Adaptive network traffic prediction system based on user feature and method thereof
Patent Information
- Application Number
- TW114106141
- Authority / Receiving Office
- TW · TW
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2025-02-19
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2045-02-18
AI Technical Summary
Traditional IT maintenance and monitoring services lack the ability to provide accurate future network traffic predictions based on different network environments, leading to inefficient resource allocation and poor customer experience.
An adaptive network traffic prediction system that analyzes network route usage behavior and user characteristics, utilizing a resource management module, data collection and feature generation module, network traffic prediction module, and comprehensive performance evaluation module to select the best prediction algorithm for each line, ensuring accurate and intelligent management of network traffic.
The system achieves accurate prediction of future network traffic, assisting in resource allocation planning and reducing anomalies, thereby ensuring network service stability and efficiency.
Smart Images

Figure TWG2TB001920462_001 
Figure TWG2TB001920462_002 
Figure TWG2TB001920462_003
Abstract
Description
Technical Field
[0001] This invention relates to an adaptive network traffic prediction system and method, and more particularly to an adaptive network traffic prediction system and method based on user characteristics. Prior Technology
[0002] Traditional IT maintenance and monitoring services are of poor quality and cannot provide future network traffic predictions based on different network environments or network lines, such as customer type, usage frequency, time period preferences, and data transmission volume. During prediction calculations, the low accuracy of predictions leads to incorrect resource allocation, resulting in not only a waste of computing power and energy, but also low accuracy and reliability of traditional network traffic prediction methods, resulting in a poor customer experience. Summary of the Invention
[0003] This invention provides an adaptive network traffic prediction system and method based on user characteristics. It can not only predict future network traffic by analyzing network route usage behavior, but also predict network traffic in different scenarios, thereby ensuring the stability of network services and management efficiency.
[0004] This invention discloses an adaptive network traffic prediction system based on user characteristics, comprising a resource management module, a data collection and feature generation module, a network traffic prediction module, and a comprehensive performance evaluation module. The resource management module stores and manages basic configuration information for each line in the telecommunications network. The data collection and feature generation module is communicatively connected to the resource management module and collects and stores network traffic data for each line in real time based on the basic configuration information, generating feature tags for each line. The network traffic prediction module is also communicatively connected to the data collection and feature generation module and includes a prediction algorithm library. This library stores multiple prediction algorithm groups corresponding to the feature tags of each line. Each prediction algorithm group includes at least two prediction algorithms. The network traffic prediction module uses the prediction algorithm groups corresponding to the feature tags of each line during the model training phase and trains the model based on historical network traffic data for each line to obtain prediction results corresponding to at least two prediction algorithms. The performance evaluation module communicates with the network traffic prediction module to obtain scores corresponding to at least two prediction algorithms based on the network traffic data of each line and the prediction results. Based on the scores, the best prediction algorithm is selected from the at least two prediction algorithms. During the model inference stage, the network traffic prediction module uses the best prediction algorithm and makes predictions based on the network traffic data of each line to obtain the best prediction results.
[0005] The present invention provides an adaptive network traffic prediction method based on user characteristics, comprising: storing and managing basic configuration information of each line in a telecommunications network; collecting and storing network traffic data of each line in real time based on the basic configuration information of each line, and generating feature labels for each line respectively; training the model in the model training phase using a set of prediction algorithms corresponding to the feature labels of each line and based on the historical network traffic data of each line to obtain prediction results corresponding to at least two prediction algorithms respectively; obtaining scores corresponding to at least two prediction algorithms based on the network traffic data of each line and the prediction results, selecting the best prediction algorithm from the at least two prediction algorithms based on the scores, and using the best prediction algorithm in the model inference phase and making predictions based on the network traffic data of each line to obtain the best prediction result.
[0006] Based on the above, this invention provides an adaptive network traffic prediction system and method based on user characteristics. This system not only predicts future network traffic by analyzing network usage behavior, assisting users in planning network resource allocation and reducing potential network traffic anomalies, but also provides predictive services and maintenance suggestions for various types of network services by predicting network traffic in different scenarios. This achieves accurate prediction and intelligent management of future network traffic, thereby ensuring the stability and efficiency of network services and management.
