingly important. With the growth of data volume and technological advancement, network operators need to effectively process and analyze large amounts of network performance data and OLT equipment performance data to optimize and improve customer satisfaction. The key to improving customer satisfaction is to quickly and accurately find the root cause of customer complaints about network use and issue corresponding work orders to resolve them.

[0051] Traditional broadband network performance data analysis methods often rely on manual intervention and simple statistical methods, which are incapable of coping with complex and large-scale broadband network performance data, resulting in low data analysis efficiency and accuracy.

[0052] In order to solve the above problems, the present application embodiment provides a method for analyzing broadband network performance data, such as Figure 1 As shown, the method comprises the following steps:

[0053] Step 101: Obtain network performance information, OLT device performance information, and label information of multiple users in a target area at each time point in a target time period;

[0054] Among them, the target area and the target time period are the area and time period with high user complaint rates, the label information indicates satisfaction or dissatisfaction with the network performance information and OLT equipment performance information at the current time point, the network performance information includes multiple first feature information, and the OLT equipment performance information includes multiple second feature information.

[0055] Network performance information includes: network speed, network latency, disconnection frequency, and other related information.

[0056] Network speed includes: download speed, upload speed, peak speed, average speed and speed fluctuation. Download speed: usually measured in Mbps (megabits per second), indicating the speed at which users download data from the network. Key to video streaming, web page loading, etc. Upload speed: measured in Mbps, indicating the speed at which users upload data to the network. Especially important for cloud backup, video conferencing, etc. Peak speed: the maximum speed when the network load is low. Average speed: the average network speed over a certain period of time, reflecting the general performance of the network. Speed ​​fluctuation: the change in speed at different times, reflecting the stability of the network.

[0057] Network latency includes: Ping response time, jitter, and route tracking. Ping response time: measured in milliseconds (ms), it measures the time it takes from sending a packet to receiving a response. It is critical for online games and real-time communications. Jitter: The rate of change in latency. High jitter can cause instability in video conferencing and online games. Route tracking: The path that a packet takes through the network to reach its destination. It helps identify delayed network nodes.

[0058] The frequency of disconnection includes: the number of interruptions, the duration of the interruption, and the recovery time. Number of interruptions: The number of times the network connection is interrupted within a specific period of time, reflecting the reliability of the network. Duration of interruption: The average duration of each interruption. Long interruptions indicate more serious network problems. Recovery time: The time required for the network to recover from an interruption to normal operation.

[0059] Other relevant information includes: bandwidth utilization, error rate and connection density. Bandwidth utilization: the percentage of network bandwidth used, reflecting whether the network is overloaded. Error rate: the frequency of errors during data transmission, including packet loss rate, etc. Connection density: the ratio of active users to OLT equipment capacity in a specific area. High density may cause network congestion.

[0060] OLT equipment performance information includes: signal strength, number of connected users, data throughput, response time, equipment operating status, OLT equipment maintenance and fault records.

[0061] Among them, signal strength: measures the optical signal strength of each OLT device, usually expressed in decibels (dBm). Signal strength indicates the quality of the connection between the user device and the OLT device. Number of connected users: refers to the number of users connected to each OLT device at the same time. A high number of users may cause network congestion. Data throughput: The amount of data processed by each OLT device, usually measured in Mbps (megabits per second). Reflects the load capacity of the network. Response time: The time it takes for the OLT device to respond to a user request. A long response time may mean network delay or congestion.

[0062] Equipment operating status: including the operating time, temperature, hardware status, etc. of the OLT equipment, used to monitor whether the equipment is operating normally. OLT equipment maintenance and fault records include: Maintenance activity records: recording the date, time and activity details of the maintenance and upgrade of the OLT equipment, such as hardware replacement, software upgrade, etc. Fault report: recording in detail the occurrence time, duration, scope of impact and fault type (such as hardware failure, software failure) of the network fault. Fault handling results: recording the specific measures and recovery time for fault resolution, reflecting the efficiency of network recovery. Maintenance plan: including scheduled maintenance plans and historical maintenance activities, which helps to analyze whether OLT equipment performance problems are related to lack of maintenance.

