Network quality prediction method, apparatus, device, medium, and product

Abstract

The method comprises: predicting a time window of network anomaly based on an association relationship between a historical business service result corresponding to a target account and historical network quality; performing association analysis on the historical business service result corresponding to the target account to obtain a historical business track; predicting future business use of the target account based on the historical business track corresponding to the target account; and predicting future network quality of the target account based on the future business use in the time window. The disclosure determines a network anomaly time period that may affect a user based on the relationship between the user and the network; meanwhile, business anomaly is independently predicted, and then the quality of experience of the user is determined according to the corresponding relationship between the business anomaly and the network anomaly, so that the independence and the correlation of network factors and business factors are retained, and the prediction is more accurate and comprehensive.

Description

Technical Field

[0001] This disclosure belongs to the field of network quality identification technology, and in particular relates to a network quality prediction method, apparatus, equipment, medium and product. Background Technology

[0002] User quality trajectory identification is a method that uses user behavior analysis probes to monitor network usage based on user behavior. However, current methods mainly focus on detection at specific times and do not consider the lag between user behavior and the perception of poor network quality, making it difficult to accurately determine the root cause of user behavior and thus failing to obtain accurate identification results. Summary of the Invention

[0003] This disclosure provides a network quality prediction method, apparatus, device, medium, and product that can improve the accuracy of network quality prediction from the user's perspective.

[0004] In a first aspect, embodiments of this disclosure provide a network quality prediction method, the method comprising: Based on the correlation between the historical business service results and historical network quality of the target account, the time window for network anomalies is predicted; Perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; Based on the historical business trajectory of the target account, predict the future business usage of the target account; Predict the future network quality of the target account based on future service usage within the time window.

[0005] In one feasible implementation, based on the correlation between the historical service results and historical network quality of the target account, the time window for network anomalies is predicted, including: User anomaly events are constructed based on service anomaly data extracted from historical service results and network quality parameters extracted from historical network quality. Based on the historical business service results of the target account in the target historical period, user abnormal events are marked within the target historical period to obtain the time distribution of user abnormal events; Based on the temporal distribution of user anomalies, the historical network quality trajectory of the target account is obtained. Based on historical network quality trajectories, predict the time window when network anomalies will occur in the target account.

[0006] In one feasible implementation, based on the historical business history of the target account, the future business usage of the target account is predicted, including: Based on historical business trajectories, extract the time series of each business where abnormal business changes occur. Based on the temporal correlation between time series corresponding to different businesses, multiple historical business combinations of the target account are obtained; Based on historical business portfolios, predict the future business portfolio of the target account and any abnormal changes in the future business portfolio.

[0007] In one feasible implementation, a correlation analysis is performed on the historical business service results corresponding to the target account to obtain the historical business trajectory, including: Extract the business types and business quality parameters of key businesses from historical business service results in time series format; Based on the parameter combinations and value ranges of the business quality parameters, multiple business states are constructed for each business type, where each business state includes at least an abnormal state. Based on the business type corresponding to the target account and the time distribution of each business status switching to the abnormal state, the historical business trajectory of the target account is obtained.

[0008] In one feasible implementation, based on the temporal correlation between time series corresponding to different businesses, multiple historical business combinations of the target account are obtained, including: For each business type, based on the switching from the corresponding business state to the abnormal state, a corresponding business abnormal period is generated; When different business types have overlapping or consecutive periods of abnormal business conditions, the different business types are treated as a single historical business combination.

[0009] In one feasible implementation, after predicting the future network quality of the target account based on future service usage within a time window, the method further includes: Based on the target account’s historical business portfolio, set a user field for the target account. The user field is used to identify the actual users using the target account. Based on user fields and future network quality, determine the network resource allocation strategy for the target account.

[0010] In one feasible implementation, predicting the future network quality of a target account based on future service usage within a time window includes: Extract the target account's business portfolio and any abnormal changes in the business portfolio within the extraction time window; For each business combination, based on the mapping relationship between abnormal business changes and the complaint behavior of target accounts, the probability of user complaints corresponding to the business combination is predicted. Based on the likelihood of user complaints, predict the future network quality of the corresponding service combinations.

[0011] In one feasible implementation, after predicting the future network quality of the corresponding service combination based on the likelihood of user complaints, the process includes: Based on the future network quality of each service combination, different levels of alarm signals are sent to the target account.

[0012] Secondly, embodiments of this disclosure provide a network quality prediction apparatus, the apparatus comprising: The network anomaly prediction module is used to predict the time window for network anomalies based on the correlation between the historical business service results and historical network quality of the target account. The business data generation module is used to perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; The business anomaly prediction module is used to predict the future business usage of a target account based on its historical business trajectory. The prediction output module is used to predict the future network quality of a target account based on future service usage within a time window.

[0013] Thirdly, embodiments of this disclosure provide a network quality prediction device, the device including: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the network quality prediction method.

[0014] Fourthly, embodiments of this disclosure provide a computer storage medium storing computer program instructions, which, when executed by a processor, implement the network quality prediction method described above.

[0015] Fifthly, embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the network quality prediction method described above.