[0007] To make the above features and advantages of the present invention more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings for detailed explanation. Simple Explanation of the Diagram
[0008] Figure 1 is a schematic diagram of an adaptive network traffic prediction system according to an embodiment of the present invention. Figure 2 is a schematic diagram of a prediction algorithm library according to an embodiment of the present invention. Figure 3A is a schematic diagram of raw time series data according to an embodiment of the present invention. Figure 3B is a schematic diagram of periodic analysis according to an embodiment of the present invention. Figure 4 is a schematic diagram of the model training phase according to an embodiment of the present invention. Figure 5 is a schematic flowchart of the model inference stage according to an embodiment of the present invention. Figure 6 is a schematic diagram of the process of retraining the network prediction model when the set conditions are triggered according to an embodiment of the present invention. Figure 7 is a schematic diagram of VPN network traffic prediction according to an embodiment of the present invention. Implementation
[0009] Some embodiments of the present invention will now be described in detail with reference to the accompanying drawings. Component symbols used in the following description are considered to be the same or similar components when they appear in different drawings. These embodiments are only a part of the present invention and do not disclose all possible implementations of the invention.
[0010] Figure 1 is a schematic diagram of an adaptive network traffic prediction system according to an embodiment of the present invention.
[0011] The adaptive network traffic prediction system 10 of this invention is applicable to various network traffic scenarios, including but not limited to VPN network traffic, broadband network traffic, and data leased line network traffic. By predicting network traffic in different scenarios, the adaptive network traffic prediction system 10 can provide predictive services and maintenance suggestions for various types of network services, achieving accurate prediction and intelligent management of future network traffic, thereby ensuring the stability and efficiency of network services.
[0012] Referring to Figure 1, the adaptive network traffic prediction system 10 may include a resource management module 101, a data collection and feature generation module 102, a network traffic prediction module 103, a comprehensive performance evaluation module 104, and a service application module 105.
[0013] Among them, the resource management module 101 can store and manage the basic configuration information of each line in the telecommunications network.
[0014] In one embodiment, the resource management module 101 can include lines requiring network traffic prediction within the management scope of the adaptive network traffic prediction system 10 and manage them. The resource management module 101 can access and store basic configuration information of the lines, including line type, speed, connection method, etc., as well as line configuration files, such as preset prediction training rules and priority levels. This invention is not limited thereto. The resource management module 101 can manage the managed lines and ensure that all managed lines can be effectively monitored and analyzed, thereby ensuring subsequent data collection, feature extraction, and prediction model training.
[0015] In another embodiment, the resource management module 101 can obtain adjustment data set by the system administrator or user through the interface provided by the service application module 105.
[0016] The data collection and feature generation module 102 is connected to the resource management module 101. It can collect and store the network traffic data of each line in real time based on the basic configuration information of each line, and generate feature tags for each line respectively.
[0017] In one embodiment, the data collection and feature generation module 102 may include a data collection and processing unit and a feature generation unit.
[0018] The data collection and processing unit can collect and store network traffic data of the managed lines in real time, arrange the network traffic data in chronological order to obtain time series data, and perform data cleaning and standardization on the time series data.
[0019] In one embodiment, the data collection and processing unit can collect network traffic data of the managed lines from the monitoring interface in real time, generate complete time-series data in chronological order, and store it in a database.
[0020] The data collection and processing unit will continuously collect network traffic data. For existing time-series data, the unit can use techniques such as median substitution to handle missing values and remove redundant values, ensuring that the number of data points collected daily matches the expected number. For example, if data is collected every 5 minutes, 288 data points can be collected daily. After the number of data points is correct, the time-series data that has undergone missing value handling is standardized to convert the data values to a uniform scale range, ensuring consistency between the upper and lower bounds of the data, facilitating subsequent analysis, calculation, and comparison.
[0021] The following explanation, in conjunction with Figures 3A and 3B, illustrates how the data collection and processing unit performs periodic analysis on the collected time series data.
[0022] Figure 3A is a schematic diagram of raw time series data according to an embodiment of the present invention. Figure 3B is a schematic diagram of periodicity analysis according to an embodiment of the present invention.
[0023] Please refer to Figures 3A and 3B. Screen 301 is a schematic diagram of the raw time series data collected by the data collection and processing unit according to time. Screen 302 is a schematic diagram of the periodic analysis after data standardization. The time series data shows periodicity from index 288 to index 2000. In other words, as can be seen from Figure 3, the period of the network traffic data of the corresponding line is every 1 day and every 7 days, that is, daily and weekly.