[0063] The above network performance information and OLT equipment performance information can be automatically collected through the network monitoring system. Performance data and logs. Alternatively, user complaint data can be collected from the customer service system and online feedback platform. Alternatively, APIs can be used to collect data from different data sources (such as OLT equipment management systems).

[0064] Step 102: convert the network performance information, OLT device performance information and label information of the multiple users at each time point into a feature matrix.

[0065] After obtaining the network feature information and OLT equipment performance information, the information needs to be preprocessed, where the preprocessing includes: data cleaning. Remove irrelevant, duplicate or incomplete records. Data conversion. Segmentation and standardization of text data (such as complaint content). Normalization or standardization of numerical data. For example, converting network speed into a unified unit of measurement. Processing missing values. Use appropriate methods to fill or delete missing values. Outlier detection. Identify and process abnormal or outlier data points. The final data can be a large matrix, that is, the feature matrix D in step 102, in which each row represents a separate data point (a time point, user's network performance data, OLT equipment data, etc.), and each column represents a feature information. This table has been sorted and cleaned to remove useless or erroneous information, ensuring that each row and column is useful and accurate, providing a solid data foundation for subsequent analysis.

[0066] Step 103: input the feature matrix into a preset target decision model to obtain importance rankings of the plurality of first feature information and the plurality of second feature information, wherein the importance rankings indicate importance rankings that affect network satisfaction.

[0067] Step 104: select multiple target feature information according to the importance ranking, and determine the correlation between each target feature information, wherein the target feature information is used to indicate feature information that has a greater impact on network satisfaction.

[0068] Step 105: Obtain complaint log information of users in the target area during the target time period;

[0069] Step 106: Input the multiple target feature information, the correlation between each target feature information and the complaint log information into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

[0070] The analysis method of broadband network performance data provided by the embodiment of the present application obtains network performance information, OLT equipment performance information and label information of multiple users in a target area at each time point in a target time period; wherein the target area and the target time period are areas and time periods with high user complaint rates, and the label information indicates satisfaction or dissatisfaction with the network performance information and OLT equipment performance information at the current time point, the network performance information includes multiple first feature information, and the OLT equipment performance information includes multiple second feature information; the network performance information, OLT equipment performance information and label information of the multiple users at each time point are converted into a feature matrix; the feature matrix is ​​input into a preset target decision model to obtain multiple first feature information and multiple second feature information. The importance ranking of information is as follows: the importance ranking indicates the importance ranking that affects network satisfaction; a plurality of target feature information is selected according to the importance ranking, and the correlation between each of the target feature information is determined, the target feature information is used to indicate the feature information that has a greater impact on network satisfaction; complaint log information of users in the target area in the target time period is obtained; the plurality of target feature information, the correlation between each of the target feature information and the complaint log information are input into a preset large language model to obtain the fundamental result of the degradation of network performance in the target area in the target time period, which can improve the analysis efficiency and accuracy of broadband network performance data, and because the fundamental result of the degradation of network performance in the target area in the target time period can be analyzed, the service satisfaction can be improved.

[0071] Optionally, the target decision model in step 1 is a decision model based on random forest, and the step 2 inputs the plurality of feature matrices into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, including:

[0072] The plurality of feature matrices are input into the random forest-based decision model, and the importance ranking of the plurality of first feature information and the plurality of second feature information is determined according to the reduction of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix.

[0073] Specifically, determining the importance ranking of the plurality of first feature information and the plurality of second feature information according to the reduction amount of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix includes:

[0074] Obtaining a reduction amount of the Gini impurity of each feature information before and after splitting on each decision tree, to obtain a plurality of reduction amounts of the Gini impurity, wherein each feature information includes the first feature information and the second feature information;

[0075] Calculating an average of the reduction amounts of the plurality of Gini impurities to obtain a target reduction amount;

[0076] The importance of each feature information is sorted according to the target reduction amount.