[0016] The network quality prediction method, apparatus, device, medium, and product disclosed herein focus on the mapping relationship between historical service results and historical network quality. This allows for accurate determination of the specific time window in which future network quality anomalies are likely to occur. Even if the user does not engage in any related behavior, such as filing a complaint about network quality, the anomaly can still be detected promptly, enabling targeted handling, overcoming delays, and providing better user service. Furthermore, this disclosure considers the differences in network quality perception among different users, focusing on the mapping relationship between users and historical service results to obtain historical service trajectories with user characteristics. These trajectories are then used to predict the user's future service usage. Subsequently, cross-validation is used to match the service usage corresponding to the time window of network quality anomalies, and the matching results are used to predict the user's perceived future network quality. This not only improves the accuracy of prediction but also further highlights the characteristics of user perception. In short, based on the relationship between users and the network, the system identifies potential periods of network anomalies that may affect users; it also independently predicts business anomalies and subsequently determines user experience quality based on the correspondence between business anomalies and network anomalies, preserving both the independence and correlation between network and business factors, resulting in more accurate and comprehensive predictions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments of this disclosure will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of a network quality prediction method provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of a network quality prediction device provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of the structure of a network quality prediction device provided in an embodiment of this disclosure. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of this disclosure will now be described in detail. To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, will provide a further detailed description. It should be understood that the specific embodiments described herein are intended only to explain this disclosure and not to limit it. For those skilled in the art, this disclosure can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this disclosure by illustrating examples.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] The acquisition, storage, use, and processing of data in this application embodiment all comply with the relevant provisions of national laws and regulations.

[0022] In the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0023] To better understand and explain the solutions provided in the embodiments of this application, some technical terms involved in the embodiments of this application will be briefly introduced below.

[0024] Poor call quality refers to the decline in call quality caused by the deterioration of wireless signal quality in mobile communication networks. It is mainly manifested in problems such as call interruption, dropped calls, and call connection failure.

[0025] QoE (Quality of Experience) is a concept in the field of communications. It can be understood as user experience or user perception, that is, the subjective feeling of end users about the performance of services provided by mobile networks. It can be expressed in a near-quantifiable way to represent end users' experience and feelings about services and networks, and reflect the gap between the current quality of services and networks and user expectations.

[0026] User quality trajectory identification continuously tracks user behavior and status changes across key dimensions to identify the evolution path and patterns of their service quality experience, thereby predicting network quality or proactively intervening with appropriate measures. Currently, this is mainly achieved by deploying different user behavior analysis probes at the core, aggregation, and access layers of the network to conduct real-time monitoring of all services based on user behavior, providing information on network usage for services and applications. For example, simplifying the QoE assignment process through an objective fuzzification process, combining entropy methods and fuzzy logic, and balancing objective attribute weights with subjective judgments enhances the portability of decisions.

[0027] However, existing methods for analyzing user behavior tend to focus on specific points in time or time slices. User behavior often lags behind the perception of network quality deterioration. This is especially true in home broadband networks; observing changes in user network perception at a specific time slice or point in time cannot fully reveal the root causes of user dissatisfaction or complaints. Events occurring throughout the entire lifecycle of a home broadband product can directly contribute to a deterioration in user perception. Furthermore, existing methods do not consider the strong correlation between different users and services. In scenarios where network users simultaneously use different services or where different users have varying perceptions of network quality for the same service, failing to distinguish between services and users can easily lead to significant errors in user experience quality assessments. This can even cause user perception alarm systems to confuse the user's perception status for a particular alarm ticket, resulting in false alarms.

[0028] To address the problems of existing technologies, embodiments of this disclosure provide a network quality prediction method, apparatus, device, medium, and product. First, this disclosure focuses on the lag in user behavior or user perception. By generating historical trajectories of user network quality anomalies and user service history, it predicts future service usage. Here, service usage can be events constructed by this disclosure that encompass user behavior, thus determining poor network quality at a given point in time before actual user complaints, effectively overcoming lag. Simultaneously, this disclosure also distinguishes between differences in network quality perception for the same user regarding different services or for different users regarding the same service, identifies the services the user is using when network quality anomalies occur, and mines the correspondence between service anomalies occurring at the same point in time and service types. Based on user service behavior, network quality is evaluated from the user's perception perspective.

[0029] This network quality prediction method can be applied to scenarios where network quality is predicted, and the subject executing this method can be an electronic device with any node in the network.

[0030] The network quality prediction method provided in the embodiments of this disclosure will be introduced first below.

[0031] Figure 1 A flowchart illustrating a network quality prediction method according to an embodiment of this disclosure is shown. Figure 1 As shown, the method may include the following steps: S110, based on the correlation between the historical business service results and historical network quality of the target account, predict the time window for network anomalies; S120, perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; S130, based on the historical business trajectory of the target account, predict the future business usage of the target account; S140 predicts the future network quality of a target account based on future service usage within a time window.

[0032] Here, historical service results refer to the service outcomes provided to one or more users in the past. These results can include summaries of user alarm tickets, user complaint tickets, and user call detail records. The users referred to here can be broadband users or Internet Protocol Television (IPTV) users.