[0024] The feature generation unit is connected to the data collection and processing unit, which can perform periodic analysis on the processed time series data and generate feature labels for each line, with each line feature label reflecting the usage behavior pattern of each line.
[0025] In one embodiment, the feature generation unit can perform periodic analysis of network usage behavior characteristics based on the processed time-series data to identify regular changes in network traffic. The feature generation unit can use autocorrelation analysis and time-domain-frequency conversion methods for analysis. Autocorrelation analysis is used to find the correlation between time points in the time-series data, while time-domain-frequency conversion methods are used to reveal periodic characteristics in the time-series data. Based on the extracted periodic characteristics or by adding de-identified customer classification features, the feature generation unit can generate specific line feature labels for each individual managed line. These line feature labels reflect the usage behavior patterns of the line, and the adaptive network traffic prediction system 10 records them and uses them in subsequent network prediction model selection to optimize the model training process.
[0026] The network traffic prediction module 103 is communicatively connected to the data collection and feature generation module 102, and may include a prediction algorithm library 20. The prediction algorithm library 20 may store multiple prediction algorithm groups corresponding to the feature tags of each line, and each prediction algorithm group may include at least two prediction algorithms.
[0027] In one embodiment, the prediction algorithm library 20 may store various time series prediction algorithms, including moving average (MA), weighted moving average (WMA), autoregressive integrated moving average model (ARIMA), linear regression, support vector regression (SVR), extreme gradient boosting regression (XGBoost), Prophet, random forest regression, or long short-term memory (LSTM), gated recurrent unit (GRU), and convolutional neural network (CNN). This invention is not limited thereto.
[0028] Figure 2 is a schematic diagram of a prediction algorithm library according to an embodiment of the present invention.
[0029] Please refer to Figure 2. The adaptive network traffic prediction system 10 pre-groups the prediction algorithms into multiple prediction algorithm groups corresponding to the feature labels of each line based on factors such as resource consumption (prediction difficulty), complexity, and architecture. This facilitates the flexible selection of suitable prediction algorithm groups for model training and model inference under different circumstances.
[0030] In one embodiment, the prediction algorithms described above can be grouped into prediction algorithm group A, prediction algorithm group B, and prediction algorithm group C based on resource consumption (prediction difficulty). Prediction algorithm group A may, for example, include the LSTM algorithm, prediction algorithm group B may, for example, include the MA algorithm, ARIMA algorithm, or random forest regression algorithm, and prediction algorithm group C may, for example, include the ARIMA algorithm or random forest regression algorithm. This invention is not limited thereto.
[0031] For route feature labels (label type 1), which reflect the route's usage pattern as full-load mode, prediction algorithm group B corresponding to each route feature label can be selected during model training. For route feature labels (label type 2), which reflect the route's usage pattern as a fixed-cycle mode, such as for customers in the banking industry, prediction algorithm group B or C corresponding to each route feature label can be selected during model training. For route feature labels (label type 3), which reflect the route's usage pattern as a daily cycle, such as for customers in the e-commerce industry, prediction algorithm group C corresponding to each route feature label can be selected during model training. For route feature labels (label type 4), which reflect the route's usage pattern as an aperiodic mode, such as for customers in the medical or aviation industries, prediction algorithm group A, B, or C corresponding to each route feature label can be selected during model training.
[0032] The network traffic prediction module 103 can select to use all the algorithms in the prediction algorithm group corresponding to the feature labels of each line during the model training phase, and perform model training based on the historical network traffic data of each line to obtain prediction results corresponding to at least two prediction algorithms respectively.
[0033] The comprehensive performance evaluation module 104 is connected to the network traffic prediction module 103. It can obtain various evaluation indicators based on the network traffic data and prediction results of each line. Based on each evaluation indicator and its corresponding weight, it can obtain scores corresponding to at least two prediction algorithms. The scores are sorted to select the best prediction algorithm from at least two prediction algorithms as the trained network prediction model for subsequent model inference.