[0077] It is understandable that analyzing the data using a random forest-based decision model can pinpoint the features that are most important to user satisfaction. Accurately identifying and defining these key indicators helps to more effectively monitor and analyze network performance, thereby directly pointing to factors that may affect user satisfaction.

[0078] A random forest-based decision model makes predictions by building multiple decision trees. As it builds these trees, it looks at which features do the best job of distinguishing the data, such as which features are more effective at distinguishing fast network users from slow network users. A feature is considered "important" if it is frequently used in making key decisions across multiple trees.

[0079] Let the feature matrix be matrix D, Ti(D) is the i-th decision tree in the random forest, and featurej is one of the features, which corresponds to the j-th column data in D [d 1j ,d 2j ,…,d mj ], in this decision tree, we split the decision tree according to this feature threshold t and calculate the reduction in Gini non-existence degree.

[0080] Before this feature split, among the m samples, they can be divided into positive samples and negative samples according to the label information, where the positive samples (satisfactory) have u before , negative samples (unsatisfactory) v before , calculate the category ratio, the proportion of positive samples is u before / m, the ratio of negative samples is v before / m, so Gini does not exist at this moment (Impurity parent )for:

[0081]

[0082] So for this column of data [d 1j,d 2j ,…,d mj ], after passing the feature threshold t, it is divided into two groups, A and B. Those greater than the threshold t are divided into group A, and those less than the threshold t are divided into group B. The positive samples (satisfactory) in group A have u Aafter , negative samples (unsatisfactory) v Aafter ; Group B positive samples (satisfactory) have u Bafter , negative samples (unsatisfactory) v Bafter indivual;

[0083] So the Gini Impurity of Group A A )for:

[0084]

[0085] So the Gini Impurity of Group B B )for:

[0086]

[0087] So after the split, Gini does not exist (Impuritychil d )for:

[0088]

[0089] In each split process, if feature featurej is used for splitting, we will calculate the improvement of the model by this split. This improvement is measured by the reduction of impurity (Gini impurity). Therefore, the importance of feature featurej in a single tree Ti can be expressed as:

[0090] Imp(T i , feature j )=Impurity paren,t -Impurity child

[0091] Then, the overall importance of feature featurej in the random forest, FI(featurej), is the average of Imp(Ti,featurej) in all trees, which can be expressed as:

[0092]

[0093] Where N is the total number of decision trees in the random forest. In this way, FI(featurej) measures the average split improvement of feature featurej across all trees in the random forest, thus reflecting the importance of the feature.

[0094] Here, the sort function can be used to sort the features from high to low according to their importance scores, and [1:k] indicates selecting the top k features after sorting. Here, k can be 3. In this way, TopFeatures contains the top k features sorted by importance score, which is the target feature information in step 104. The selected target feature information can be used to guide subsequent analysis and decision-making.

[0095] Optionally, determining the correlation between each of the target feature information includes: calculating the Pearson correlation coefficient between any two of the target feature information;

[0096] The correlation between each target feature information is determined according to the Pearson correlation coefficient.

[0097] Specifically, for multiple target feature information, we want to know whether any two target feature information affect each other, and whether this influence is positive or negative. For example, assuming that the two target feature information are network latency and user complaints, we calculate the correlation between the two to see if they are related. If the correlation value is close to 1 or -1, it indicates that there is a strong positive or negative correlation between them. If the correlation value is close to 0, it indicates that there is no obvious correlation between them. In this way, we can understand which feature information has the greatest impact on user experience.