[0033] Based on this, focusing on the mapping relationship between historical service results and historical network quality allows for the accurate identification of specific time windows when future network quality anomalies are likely to occur. Even if users do not engage in any related behaviors, such as complaints about network quality, anomalies can still be detected promptly, enabling targeted intervention, overcoming delays, and providing better user service. Furthermore, this disclosure considers the differences in network quality perception among different users, focusing on the mapping relationship between users and historical service results to obtain historical service trajectories with user characteristics. These trajectories are then used to predict the user's future service usage. Subsequently, cross-validation is used to match the service usage within the time window of network quality anomalies, and the matching results are used to predict the user's perceived future network quality. This not only improves prediction accuracy but also further highlights the characteristics of user perception. In short, based on the relationship between users and the network, it identifies the time periods of network anomalies that may affect users; it also independently predicts service anomalies and subsequently determines user experience quality based on the correspondence between service and network anomalies, preserving both the independence and correlation between network and service factors, resulting in more accurate and comprehensive predictions.

[0034] The steps described above are explained in detail below.

[0035] Regarding S110, this disclosure first identifies potential network anomalies during specific time periods by analyzing the relationship between the target user and the network. Furthermore, it considers the time lag between the actual occurrence of network anomalies and the user's perceived anomalies, and predicts whether future network quality anomalies will occur based on historical network quality data.

[0036] In one optional implementation, based on the correlation between the historical service results and historical network quality of the target account, the time window for network anomalies is predicted, including: User anomaly events are constructed based on service anomaly data extracted from historical service results and network quality parameters extracted from historical network quality. Based on the historical business service results of the target account in the target historical period, the user abnormal events are marked within the target historical period to obtain the time distribution of the user abnormal events; Based on the temporal distribution of user anomalies, the historical network quality trajectory of the target account is obtained. Based on historical network quality trajectories, predict the time window when network anomalies will occur in the target account.

[0037] Here, service anomaly data can be user alarm tickets. Furthermore, network quality parameters can be determined from various aspects. For example, network quality parameters may include key indicators such as ONU received optical power, uplink traffic, downlink traffic, TCP retransmission rate, TCP connection latency, video stuttering / screen flickering rate, peak CPU utilization, and peak memory utilization. The network here can be a home network; for home networks, specific parameters such as optical modem CPU utilization, optical modem memory utilization, WAN port uplink traffic, and WAN port downlink traffic can also be used.

[0038] When constructing user anomaly event data, the temporal distribution of these events can be determined at different granularities, such as hours or days. Using user accounts (e.g., broadband accounts), the affected aspects and subsequent performance of each user are presented on a timeline along with their network quality parameters. These parameters are then used to mark user anomaly events occurring at the same point in time, resulting in a historical network quality trajectory. By analyzing the user anomaly events on this historical trajectory, we can help identify the specific sources of user malfunctions or disruptions to user experience, and determine the time windows in which future network anomalies are likely to occur.

[0039] Table 1 shows a summary table of user abnormal events provided in one embodiment of this disclosure. As shown in Table 1, the user events constructed in this disclosure include multiple types, such as fault events, alarm events, complaint events, operation events, and indicator-related events. Furthermore, multiple events with different names are set for each event type, and the content of events with the same name also differs due to the set judgment values / thresholds. A level is set for each specific event. Taking the ONU fault alarm event as an example, alarm rules are preset based on network quality parameters related to ONU faults. When the alarm rules are met, an ONU fault alarm event is confirmed to have occurred. At this time, the event type is fault, the level is severe, and it requires high attention.

[0040] The established levels are used to evaluate the severity of events, thereby characterizing the degree of network anomaly. Specifically, the number of events of different severity levels is used to evaluate the degree of broadband quality anomaly for the broadband user at the current point in time (e.g., today). The anomaly degree is divided into three levels: high, medium, and low. A high anomaly degree means that there are one or more severe events or five or more minor events on that day; a medium anomaly degree means that there are one or more minor events on that day; and a low anomaly degree means that there are no events on that day. Here, according to the anomaly degree from largest to smallest, the levels are: severe, trending, minor, and no action required. The no-action level refers to normal operation related to the equipment that actually occurred, and no action is needed in this case.

[0041] Table 1 Summary of User Abnormal Events

[0042]

[0043] It should be noted that predicting the time window for network anomalies to occur in a target account based on historical network quality trajectories can be achieved through anomaly prediction models.

[0044] In a specific example, this can be achieved as follows: Build and train an anomaly prediction model. Access user quality data from all customers over the past three years, read all complaint events, and select the network quality trajectory data of the user from the preceding two months based on the time of the complaint event. Record the number of occurrences of high, medium, and low anomalies in this sample, as well as the consecutive time periods with high anomalies, and assign labels to the samples. Use the number of high, medium, and low anomalies and the consecutive time periods with high anomalies in the two months preceding each user's complaint event as a training sample to create a dataset. Build an anomaly prediction model for user complaint events, dividing the entire dataset into training and test sets to train the model, enabling it to predict the likelihood of user complaints at a given point in time.