[0034] In one embodiment, the evaluation indicators may include, but are not limited to, the coefficient of determination (R²), root-mean-square error (RMSE), and mean absolute error (MAE). The comprehensive performance evaluation module 104 weights each evaluation indicator according to its importance, and the weights can be set by the system administrator based on the characteristics of different network scenarios. Since different network environments emphasize different evaluation indicators, and the reason for using multiple evaluation indicators is to consider the different focuses of each indicator, such as the model's explanatory power, the relative proportion of prediction errors, and the magnitude of extreme value impacts. After weighting each evaluation indicator with specific weights, a more reasonable comprehensive score is calculated.
[0035] The network traffic prediction module 103 will select one or more of the best prediction algorithms (i.e. network prediction models) from the trained prediction algorithms based on the scores provided by the comprehensive performance evaluation module 104, and perform model inference. In the model inference stage, the best prediction algorithm is used and prediction is made based on the network traffic data of each line to obtain the best prediction results.
[0036] Service application module 105 serves as the interface between the adaptive network traffic prediction system 10 and the user, presenting network traffic monitoring and prediction results to the user or system administrator in an intuitive manner. Service application module 105 is communicatively connected to the resource management module and the performance evaluation module, including a user setting interface unit and a traffic display interface unit.
[0037] The user setting interface unit provides an interface for users to set the trigger conditions for model training or the setting conditions and triggering methods. It may include the following settings: prediction function on / off and parameter adjustment, warning function settings, notification methods, retraining mechanism trigger settings, etc.
[0038] Forecast function on / off and parameter adjustment: Users can enable or disable the forecast function as needed, and adjust the forecast range, time period, etc.
[0039] Alert Function Settings: Users can customize traffic warning notification conditions, and issue an alert notification when the predicted traffic reaches specific conditions.
[0040] Notification method: Users can customize how to receive notifications (such as email, alert sound, etc.) when a certain notification condition is met.
[0041] Retraining Mechanism Trigger Settings: Users can configure the restart conditions and trigger methods for the trained model to ensure that the model can continuously adapt to the latest data changes and periodic characteristics. Users can set trigger conditions including but not limited to the following: fixed-time trigger; variable-time trigger; model performance deviation trigger; system preset trigger.
[0042] The traffic display interface unit communicates with the user-configured interface unit to provide a visual presentation of the interface and display the current real-time traffic data and predicted traffic data (i.e., the best prediction result). Users can adjust the visual presentation of the interface according to their personal preferences, including but not limited to the following adjustments: chart type; chart lines; display ratio of actual traffic data and predicted traffic data; and colors. This helps users to understand the network status more intuitively, thereby supporting decision-making, optimizing resource allocation, and improving network management efficiency.
[0043] In one embodiment, users can view real-time traffic monitoring data and predicted future traffic data through the traffic display interface unit, and refer to the prediction results of the adaptive network traffic prediction system 10 as a basis for network setup and management decisions.
[0044] In one embodiment of the present invention, the resource management module 101, the data collection and feature generation module 102, the network traffic prediction module 103, the comprehensive performance evaluation module 104, the service application module 105, the data collection and processing unit, the feature generation unit, the user setting interface unit, and the traffic display interface unit are implemented using a central processing unit (CPU) or other programmable general-purpose or special-purpose microprocessors. In another embodiment of the present invention, the resource management module 101, the data collection and feature generation module 102, the network traffic prediction module 103, the comprehensive performance evaluation module 104, the service application module 105, the data collection and processing unit, the feature generation unit, the user setting interface unit, and the traffic display interface unit may also be implemented using the same processor and loaded with different modules; the present invention is not limited thereto.
[0045] The resource management module 101, data collection and feature generation module 102, network traffic prediction module 103, comprehensive performance evaluation module 104, service application module 105, data collection and processing unit, feature generation unit, user setting interface unit, and traffic display interface unit can be implemented through software, firmware, hardware circuits, or any combination thereof. This disclosure does not limit the implementation method of the resource management module 101, data collection and feature generation module 102, network traffic prediction module 103, comprehensive performance evaluation module 104, service application module 105, data collection and processing unit, feature generation unit, user setting interface unit, and traffic display interface unit.
[0046] The adaptive network traffic prediction method described in Figures 4 to 6 of this invention will be explained below with reference to the various devices, components, and modules shown in Figure 1. The processes in Figures 4 to 6 may be adjusted according to the implementation situation and are not limited thereto.