[0098] To calculate the correlation between these target feature information, the Pearson correlation coefficient can be used. For any two target feature information featurei and featurej, their correlation Corr(featurei,featurej) can be expressed as:

[0099]

[0100] For multiple target feature information, the output of the correlation calculation can be expressed as a correlation matrix R, where each element Rij represents the correlation between feature featurei and featurej:

[0101]

[0102] In the above formula, N is the total number of observations, that is, the number of rows in the matrix R, and featurei,n is the value of featurei in the nth row. and are the average values ​​of featurei and featurej respectively.

[0103] The matrix R shows the mutual correlation between all target feature information and can be used to further analyze which feature combinations have the greatest impact on user satisfaction.

[0104] Optionally, before obtaining network performance information of a plurality of users in the target area at each time point in the target time period, the method further includes:

[0105] Obtain the geographic location and complaint time of each complaining user;

[0106] Performing a first clustering process on the geographical location of each of the complaining users to obtain the target area;

[0107] A second clustering process is performed on the complaint times of the complaining users in the target area to obtain a target time period for the target area.

[0108] Specifically, the first clustering process of the geographical location of each of the complaining users includes:

[0109] The geographic location of each of the complaining users is input into a preset clustering model, and a first clustering process is performed on the geographic location of each of the complaining users. The clustering model is a model based on a K-means clustering algorithm.

[0110] Use the k-means clustering algorithm to analyze complaint data and identify areas and time periods where complaints are concentrated. Being able to clearly identify the geographic and temporal distribution of complaints helps determine which areas and time periods require priority attention. Input data mainly includes the geographic location and time information of user complaints. Specifically, it can include: Geographic location: the specific location of the user complaint, such as longitude and latitude or a specific geographic area code. Time information: the time when the user complaint occurred, which can be a specific date and time.

[0111] Assume that we have n complaint records, each record contains a geographic location (longitude longi and latitude lati) and time timei. The input data D can be expressed as:

[0112] D={(long1, lat1, time1), (long2, lat2, time2),..., (long n ,lat n ,time n )}

[0113] K-means cluster analysis is a method of dividing data points into several groups, where each group contains similar data points. We will divide the complaint records into several groups based on their geographical location and time. The K-means algorithm will try to find the center point of the group so that the complaint records within each group are as similar as possible (that is, the closer to the center point, the better), while different groups are as different as possible.

[0114] In the actual implementation, the center points are initialized. K initial cluster centers C = {c1, c2, ..., cK} are selected. These are randomly selected data points. Suppose we choose K = 3, that is, we want to divide complaints into 3 groups. First, 3 complaint records are randomly selected as the initial center points of each group.

[0115] Assign data points to the nearest cluster center. For each data point di in the data matrix D, find the cluster center closest to it. This assignment can be expressed mathematically as:

[0116] a i =argmin||d i -c i || 2

[0117] Where ai is the index of the cluster center to which point di is assigned. Each complaint record has a geographic location and a time point. We calculate the distance of each complaint record to the three center points and assign it to the group represented by the nearest center point. Distance can be imagined as geographic distance and temporal proximity.

[0118] Update the cluster center. For each cluster, update the cluster center to the average position of all points in the cluster. The updated cluster center cj can be expressed mathematically as:

[0119]

[0120] Among them, Sj is the set of all data points assigned to the cluster center cj, that is, the average location of the group is calculated as the new cluster center. For each group, we now update the center point, that is, the average geographic location and average time of all complaint records in the group. Repeat the above process iteratively. Until the cluster center no longer changes significantly or reaches a predetermined number of iterations. Or, until the group center no longer changes significantly or reaches a predetermined number of iterations. After each iteration, the center point of the group will more accurately reflect the hot spots and times of complaints.