[0045] The anomaly prediction model employs a network model combining multi-source convolutional neural networks and support vector machines; the model training process is not detailed here. After training, the anomaly prediction model takes into account the number of high, medium, and low anomalies within a user's network over two months, as well as the consecutive periods of high anomaly. It then outputs whether the user will file a complaint and the time window in which such a complaint might occur. Here, if a user files a complaint, it is considered that the network quality is abnormal at that time, and the output time window for the complaint is the time window in which the network anomaly occurs.

[0046] This deep integration of network quality and user experience facilitates the provision of highly reliable and personalized services. First, it overcomes the limitations of holistic network monitoring by starting from the user's business perception, correlating service anomalies with network parameters to construct user-specific anomaly events, making predictions more flexible and accurate. Second, by constructing historical network quality trajectories, it establishes a temporal correlation pattern between user business anomalies and network quality, ensuring predictions are based on causality rather than just correlation, improving interpretability and reliability, and providing a high-quality data foundation for subsequent prediction processes. It can predict potential network anomaly time windows, allowing administrators to intervene proactively, such as switching networks, pre-allocating resources, and sending alerts, shifting from passive response to proactive protection and effectively improving user experience.

[0047] Regarding S120, the analysis is still based on the historical business service results used in S110 to ensure data source consistency. Combining the historical business service results, the user's historical business activities over time are determined.

[0048] In one optional implementation, a correlation analysis is performed on the historical business service results corresponding to the target account to obtain the historical business trajectory, including: Extract the business types and business quality parameters of key businesses from historical business service results in time series format; Based on the parameter combinations and value ranges of the business quality parameters, multiple business states are constructed for each business type, where each business state includes at least an abnormal state. Based on the business type corresponding to the target account and the time distribution of each business status switching to the abnormal state, the historical business trajectory of the target account is obtained.

[0049] Here, key business functions can be web pages, games, and videos; business quality parameters can be key indicators such as latency, success rate, and speed. Based on these parameters, business types are categorized into anomaly levels: "with anomaly" or "without anomaly." Anomaly level indicates that when anomalies are present, the business status of that type is considered abnormal.

[0050] Next, acquire business usage time data, record business types and the time when each business type switches to an abnormal state at granularity such as hours and days, determine business events, and generate historical business trajectories corresponding to the target timeline based on all business events that occur on the target timeline.

[0051] Table 2 shows a summary table of business events provided in one embodiment of this disclosure. As shown in Table 2, each business type includes multiple business events. Taking the network loading latency event as an example, it represents the total time from when a user initiates a page access request to when 90% of the page is loaded (excluding cases where the page fails to load), with an accuracy to the millisecond; the corresponding business type is a webpage. For each Top N webpage, the webpage loading latency is calculated using the following formula: Webpage loading latency = (Time when the page access request was initiated at the point where 90% of the page is loaded) / Total number of successful accesses multiplied by 100%, and then the average webpage loading latency value of the Top N webpages is obtained. For wired usage scenarios, a webpage loading latency exceeding 2.5 seconds; for WIFI usage scenarios, a webpage loading latency exceeding 3.2 seconds is considered a network loading latency event.

[0052] Table 2 Summary Table of Business Events

[0053]

[0054] In this way, by combining multiple business quality parameters to define business status, we can more comprehensively and accurately depict the user's actual business experience and avoid the one-sidedness of judging by a single parameter. We focus on the transition of business status to abnormal status, taking into account both the time-varying attributes of the status and the abnormal situations of the business. This allows us to determine the process of changes in the user's business status over time, so that the historical business trajectory contains the serialized characteristics of time dimension and status change, providing an accurate and comprehensive data foundation for subsequent correlation analysis.

[0055] Regarding S130, after obtaining historical business trajectories, the user's future business usage is predicted based on these trajectories to avoid lag in user perception and behavior.

[0056] In one optional implementation, based on the historical business history of the target account, the future business usage of the target account is predicted, including: Based on historical business trajectories, extract the time series of each business where abnormal business changes occur. Based on the temporal correlation between time series corresponding to different businesses, multiple historical business combinations of the target account are obtained; Based on historical business portfolios, predict the future business portfolio of the target account and any abnormal changes in the future business portfolio.

[0057] It should be noted that this disclosure does not predict abnormal changes in business based on business type, but rather divides all businesses under the target account into multiple business groups based on the time correlation of businesses. Each business group includes at least one business type. This approach uses business groups to predict future business usage and focuses on abnormal changes in future business usage by predicting abnormal changes in these business groups.

[0058] In a specific example, by acquiring historical business trajectory data, time-series-based data mining algorithms, such as autoregressive model (AR), moving average model (MA), and autoregressive moving average model (ARMA), are used to predict the periodic development trend of historical business trajectory data, thereby obtaining the future business portfolio of the target account and abnormal changes in the future business portfolio.

[0059] In this way, predicting future data based on historical data enables a shift from passively responding to problems to proactively warning of risks. By identifying business combinations based on real, data-driven temporal correlations between business operations, and using these business combinations as the smallest unit to predict business anomalies, the prediction results are more aligned with user habits (for example, if user A habitually watches videos after playing games, anomalies in both may be correlated), significantly improving the accuracy and practicality of the predictions.