[0047] Figure 4 is a schematic diagram of the model training phase according to an embodiment of the present invention.
[0048] Please refer to Figure 4. During the model training phase, in step S401, the resource management module 101 manages a line in the telecommunications network whose traffic is to be predicted and stores the basic configuration information of the line.
[0049] In step S402, the data collection and processing unit collects and stores the network traffic data of the line in real time based on the basic configuration information of the line.
[0050] In step S403, the data collection and processing unit arranges the network traffic data of the line in chronological order to obtain time series data.
[0051] In step S404, the data collection and processing unit performs data cleaning on the time series data. For example, the data collection and processing unit can remove redundant values and fill in missing values on the time series data.
[0052] In step S405, the data collection and processing unit performs data standardization processing on the time series data.
[0053] In step S406, the feature generation unit performs periodic analysis on the processed time series data.
[0054] In step S407, the feature generation unit generates a line feature label for the corresponding line, wherein the line feature label reflects the usage behavior pattern of the line.
[0055] In step S408, the network traffic prediction module 103 selects and uses the prediction algorithm group corresponding to the line feature label.
[0056] In step S409, the network traffic prediction module 103 uses all the prediction algorithms in the prediction algorithm group and trains the model based on the historical network traffic data of the line to obtain the prediction results corresponding to each prediction algorithm.
[0057] In step S410, the comprehensive performance evaluation module 104 obtains the score corresponding to each prediction algorithm based on the network traffic data of the line and the prediction results of each prediction algorithm.
[0058] The adaptive network traffic prediction system 10 can select the best prediction algorithm (i.e., network traffic prediction model) based on the obtained scores during the model training phase and execute the model inference phase shown in Figure 5.
[0059] Figure 5 is a schematic flowchart of the model inference stage according to an embodiment of the present invention.
[0060] Please refer to Figure 5. During the model inference stage, in step S501, the data collection and processing unit collects and stores the network traffic data of the line in real time based on the basic configuration information of the line.
[0061] In step S502, the data collection and processing unit arranges the network traffic data of the line in chronological order to obtain time series data.
[0062] In step S503, the data collection and processing unit cleans the time series data. For example, the data collection and processing unit can remove redundant values and fill in missing values in the time series data.
[0063] In step S504, the data collection and processing unit performs data standardization processing on the time series data.
[0064] In step S505, the network traffic prediction module 103 uses the best prediction algorithm and makes a prediction based on the network traffic data of the line, that is, performs model inference, and the user can view the prediction results through the traffic display interface unit of the service application module 105.
[0065] In step S506, the comprehensive performance evaluation module 104 obtains the best prediction result, i.e., it performs the evaluation.
[0066] Figure 6 is a schematic diagram of the process of retraining the network prediction model when the set conditions are triggered according to an embodiment of the present invention.
[0067] Please refer to Figure 6. In step S601, when the prediction function of the line is enabled, the user setting interface unit of the service application module 105 adopts the preset setting conditions of the training network prediction model.
[0068] In step S602, the user adjusts the settings for retraining the network prediction model and specifies custom rules through the user settings interface unit.
[0069] In step S603, the set conditions are triggered, which triggers the network traffic prediction module 103 to perform retraining.
[0070] In step S604, the network traffic prediction module 103 determines whether the conditions for regenerating line feature labels are met.
[0071] If the conditions for regenerating line feature labels are met, in step S605, the feature generation unit performs periodic analysis on the processed time series data.
[0072] In step S606, the feature generation unit generates a new line feature label for the corresponding line.
[0073] In step S607, the network traffic prediction module 103 selects and uses a new prediction algorithm group corresponding to the new line feature label.
[0074] In step S608, the network traffic prediction module 103 uses all the prediction algorithms in the new prediction algorithm group and trains the model based on the historical network traffic data of the line to obtain the prediction results corresponding to each prediction algorithm.
[0075] In step S609, the comprehensive performance evaluation module 104 obtains the scores corresponding to each prediction algorithm and enters the subsequent new model inference stage.
[0076] If the conditions for regenerating the line feature label are not met, proceed to step S610.
[0077] In step S610, the network traffic prediction module 103 selects and uses the prediction algorithm group corresponding to the line feature label.
[0078] In step S611, the network traffic prediction module 103 uses all the prediction algorithms in the prediction algorithm group and trains the model based on the historical network traffic data of the line to obtain the prediction results corresponding to each prediction algorithm.