[0121] The cluster centers cj output by the K-means clustering algorithm represent the center locations of the K groups in the data set. These cluster centers can be used to identify hot spots for operator user complaints. The output of K-means clustering is a set of cluster centers cj and a label indicating which cluster each record belongs to. Assuming we have divided into k clusters, the output can be expressed as:

[0122] C={(long1, lat1, time1), (long2, lat2, time2),..., (long k ,lat k ,timek )}

[0123] We get 3 groups, each representing a specific type of complaint pattern. For example: Group 1 may represent complaints in the downtown area during the morning rush hour on weekdays. Group 2 may represent complaints in the suburbs on weekend nights. Group 3 may represent complaints in the commercial area in the afternoon. Finally, the heat map and the peak period of complaints are output. Using the output of K-means, we can create a heat map on the map to visualize the hot spots of complaints. The center point (clongj, clatj) of each cluster represents a hot spot area, and the number of complaints in the area can be represented by the number of complaint records contained in this cluster. On the heat map, these hot spots can be marked with different colors or sizes. The warmer the color or the larger the size, the higher the number of complaints in the area. In this way, we can see at a glance which areas are high-incidence areas of complaints. Similarly, the center point (ctimei) of each cluster represents a hot time period. These hot time periods can be marked with different colors or sizes. The warmer the color or the larger the size, the higher the number of complaints in the time period.

[0124] Optionally, before inputting the plurality of target feature information, the correlations between the target feature information, and the complaint log information into a preset large language model, the method further includes:

[0125] Acquire a training set, wherein the training set includes a plurality of training samples and label information of each training sample, wherein one training sample includes a plurality of target feature information, correlations between each target feature information, and the complaint log information, wherein the label information is used to indicate a real text processing result of the training sample, and the real text processing result indicates a fundamental result of the degradation of network performance of the target area in the target time period;

[0126] The training set is input into a preset initial large language model, and based on the difference between the predicted text processing result and the real text processing result corresponding to the training sample, the training loss corresponding to the large language model is obtained. The large language model is iteratively updated multiple times according to the training loss until the training loss meets the preset conditions, so as to obtain the target large language model, and the prompt rules of the text processing results are pre-configured in the initial large language model.

[0127] It is understandable that the above training process is to fine-tune the original large language model based on the current training set to adapt to the current application scenario.

[0128] In the actual implementation process, combined with the above analysis results, the large language model is used to deeply analyze the network logs in the identified key time periods and areas to find out the specific root causes of the problem. This module can understand the problem more comprehensively and formulate more effective solution strategies.

[0129] The design of the above large language model includes: Input processing: receiving the output of the feature analysis module, including multiple target feature information, their correlations, and hot spots and time periods of complaints. Log screening: based on the input information, filter out logs of key areas and time periods from the network log database. Model integration: integrating the large language model and configuring appropriate prompts to guide the model to conduct focused log analysis. Analysis engine: input the filtered log data into the model for in-depth analysis. Output generation: extracting the results of the model analysis and generating an easy-to-understand report, including the root cause of the problem and the recommended solution.

[0130] In this step, the large language model integrates all previous findings, including analysis of key features, identification of hot spots, and patterns in network logs, to find the root causes of the concentration of user complaints.

[0131] The specific processing process includes: combining patterns extracted from network logs (such as traffic peaks, frequent disconnections or delays) with target feature information (such as network delays, disconnection frequency) identified by machine learning methods. In addition, consider user complaint data, especially focusing on hot spots and time periods of complaints. Compare the overlaps in time and location between different data sources. For example, check whether network performance issues (such as OLT equipment overload) coincide with high complaint periods and areas.

[0132] Look for common patterns or trends in all the data, which may be key clues to the problem. For example, is there an increase in complaints from a particular OLT device every time the traffic reaches a peak? Determine whether there is a causal relationship. For example, can it be reasonably inferred that the increase in network latency is the direct cause of the increase in user complaints. Based on the existing data and patterns, formulate hypotheses about the root cause of the problem. For example, assume that the high complaint rate in a certain area is due to the obsolete OLT equipment that cannot handle high traffic demands. Integrate all analyses and hypotheses to propose one or more possible root causes. These reasons should be able to explain all observed patterns and data.