[0060] Furthermore, this disclosure determines which businesses belong to the same historical business combination by examining the temporal correlation between the business status of different business types changing to an abnormal state.

[0061] In one optional implementation, based on the temporal correlation between time series corresponding to different services, multiple historical service combinations of the target account are obtained, including: For each business type, based on the switching from the corresponding business state to the abnormal state, a corresponding business abnormal period is generated; When different business types have overlapping or consecutive periods of abnormal business conditions, the different business types are treated as a single historical business combination.

[0062] In a specific example, this disclosure uses the Markov Monte Carlo (MCMC) method to calculate the correlation between two business types. In the MCMC method, business X has k states, and the business state corresponding to each business is either a normal state or an abnormal state. Therefore, business X at each time step can take either a normal state or an abnormal state, which can be viewed as a probability distribution. That is, at time t, the business state is a random variable, and each state value corresponds to a probability.

[0063] Table 3 shows a service state probability table provided in one embodiment of this disclosure. As shown in Table 3, the total service X includes service X1, service X2, and service X3. Generally, each service corresponds to multiple service events, and each service event has a corresponding service state. Therefore, each service type corresponds to multiple service states, namely state 1, ..., state k. a1 represents the probability of service type X1 in state 1 at a certain moment, and P1 represents the probability distribution of service type X1, P1 = (a1, a2, ..., ak). In this disclosure, X1 represents web page service, X2 represents video service, and X3 represents game service. Taking web page service as an example, the states of each service event of web page service—web page loading latency, web page opening success rate, and web page resource download success rate—are normal or abnormal, respectively. Therefore, web page service has six states.

[0064] Table 3 Business Status Probability Table

[0065] Based on the above, a matrix is ​​constructed from the probabilities of transitions between service states. Let i and j be two states of a Markov chain, and Tij represent the probability of transitioning from state i to state j, then Tij = P(X=j|X=i). The transition probabilities Tij between all states form a state transition probability matrix. Based on the abnormal changes of each service in historical data, the probability of abnormal transitions is calculated. When the probability of abnormal transitions between two services, such as webpage X1 and game X3, is close within a certain time period, that is, when these two services simultaneously experience abnormalities within a similar time frame, it is determined that the two services have a strong temporal correlation within that time period, possibly using the same device or the same broadband user A, and are therefore considered as the same service combination. Meanwhile, service video X2 may primarily be used by another user B, and is therefore considered as another service combination.

[0066] Furthermore, when a business portfolio contains multiple business types, it's possible to further determine whether there are traffic overlaps between these business types. Real-time traffic change data can be used to assess whether there are overlaps between the multiple services. If overlaps are found, it indicates a strong temporal correlation between the two services within that time period, thereby improving the accuracy of business portfolio analysis.

[0067] Specifically, although Webpage X1 and Game X3 consume data simultaneously, their peak data usage alternates, indicating a connection behavior. Connecting wired game data with wireless video / webpage data is the first priority for confirmation.

[0068] Thus, compared to abnormal time points, comparing abnormal time periods can more robustly capture the correlation between services, avoiding oversensitivity to short-lived, isolated anomalies and instead focusing on persistent anomaly patterns, thereby making the identified service combinations more stable and representative. Furthermore, overlapping time periods reflect concurrent network problems encountered when users simultaneously use multiple services (such as downloading and watching videos simultaneously), or when the same network failure event affects multiple services at the same time. Continuous time periods reflect the sustained impact of network problems or user device issues during service switching when users continuously engage in different service activities (such as starting a video call after finishing a game). The established criteria for determining the overlap or continuity of abnormal time periods are easily implemented through time interval comparison algorithms, making them relatively simple and comprehensive. Finally, the identified service combinations and their associated patterns can be directly used for network fault root cause analysis. For example, frequently overlapping combinations of anomalies may point to shared bandwidth congestion, while specific continuous patterns may point to intermittent problems with user-side devices.

[0069] Regarding S140, the predicted future service usage is matched with the predicted time window for network anomalies. Based on the matching result, the perceived future network quality for the target account is predicted. For example, if future service usage is normal within the predicted time window, it indicates good future network quality. As for the specific matching method, a second time window for future service usage anomalies can be extracted. The future network quality is determined by the temporal relationship between the network quality anomaly time window and the second time window for service usage anomalies. Many matching methods exist, and will not be elaborated upon here.

[0070] This disclosure is based on matching future service usage with the time window of network anomalies in the service portfolio, and introduces user complaint behavior. Through user complaint behavior, it emphasizes the user's unique perception of the anomaly, thereby making the predicted network quality more user-specific and personalized.

[0071] In one alternative implementation, predicting the future network quality of a target account based on future service usage within a time window includes: Extract the target account's business portfolio and any abnormal changes in the business portfolio within the extraction time window; For each business combination, based on the mapping relationship between abnormal business changes and the complaint behavior of target accounts, the probability of user complaints corresponding to the business combination is predicted. Based on the likelihood of user complaints, predict the future network quality of the corresponding service combinations.