[0079] In step S612, the comprehensive performance evaluation module 104 obtains the scores corresponding to each prediction algorithm and enters the subsequent model inference stage.
[0080] The adaptive network traffic prediction system 10 has a mechanism that allows it to automatically re-enter the model training phase at fixed times (such as monthly) or variable times (such as sporting events) set by the user or generated by the system during the model inference phase. This ensures that the network prediction model can continuously adapt to the latest traffic changes and periodic characteristics, avoid deviations over time, and maintain the accuracy and reliability of the network traffic prediction method.
[0081] The following section, in conjunction with Figure 7, explains in detail how to perform network traffic prediction in a VPN network scenario when a single VPN network line leased by a customer is used as the target for traffic prediction.
[0082] Figure 7 is a schematic diagram of VPN network traffic prediction according to an embodiment of the present invention.
[0083] Please refer to Figure 7. System administrators or users can configure and enable the network traffic prediction function of the VPN line through the service application module 105.
[0084] Resource management module 101 incorporates the VPN line into the network traffic prediction function and loads the basic configuration information of the VPN line.
[0085] The data collection and feature generation module 102's data collection and processing unit collects network traffic data in real time based on the VPN line. According to the monitoring requirements of the VPN line type, the system administrator configures the service application module 105 to collect data in real time at each interval. Network traffic data is captured every minute and stored in a database to generate time-series data, such as... ,in Representative at the The traffic value recorded during each capture, where n is a positive integer. In this embodiment, it can be set to record once every 5 minutes, and the adaptive network traffic prediction system 10 will collect 288 network traffic data records per day.
[0086] The data collection and feature generation module 102's data collection and processing unit checks the completeness of the aforementioned time series data and processes redundant and missing values. In this embodiment, it ensures that 288 traffic data records are collected daily. If redundant or missing values appear in the collected data due to network equipment problems, the median substitution method is used to process the data, remove redundant values and fill in missing values, and maintain the continuity of the time series.
[0087] Subsequently, the data collection and processing unit of the data collection and feature generation module 102 standardizes the time series data that has been ensured to be complete. To ensure that the algorithm can run effectively in subsequent numerical calculations and to avoid bias caused by differences in traffic data scale, the adaptive network traffic prediction system 10 transforms all network traffic data to a value between 0 and 1 using the following formula 1. Formula 1 is as follows:
[0088] in, In order to be in Raw traffic data at specific times and These are the maximum and minimum values in the time series of network traffic data for this segment. This refers to standardized network traffic data.
[0089] The feature generation unit of the data collection and feature generation module 102 performs time period analysis. In this case, autocorrelation analysis is used for period analysis. The adaptive network traffic prediction system 10 pairs time series data. Perform periodic analysis to understand the periodic state in the data.
[0090] Referring to Figure 3, the adaptive network traffic prediction system 10 analyzes the correlation between different time points in the time series data, and identifies periodic patterns by calculating the degree of correlation between the time series data. Based on Formula 2, for given time series data... The periodicity characteristics of this material are calculated using Formula 2 as follows:
[0091] in, Representing different time lags, It is the lag range of the analysis. Time series data Lag The correlation function.
[0092] Based on the above periodic feature analysis, the adaptive network traffic prediction system 10 generates time periodic features to provide a basis for the subsequent generation of line feature labels. In this case, multiples of 288 time points are used as the main periodic sampling source, namely the 288th, 576th, ..., 2016th, ... time points.
[0093] The feature generation unit of the data collection and feature generation module 102 assigns and stores line tags. By comparing the periodic characteristics of the line traffic, the adaptive network traffic prediction system 10 generates line feature tags for the VPN line based on these periodic characteristics and other manually configured data characteristics of the customer applying for the VPN line, and records the tag information.
[0094] During the model training phase, the network traffic prediction module 103 uses the historical traffic time series data of the VPN line and divides it into training set, validation set and test set. At the same time, it uses all the prediction algorithms in the prediction algorithm group corresponding to the line feature label to carry out training and stores the results.