[0133] For example, suppose that in an urban area, operators receive a large number of complaints about slow mobile network speeds. Through the data collection and feature analysis module, the top k important features and correlations, hot spots and times of complaints are identified, including: Network delay: significantly increased during the evening peak period. Disconnection frequency: abnormally high in specific areas (such as the city center). User density: high user density in the city center area is positively correlated with network performance issues. Then based on this, we give the large language model appropriate prompt information, for example: The prompt can be designed as: "Based on the following information, analyze the possible root causes of network problems: The most important network performance features are [network delay, disconnection frequency], and the correlation between them is [positive correlation]. The main complaints are concentrated in the [city center area], especially in the [evening peak period]. Please review the network logs for the corresponding period to identify the specific reasons that may lead to performance degradation."

[0134] After performing correlation analysis, the large language model found that network latency and disconnection frequency were positively correlated during the evening peak hours and in the city center. Using the large language model to deeply analyze the network logs in these time periods and areas, the model revealed that insufficient OLT equipment capacity was the main reason for the decline in network performance. Especially during the evening peak hours, due to the surge in the number of users, the existing OLT equipment could not effectively handle the high traffic demand. In addition, the analysis also pointed out that the OLT equipment in this area had not been upgraded or maintained recently, resulting in its performance being insufficient to cope with the traffic pressure during the peak period.

[0135] Based on these findings, operators can take specific measures, such as increasing OLT equipment capacity in city center areas, optimizing network traffic management strategies, or performing necessary equipment upgrades and maintenance to improve network performance and customer satisfaction.

[0136] like Figure 2 As shown, the analysis device of broadband network performance data provided by the embodiment of the present application includes:

[0137] The first acquisition module 11 is used to acquire network performance information, OLT device performance information and label information of multiple users in the target area at each time point in the target time period;

[0138] The target area and the target time period are areas and time periods with high user complaint rates, the tag information indicates satisfaction or dissatisfaction with the network performance information and the OLT device performance information at the current time point, the network performance information includes a plurality of first feature information, and the OLT device performance information includes a plurality of second feature information;

[0139] The first processing module 12 is used to convert the network performance information, OLT device performance information and label information of the multiple users at each time point into a feature matrix;

[0140] The second processing module 13 is used to input the feature matrix into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, wherein the importance ranking indicates the importance ranking affecting the network satisfaction;

[0141] A determination module 14 is used to select a plurality of target feature information according to the importance ranking, and determine the correlation between each of the target feature information, wherein the target feature information is used to indicate feature information having a greater impact on network satisfaction;

[0142] The second acquisition module 15 is used to acquire the complaint log information of users in the target area during the target time period;

[0143] The third processing module 16 is used to input the multiple target feature information, the correlation between each target feature information and the complaint log information into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

[0144] In one embodiment, the second processing module 13 is specifically used for:

[0145] The plurality of feature matrices are input into the random forest-based decision model, and the importance ranking of the plurality of first feature information and the plurality of second feature information is determined according to the reduction of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix.

[0146] In one embodiment, the second processing module 13 is specifically used for:

[0147] Obtaining a reduction amount of the Gini impurity of each feature information before and after splitting on each decision tree, to obtain a plurality of reduction amounts of the Gini impurity, wherein each feature information includes the first feature information and the second feature information;

[0148] Calculating an average of the reduction amounts of the plurality of Gini impurities to obtain a target reduction amount;

[0149] The importance of each feature information is sorted according to the target reduction amount.

[0150] In one embodiment, the determination module 13 is specifically configured to:

[0151] Calculate the Pearson correlation coefficient between any two target feature information;

[0152] The correlation between each target feature information is determined according to the Pearson correlation coefficient.

[0153] In one embodiment, the first acquisition module 11 is further used for:

[0154] Obtain the geographic location and complaint time of each complaining user;

[0155] Performing a first clustering process on the geographical location of each of the complaining users to obtain the target area;

[0156] A second clustering process is performed on the complaint times of the complaining users in the target area to obtain a target time period for the target area.