[0072] In a specific example, the aforementioned process has already identified a first service combination consisting of web page X1 and game X3, and a second service combination consisting of video X2. Based on each service combination, the probability of a user complaining under that service combination is determined according to the correlation between the complaint behavior of the target account and the service anomaly. Here, a higher complaint probability indicates a greater likelihood of user complaints when the service combination experiences anomalies, and therefore, that service combination should be given more attention. Consequently, future network quality should be predicted based on this service combination.

[0073] More specifically, the probability of complaints is calculated separately for different service combinations: the correlation between abnormal data and user complaints for webpage X1 and game X3 is calculated as the complaint probability T1 for the first service combination. The correlation between abnormal data and user complaints for video X2 is calculated as the complaint probability T2 for the second service combination. If the difference between probabilities T1 and T2 is small, all services are retained, and network quality is predicted based on webpage X1, video X2, and game X3; if the difference between probabilities T1 and T2 is large, network quality is predicted based on the service combination with the higher probability value.

[0074] For example, the above process is implemented by building, training, and using a business forecasting model.

[0075] Features of historical business trajectory data are extracted based on the time dimension, including quarterly, monthly, weekly, holiday, and weekend characteristics. Different anomalies and their frequencies are used as features for different businesses, and the historical frequency of anomalies for different businesses is used as labels. A regression model is then built to predict the number of anomalies. Next, user quality data from all customers over the past three years is used. For each user complaint event, based on the time of the event, the business trajectory data for the preceding two months is selected from the user's business data. The business types and frequencies of the first and second business combinations executed within those two months are recorded and labeled. The business types and frequencies of the first and second business combinations in the two months preceding each user's complaint event are used as training samples to build a dataset, which is then divided into training and test sets. A business prediction model for user complaint events is then constructed and trained on the training and test sets to enable it to predict the likelihood of complaints when users use the first and second business combinations. The business prediction model and the anomaly prediction model can have the same structure, which will not be elaborated here.

[0076] In this way, by predicting the likelihood of complaints that are strongly correlated with users' subjective feelings, network quality can be reflected more accurately. The entire prediction process first predicts whether network anomalies will occur and the status of critical services. Then, it determines whether the network anomalies will affect the business mix of critical services, and whether the affected business mix is ​​likely to trigger user complaints. This triple-judgment decision-making fundamentally reduces false alarms caused by network fluctuations that are imperceptible to users, ensuring that every alarm corresponds to a real risk perceived by users, and greatly improving the accuracy of network quality prediction.

[0077] The aforementioned service combinations can represent the various service usage habits of the target account. From another perspective, users' long-term behavior remains largely unchanged. Based on predictions using historical long-term data, each service combination can also correspond to a user, meaning multiple users share the target account. However, each user's service usage differs, requiring corresponding data processing to improve the quality of the services disclosed herein.

[0078] In one alternative implementation, after predicting the future network quality of the target account based on future service usage within a time window, the method further includes: Based on the target account’s historical business portfolio, set a user field for the target account. The user field is used to identify the actual users using the target account. Based on user fields and future network quality, determine the network resource allocation strategy for the target account.

[0079] Here, "user" refers to the actual user of the target account. For example, if the first service combination (webpage X1 and game X3) appears during time period t, it means that user A is using the target account at that time. If the second service combination (video X2) appears during time period t, it means that user B is using the target account at that time. User A and user B can use the target account simultaneously or at different times. Since this disclosure uses service combinations as the smallest unit for prediction and processing, the predictions for user A and user B are independent of each other.

[0080] For example, a mapping relationship is generated between user network quality anomaly events and user service types existing at the same point in time. The user's service combination at the time the network quality anomaly event occurred is then identified; the service combination can be a single service or a set of multiple services. A quality mapping model that distinguishes users by service combination and links network anomaly datasets to user alarm tickets reflects the correspondence between user network quality anomaly events and users corresponding to different user service types.

[0081] Thus, this disclosure, based on user behavior mining, identifies multiple users potentially existing under the same account based on the strength of their business correlation. Then, it categorizes the businesses associated with different users, establishing business combinations. The constructed business prediction model also possesses the ability to predict the likelihood of complaints from different users within a target account. Based on user fields and future network quality, it can directly serve refined operational actions such as network resource pre-scheduling, targeted user care, and differentiated service guarantees. For example, if it predicts that a target account will exhibit a specific business combination of "games + videos" during a specific time period, then it is identified as user A. During peak evening hours with a high risk of anomalies, the corresponding network bandwidth and routing quality can be guaranteed for user A in advance.

[0082] In addition, to improve service quality, after predicting the future network quality of corresponding service combinations based on the likelihood of user complaints, the following measures are taken: Based on the future network quality of each service combination, different levels of alarm signals are sent to the target account.

[0083] It should be noted that since there may be multiple users for the target account, each user's perception of network quality will differ. The perceived network quality value for the same numerical value will vary among different users. Therefore, it is necessary to consider factors such as the warning method and frequency to issue alarm signals corresponding to each user.