[0095] The comprehensive performance evaluation module scores all trained prediction algorithms to assess the effectiveness of the traffic prediction model. The evaluation metrics are calculated by comparing the prediction results of each algorithm with actual network traffic data. In this embodiment, the following approach is used: , , , , The scores are calculated by weighting each evaluation indicator, as shown in Formula 3 below:
[0096] in, This indicates the weight of the evaluation indicators, and , After the scoring is completed, the system obtains the scores of each time series prediction algorithm used to predict the VPN circuit, sorts and records the scores, and obtains the best prediction algorithm (i.e. the best traffic prediction model) based on the scores.
[0097] During the model inference process of the network traffic prediction module 103, the adaptive network traffic prediction system 10 will no longer use all algorithms, but will only use a few (one in this example) best traffic prediction models with the best scores to continuously make predictions and store the prediction results.
[0098] Users can view the prediction results in the service application module 105 (as shown in Figure 7), thereby supporting decision-making, optimizing resource allocation, and improving network management efficiency.
[0099] Based on the above, this invention provides an adaptive network traffic prediction system and method based on user characteristics. This system not only predicts future network traffic by analyzing network usage behavior, assisting users in planning network resource allocation and reducing potential network traffic anomalies, but also provides predictive services and maintenance suggestions for various types of network services by predicting network traffic in different scenarios. This achieves accurate prediction and intelligent management of future network traffic, thereby ensuring the stability and efficiency of network services and management.
[0100] Although the present invention has been disclosed above by way of embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.
[0101] 10: Adaptive Network Traffic Prediction System 101: Resource Management Module 102: Data Collection and Feature Generation Module 103: Network Traffic Prediction Module 104: Comprehensive Performance Evaluation Module 105: Service Application Module 301, 302: Screen S401, S402, S403, S404, S405, S406, S407, S408, S409, S410, S501, S502, S503, S504, S505, S506, S601, S602, S603, S604, S605, S606, S607, S608, S609: Steps
Claims
1. An adaptive network traffic prediction system based on user characteristics, comprising a resource management module, a data collection and feature generation module, a network traffic prediction module, and a comprehensive performance evaluation module, wherein, The resource management module stores and manages the basic configuration information of each line in the telecommunications network. The data collection and feature generation module communicates with the resource management module to collect and store network traffic data for each line in real time based on the basic configuration information, and generates feature tags for each line. The network traffic prediction module communicates with the data collection and feature generation module and includes a prediction algorithm library. This library stores multiple prediction algorithms, which are pre-grouped into groups corresponding to the feature tags of each line based on resource consumption and complexity. Each line feature tag reflects the usage behavior pattern of the line, which can be a full-load mode, a fixed usage cycle mode, a daily cycle mode, or a non-cycle mode. Each prediction algorithm group includes at least two prediction algorithms. The network traffic prediction module is used during the model training phase. A set of prediction algorithms corresponding to the feature labels of each line is trained on the model based on the historical network traffic data of each line to obtain prediction results corresponding to the at least two prediction algorithms respectively; and the comprehensive performance scoring module is connected to the network traffic prediction module to obtain scores corresponding to the at least two prediction algorithms based on the network traffic data of each line and the prediction results, select the best prediction algorithm from the at least two prediction algorithms based on the score, and the network traffic prediction module uses the best prediction algorithm in the model inference stage and makes predictions based on the network traffic data of each line to obtain the best prediction results.
2. The adaptive network traffic prediction system as described in claim 1, wherein the data collection and feature generation module comprises: The data collection and processing unit is used to collect and store network traffic data of each line in real time, arrange the network traffic data in chronological order to obtain time series data, and perform data cleaning and standardization on the time series data; and the feature generation unit is connected to the data collection and processing unit to perform periodic analysis on the processed time series data, and generate feature tags for each line respectively, wherein each line feature tag reflects the usage behavior pattern of each line.
3. The adaptive network traffic prediction system as described in claim 1, wherein the prediction algorithm is a moving average (MA), a weighted moving average (WMA), an autoregressive integrated moving average model (ARIMA), a linear regression, a support vector regression (SVR), an extreme gradient boosting regression (XGBoost), Prophet, a random forest regression algorithm, or a long short-term memory (LSTM), a gated recurrent unit (GRU), or a convolutional neural network (CNN).
4. The adaptive network traffic prediction system as described in claim 1, wherein the comprehensive performance evaluation module is further configured to obtain each evaluation index based on the network traffic data of each line and the prediction result, obtain the score corresponding to the at least two prediction algorithms based on each evaluation index and the corresponding weight, and sort the scores to select the best prediction algorithm from the at least two prediction algorithms.