[0157] In one embodiment, the first acquisition module 11 is specifically used for:

[0158] The geographic location of each of the complaining users is input into a preset clustering model, and a first clustering process is performed on the geographic location of each of the complaining users. The clustering model is a model based on a K-means clustering algorithm.

[0159] In one embodiment, the apparatus further comprises a training module 17, wherein the training module 17 is configured to:

[0160] Acquire a training set, wherein the training set includes a plurality of training samples and label information of each training sample, wherein one training sample includes a plurality of target feature information, correlations between each target feature information, and the complaint log information, wherein the label information is used to indicate a real text processing result of the training sample, and the real text processing result indicates a fundamental result of the degradation of network performance of the target area in the target time period;

[0161] The training set is input into a preset initial large language model, and based on the difference between the predicted text processing result and the real text processing result corresponding to the training sample, the training loss corresponding to the large language model is obtained. The large language model is iteratively updated multiple times according to the training loss until the training loss meets the preset conditions, so as to obtain the target large language model, and the prompt rules of the text processing results are pre-configured in the initial large language model.

[0162] The broadband network performance data analysis device provided in this embodiment can execute the above-mentioned broadband network performance data analysis method embodiment, and its implementation principle and technical effect are similar, which will not be elaborated here.

[0163] For the specific definition of the analysis device for broadband network performance data, please refer to the definition of the analysis method for broadband network performance data above, which will not be repeated here. Each module in the above-mentioned analysis device for broadband network performance data can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor of the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0164] The execution subject of the broadband network performance data analysis method provided in the embodiment of the present application may be an electronic device, which may be a computer device, a terminal device, a server or a server cluster, and the embodiment of the present application does not specifically limit this.

[0165] Figure 3 This is a schematic diagram of the internal structure of an electronic device provided in an embodiment of the present application. Figure 3 As shown, the electronic device includes a processor and a memory connected via a system bus. The processor is used to provide computing and control capabilities. The memory may include a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The computer program can be executed by the processor to implement the steps of the analysis method for broadband network performance data provided in each of the above embodiments. The internal memory provides a cache operating environment for the operating system and computer program in the non-volatile storage medium.

[0166] Those skilled in the art will understand that Figure 3 The internal structure diagram of the electronic device shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0167] In another embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for analyzing broadband network performance data in the embodiment of the present application are implemented.

[0168] In another embodiment of the present application, a computer program product is also provided. The computer program product includes computer instructions. When the computer instructions are executed on the analysis device for broadband network performance data, the analysis device for broadband network performance data executes each step of the analysis method for broadband network performance data in the method flow shown in the above method embodiment.

[0169] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using a software program, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer execution instructions on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. Computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server, data center, etc. that contains one or more servers that can be integrated with a medium. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a DVD), or a semiconductor medium (eg, a solid state disk (SSD)).

[0170] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for analyzing broadband network performance data, characterized in that: The method comprises: Obtain network performance information, OLT device performance information, and label information of multiple users in the target area at each time point in the target time period; The target area and the target time period are areas and time periods with high user complaint rates, the tag information indicates satisfaction or dissatisfaction with the network performance information and the OLT device performance information at the current time point, the network performance information includes a plurality of first feature information, and the OLT device performance information includes a plurality of second feature information; Converting the network performance information, OLT device performance information and label information of the plurality of users at each time point into a feature matrix; Inputting the feature matrix into a preset target decision model to obtain importance rankings of the plurality of first feature information and the plurality of second feature information, wherein the importance rankings indicate importance rankings that affect network satisfaction; Selecting a plurality of target feature information according to the importance ranking, and determining the correlation between each of the target feature information, wherein the target feature information is used to indicate feature information having a greater impact on network satisfaction; Obtain complaint log information of users in the target area during the target time period; The multiple target feature information, the correlation between each target feature information and the complaint log information are input into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

2. The method according to claim 1, characterized in that The target decision model is a decision model based on random forests, and the feature matrix is ​​input into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, including: The plurality of feature matrices are input into the random forest-based decision model, and the importance ranking of the plurality of first feature information and the plurality of second feature information is determined according to the reduction of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix.