[0084] In a specific example, based on network quality anomaly event prediction and user service prediction, alarms are issued for single services or sets of services that will occur at future network quality anomaly event times, according to the alarm mapping relationship. By differentiating the alarm status of different services, the user's network quality perception can be distinguished, enabling accurate anomaly alarms for anomalies that are clearly perceived by the user.

[0085] In this way, alarm signals can be issued in a way that is tailored to the business portfolio (users), avoiding frequent disturbances to low-awareness users, reducing false alarms and wasted computing power, or increasing service density for high-awareness users, thereby improving user experience. For example, for high-risk users with a high probability of complaints, proactive network optimization, resource allocation, or early customer care (such as sending optimization suggestion SMS messages) can be triggered, turning passive complaints into proactive service and improving customer satisfaction; for low-risk users, unnecessary maintenance resource investment can be avoided, enabling precise resource allocation.

[0086] In summary, this disclosure introduces historical trajectories of user network quality anomalies and user service-related historical trajectories. It establishes an alarm mapping relationship between single or multiple service anomalies and the anomaly dataset, pointing to user alarm work orders for home network quality parameters, based on the correspondence between user network quality anomalies at the same point in time and user service types and anomaly levels. At the time of occurrence of network quality anomalies predicted by the anomaly prediction model, the corresponding single or multiple service sets predicted by the service prediction model are identified. Only alarms are issued if they conform to the alarm mapping of the quality difference mapping model; otherwise, no alarm is issued, reducing false alarms. This improves the accuracy of network quality prediction and service quality.

[0087] Figure 2 This is a schematic diagram of a network quality prediction device provided in an embodiment of this disclosure. Figure 2 As shown, the network quality prediction device 200 may include a network anomaly prediction module 210, a service data generation module 220, a service anomaly prediction module 230, and a prediction result output module 230.

[0088] The network anomaly prediction module 210 is used to predict the time window for network anomalies based on the correlation between the historical business service results and historical network quality of the target account. The business data generation module 220 is used to perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; The business anomaly prediction module 230 is used to predict the future business usage of the target account based on the historical business trajectory of the target account. The prediction output module 240 is used to predict the future network quality of the target account based on future service usage within a time window.

[0089] Optionally, the network anomaly prediction module 210 is also used for: User anomaly events are constructed based on service anomaly data extracted from historical service results and network quality parameters extracted from historical network quality. Based on the historical business service results of the target account in the target historical period, user abnormal events are marked within the target historical period to obtain the time distribution of user abnormal events; Based on the temporal distribution of user anomalies, the historical network quality trajectory of the target account is obtained. Based on historical network quality trajectories, predict the time window when network anomalies will occur in the target account.

[0090] Optionally, the business anomaly prediction module 230 is also used for: Based on historical business trajectories, extract the time series of each business where abnormal business changes occur. Based on the temporal correlation between time series corresponding to different businesses, multiple historical business combinations of the target account are obtained; Based on historical business portfolios, predict the future business portfolio of the target account and any abnormal changes in the future business portfolio.

[0091] Optionally, the business data generation module 220 is also used for: Extract the business types and business quality parameters of key businesses from historical business service results in time series format; Based on the parameter combinations and value ranges of the business quality parameters, multiple business states are constructed for each business type, where each business state includes at least an abnormal state. Based on the business type corresponding to the target account and the time distribution of each business status switching to the abnormal state, the historical business trajectory of the target account is obtained.

[0092] Optionally, the business anomaly prediction module 230 is also used for: For each business type, based on the switching from the corresponding business state to the abnormal state, a corresponding business abnormal period is generated; When different business types have overlapping or consecutive periods of abnormal business conditions, the different business types are treated as a single historical business combination.

[0093] Optionally, the prediction result output module 240, after predicting the future network quality of the target account based on future service usage within a time window, is further used for: Based on the target account's historical business portfolio, set a user field for the target account. The user field represents the actual users using the target account. Based on user fields and future network quality, determine the network resource allocation strategy for the target account.

[0094] Optionally, the prediction result output module 240 is used for: Extract the target account's business portfolio and any abnormal changes in the business portfolio within the extraction time window; For each business combination, based on the mapping relationship between abnormal business changes and the complaint behavior of target accounts, the probability of user complaints corresponding to the business combination is predicted. Based on the likelihood of user complaints, predict the future network quality of the corresponding service combinations.

[0095] Optionally, after predicting the future network quality of the corresponding service combination based on the probability of user complaints, the prediction result output module 240 is further used for: Based on the future network quality of each service combination, different levels of alarm signals are sent to the target account.

[0096] Figure 3 A schematic diagram of the hardware structure for network quality prediction provided in an embodiment of this disclosure is shown.

[0097] The network quality prediction device may include a processor 301 and a memory 302 storing computer program instructions.

[0098] Specifically, the processor 301 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this disclosure.

[0099] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 302 may include removable or non-removable (or fixed) media, or memory 302 may be non-volatile solid-state memory. Memory 302 may be internal or external to the integrated gateway disaster recovery device.

[0100] In one instance, memory 302 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0101] Memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Therefore, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this disclosure.