5. The adaptive network traffic prediction system as described in claim 1, wherein the adaptive network traffic prediction system further includes a service application module that is communicatively connected to the resource management module and the performance evaluation module, respectively, and the service application module includes: The user setting interface unit provides an interface for users to set the triggering conditions and triggering methods for model training; And a flow display interface unit, which communicates with the user setting interface unit to provide a visual presentation of the interface and display the best prediction result.
6. An adaptive network traffic prediction method based on user characteristics, comprising: Stores and manages basic configuration information for each line in the telecommunications network; Based on the basic configuration information of each line, network traffic data of each line is collected and stored in real time, and feature tags are generated for each line respectively. The system stores multiple prediction algorithms in a prediction algorithm library. Based on resource consumption and complexity, these multiple prediction algorithms are pre-grouped into multiple prediction algorithm groups corresponding to the feature labels of each line. Each prediction algorithm group includes at least two prediction algorithms. The feature labels of each line reflect the usage behavior pattern of each line, which can be a full-load mode, a fixed usage cycle mode, a daily cycle mode, or an acyclic mode. During the model training phase, the prediction algorithm groups corresponding to the feature labels of each line are used to train the model based on the historical network traffic data of each line to obtain prediction results corresponding to the at least two prediction algorithms. Based on the network traffic data of each line and the prediction results, scores are obtained for the at least two prediction algorithms. Based on the scores, the best prediction algorithm is selected from the at least two prediction algorithms. During the model inference phase, the best prediction algorithm is used to make predictions based on the network traffic data of each line to obtain the best prediction result.
7. The adaptive network traffic prediction method as described in claim 6, wherein the step of collecting and storing network traffic data of each line in real time based on the basic configuration information of each line, and generating feature tags for each line respectively, further includes: The system collects and stores network traffic data for each line in real time, arranges the network traffic data in chronological order to obtain time series data, cleans and standardizes the time series data, performs periodic analysis on the processed time series data, and generates feature tags for each line, where each feature tag reflects the usage behavior pattern of each line.
8. The adaptive network traffic prediction method as described in claim 6, wherein the prediction algorithm is a moving average (MA), a weighted moving average (WMA), an autoregressive integrated moving average model (ARIMA), a linear regression, a support vector regression (SVR), an extreme gradient boosting regression (XGBoost), Prophet, a random forest regression algorithm, or a long short-term memory (LSTM), a gated recurrent unit (GRU), or a convolutional neural network (CNN).
9. The adaptive network traffic prediction method as described in claim 6, wherein the step of obtaining scores corresponding to the at least two prediction algorithms based on the network traffic data of each line and the prediction result, and selecting the best prediction algorithm from the at least two prediction algorithms based on the scores, further includes: Based on the network traffic data of each line and the prediction results, each evaluation index is obtained. Based on each evaluation index and its corresponding weight, the score corresponding to each of the at least two prediction algorithms is obtained. The scores are then sorted to select the best prediction algorithm from the at least two prediction algorithms.
10. The adaptive network traffic prediction method as described in claim 6, wherein the method further comprises: Set the trigger conditions for training the model; In response to the triggering condition being triggered, the feature tags of each line are loaded, or new network traffic data of each line is collected and stored in real time according to the triggering condition, and new feature tags corresponding to each line are regenerated; during the model training phase, the prediction algorithm group corresponding to each new line feature tag is used and the model is trained according to the historical network traffic data of each line to obtain new prediction results corresponding to the at least two prediction algorithms; and new scores corresponding to the at least two prediction algorithms are obtained according to the new network traffic data of each line and the new prediction results, and a new best prediction algorithm is selected from the at least two prediction algorithms according to the new scores. During the model inference phase, the new best prediction algorithm is used and prediction is performed according to the new network traffic data of each line to obtain a new best prediction result.
Citation Information
Patent Citations
Network flow prediction method based on attention multi-component space-time cross-domain neural network model
CN112532439A
Network traffic prediction method and system
CN117354172A
Network flow analysis method and system based on machine learning
CN118764278A
System and methods for automatically validating evidence of traffic violations using automatically detected context features
US11776276B1