3. The method according to claim 2, characterized in that The step of determining the importance ranking of the plurality of first feature information and the plurality of second feature information according to the reduction amount of the Gini impurity of each decision tree in the random forest for each feature information in the feature matrix comprises: Obtaining a reduction amount of the Gini impurity of each feature information before and after splitting on each decision tree, to obtain a plurality of reduction amounts of the Gini impurity, wherein each feature information includes the first feature information and the second feature information; Calculating an average of the reduction amounts of the plurality of Gini impurities to obtain a target reduction amount; The importance of each feature information is sorted according to the target reduction amount.

4. The method according to claim 1, characterized in that: The determining the correlation between each of the target feature information includes: Calculate the Pearson correlation coefficient between any two target feature information; The correlation between each target feature information is determined according to the Pearson correlation coefficient.

5. The method according to claim 1, characterized in that Before obtaining network performance information of a plurality of users in the target area at each time point in the target time period, the method further includes: Obtain the geographic location and complaint time of each complaining user; Performing a first clustering process on the geographical location of each of the complaining users to obtain the target area; A second clustering process is performed on the complaint time of the complaining users in the target area to obtain a target time period for the target area.

6. The method according to claim 5, characterized in that The first clustering process of the geographical location of each of the complaining users includes: The geographic location of each of the complaining users is input into a preset clustering model, and a first clustering process is performed on the geographic location of each of the complaining users. The clustering model is a model based on a K-means clustering algorithm.

7. The method according to claim 1, characterized in that Before inputting the plurality of target feature information, the correlation between each target feature information and the complaint log information into a preset large language model, the method further includes: Acquire a training set, wherein the training set includes a plurality of training samples and label information of each training sample, wherein one training sample includes a plurality of target feature information, correlations between each target feature information, and the complaint log information, wherein the label information is used to indicate a real text processing result of the training sample, and the real text processing result indicates a fundamental result of the degradation of network performance of the target area in the target time period; The training set is input into a preset initial large language model, and based on the difference between the predicted text processing result and the real text processing result corresponding to the training sample, the training loss corresponding to the large language model is obtained. The large language model is iteratively updated multiple times according to the training loss until the training loss meets the preset conditions, so as to obtain the target large language model, and the prompt rules of the text processing results are pre-configured in the initial large language model.

8. A device for analyzing broadband network performance data, characterized in that: The device comprises: A first acquisition module is used to acquire network performance information, OLT device performance information and label information of multiple users in a target area at each time point in a target time period; The target area and the target time period are areas and time periods with high user complaint rates, the tag information indicates satisfaction or dissatisfaction with the network performance information and the OLT device performance information at the current time point, the network performance information includes a plurality of first feature information, and the OLT device performance information includes a plurality of second feature information; A first processing module, configured to convert the network performance information, OLT device performance information and label information of the plurality of users at each time point into a feature matrix; A second processing module is used to input the feature matrix into a preset target decision model to obtain the importance ranking of the plurality of first feature information and the plurality of second feature information, wherein the importance ranking indicates the importance ranking affecting network satisfaction; A determination module, used for selecting a plurality of target feature information according to the importance ranking, and determining the correlation between each of the target feature information, wherein the target feature information is used for indicating feature information having a greater impact on network satisfaction; The second acquisition module is used to acquire the complaint log information of users in the target area during the target time period; The third processing module is used to input the multiple target feature information, the correlation between each target feature information and the complaint log information into a preset large language model to obtain the fundamental result of the network performance degradation of the target area in the target time period.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for analyzing broadband network performance data according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method for analyzing broadband network performance data as described in any one of claims 1 to 7 is implemented.

Images