[0102] The processor 301 reads and executes computer program instructions stored in the memory 302 to achieve... Figure 1 The network quality prediction method in the illustrated embodiment.

[0103] In one example, the network quality prediction device may also include a communication interface 303 and a bus 304. Wherein, for example... Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0104] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this disclosure.

[0105] Bus 304 includes hardware, software, or both, that couples components of a network quality prediction device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in embodiments of this disclosure, this disclosure contemplates any suitable bus or interconnect.

[0106] Furthermore, in conjunction with the network quality prediction methods described in the above embodiments, this disclosure can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the network quality prediction methods described in the above embodiments.

[0107] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the network quality prediction methods described in the above embodiments.

[0108] It should be clarified that this disclosure is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this disclosure is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this disclosure.

[0109] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this disclosure are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0110] It should also be noted that the exemplary embodiments mentioned in this disclosure describe methods or systems based on a series of steps or apparatus. However, this disclosure is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0111] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0112] The above description is merely a specific embodiment of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this disclosure is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this disclosure, and these modifications or substitutions should all be covered within the protection scope of this disclosure.

Claims

1. A network quality prediction method, characterized in that, The method includes: Based on the correlation between the historical business service results and historical network quality of the target account, the time window for network anomalies is predicted; Perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; Based on the historical business trajectory corresponding to the target account, predict the future business usage of the target account; Based on the future service usage within the time window, predict the future network quality of the target account.

2. The network quality prediction method according to claim 1, characterized in that, The prediction of the time window for network anomalies based on the correlation between the historical service results and historical network quality of the target account includes: Based on the service anomaly data extracted from the historical service results and the network quality parameters extracted from the historical network quality, user anomaly events are constructed. Based on the historical business service results of the target account in the target historical period, the user abnormal events are marked within the target historical period to obtain the time distribution of the user abnormal events; Based on the temporal distribution of the user's abnormal events, the historical network quality trajectory of the target account is obtained; Based on the historical network quality trajectory, the time window for the target account to experience network anomalies is predicted.

3. The network quality prediction method according to claim 1, characterized in that, The step of predicting the future business usage of the target account based on the historical business trajectory corresponding to the target account includes: Based on the historical business trajectory, extract the time series of each business where abnormal business changes occur; Based on the temporal correlation between the time series corresponding to different businesses, multiple historical business combinations of the target account are obtained; Based on the historical business portfolio, predict the future business portfolio of the target account and any abnormal changes in the business portfolio of the future business portfolio.

4. The network quality prediction method according to claim 3, characterized in that, A correlation analysis is performed on the historical business service results corresponding to the target account to obtain the historical business trajectory, including: Extract the business types and business quality parameters of key businesses from historical business service results in time series format; Based on the parameter combinations and value ranges of the aforementioned service quality parameters, multiple service states are constructed for each service type, wherein each service state includes at least an abnormal state. Based on the business type corresponding to the target account and the time distribution of each business state switching to the abnormal state, the historical business trajectory of the target account is obtained.

5. The network quality prediction method according to claim 4, characterized in that, The method of obtaining multiple historical business combinations for the target account based on the temporal correlation between the time series corresponding to different businesses includes: For each business type, a corresponding business exception period is generated based on the switching from the corresponding business state to the exception state. When the abnormal periods of different business types overlap or are consecutive, the different business types are considered as a single historical business combination.

6. The network quality prediction method according to claim 5, characterized in that, After predicting the future network quality of the target account based on the future service usage within the time window, the method further includes: Based on the target account’s historical business combinations, a user field is set for the target account, and the user field is used to characterize the actual users using the target account; Based on the user field and the future network quality, a network resource allocation strategy is determined for the target account.

7. The network quality prediction method according to claim 3, characterized in that, The prediction of the future network quality of the target account based on the future service usage within the time window includes: Extract the business combination of the target account within the time window and the abnormal changes in the business combination; For each business combination, based on the mapping relationship between the abnormal changes in the business and the complaint behavior of the target account, the probability of user complaints corresponding to the business combination is predicted; Based on the likelihood of user complaints, predict the future network quality of the corresponding service combination.

8. The network quality prediction method according to claim 7, characterized in that, After predicting the future network quality of the corresponding service combination based on the likelihood of user complaints, the following steps are included: Based on the future network quality of each service combination, different levels of alarm signals are sent to the target account.

9. A network quality prediction device, characterized in that, The device includes: The network anomaly prediction module is used to predict the time window for network anomalies based on the correlation between the historical business service results and historical network quality of the target account. The business data generation module is used to perform correlation analysis on the historical business service results corresponding to the target account to obtain the historical business trajectory; The business anomaly prediction module is used to predict the future business usage of the target account based on the historical business trajectory corresponding to the target account. The prediction result output module is used to predict the future network quality of the target account based on the future service usage within the time window.

10. A network quality prediction device, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the network quality prediction method as described in any one of claims 1-8.

11. A computer storage medium, characterized in that, The computer storage medium stores computer program instructions, which, when executed by a processor, implement the network quality prediction method as described in any one of claims 1-8.

12. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the network quality prediction method as described in any one of claims 1-8.

Images