Data processing method and network management device
By collecting 6-dimensional traffic feature vectors and lightweight AI models, combined with hierarchical filtering and scenario-based scheduling, the problem of traffic scheduling in complex network environments was solved, achieving high precision, stability and adaptability, and improving the scheduling quality of network management devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-09
- Publication Date
- 2026-04-10
AI Technical Summary
Existing technologies struggle to achieve high-precision and high-stability traffic scheduling in complex network environments, exhibiting issues such as insufficient noise identification accuracy, poor scheduling strategy stability, inadequate scenario adaptability, and limitations in decision-making under distributed scenarios.
By collecting 6-dimensional traffic feature vectors, combining a lightweight CNN-LSTM hybrid model and an improved ARIMA model, hierarchical filtering strategies and scenario-based scheduling are implemented to achieve accurate differentiation between noise and real traffic. Furthermore, the scheduling strategy is optimized through edge-center collaborative communication.
It improves the accuracy and stability of traffic scheduling, reduces the probability of triggering invalid scheduling, adapts to different scenario requirements, and enhances the global optimization capability of distributed networks.
Smart Images
Figure CN121842279A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer network technology, and in particular to a data processing method and a network management device. Background Technology
[0002] With the rapid development of computer network technology, especially the widespread adoption of distributed networks and the Internet of Things (IoT), the volume of traffic carried by network links continues to grow, and the types of traffic are becoming increasingly complex. As the core equipment ensuring stable network operation, the rationality of traffic scheduling by network management devices directly determines the network's transmission efficiency, link stability, and service quality. Real-time traffic data is the core basis for network management devices to formulate scheduling strategies. How to accurately process interference information in real-time traffic data and avoid link jitter caused by frequent adjustments to scheduling strategies has become a key technical problem that urgently needs to be solved in this field.
[0003] In existing technologies, the data processing during traffic scheduling mainly adopts the following two types of schemes: First, a traffic noise filtering scheme based on a fixed threshold, which judges and filters traffic fluctuations exceeding the threshold as interference by setting a preset traffic change threshold, and then triggers scheduling strategy adjustments based on the filtered traffic data; Second, a traffic preprocessing scheme based on a simple smoothing algorithm, which uses basic algorithms such as moving average to smooth real-time traffic data, reduces the impact of instantaneous fluctuations on scheduling decisions, and reduces the frequency of strategy adjustments by setting a single scheduling adjustment interval constraint.
[0004] However, the aforementioned existing technologies have many shortcomings in practical applications and are difficult to meet the high-precision and high-stability scheduling requirements in complex network environments:
[0005] 1. Insufficient accuracy in noise identification: Existing fixed threshold filtering schemes determine noise based solely on the magnitude or duration of a single traffic change, which cannot cope with diverse traffic fluctuations in complex network environments (such as small packet interactions between heterogeneous terminals, sudden protocol handshake traffic, etc.), and are prone to problems such as "misjudging real traffic changes" or "missing instantaneous noise"; although simple smoothing algorithms can reduce fluctuations, they lose the true trend characteristics of traffic, resulting in a mismatch between scheduling strategies and actual traffic demands.
[0006] 2. Poor stability of scheduling strategies: Existing technologies all adopt a "passive response" scheduling triggering mechanism, which only judges whether to adjust the strategy based on real-time filtered traffic data, lacking the ability to predict short-term traffic trends. When traffic experiences small, high-frequency fluctuations, even after smoothing, scheduling strategy adjustments will still be frequently triggered, causing network link jitter. This is especially problematic in scenarios with high requirements for link stability, such as real-time audio and video, and industrial control, which seriously affects service quality.
[0007] 3. Insufficient scenario adaptability: Existing technologies use fixed scheduling constraint parameters (such as adjustment intervals and change thresholds), which cannot adapt to the differentiated needs of different application scenarios. For example, low-latency sensitive scenarios (such as real-time communication) require rapid scheduling response, while high-stability sensitive scenarios (such as financial transactions) require strict control of scheduling frequency. Fixed parameters will lead to insufficient real-time performance or lack of stability in some scenarios.
[0008] 4. Limitations of decision-making in distributed scenarios: In distributed networks, existing technologies mostly adopt a single edge node independent decision-making approach for traffic data processing and scheduling, without establishing a collaborative mechanism between edge nodes and central management nodes. When multiple edge nodes experience link jitter, independent parameter adjustments by a single node can easily lead to conflicts in global scheduling strategies or cause a decrease in overall network transmission efficiency due to over-filtering; while when only a single node experiences jitter, there is a lack of targeted local parameter optimization methods.
[0009] Therefore, a data processing method and a network management device are proposed. Summary of the Invention
[0010] This application aims to at least partially solve one of the technical problems in the aforementioned technologies.
[0011] To achieve the above objectives, the first aspect of this application proposes a data processing method, comprising the following steps:
[0012] 1) Collect real-time traffic data of the target link and extract a 6-dimensional traffic feature vector from the real-time traffic data. The 6-dimensional traffic feature vector specifically includes the duration of change, the slope of change, the coordination of associated links, the traffic pulse width, the data packet size distribution and the port protocol type.
[0013] 2) Input the 6-dimensional traffic feature vector into a pre-set lightweight CNN-LSTM hybrid model, and output the noise probability and the true change probability of the corresponding real-time traffic data through the hybrid model;
[0014] 3) Implement a hierarchical filtering strategy based on noise probability to perform weighted smoothing on real-time traffic data and obtain smoothed traffic values;
[0015] 4) Based on the smoothed flow value, short-term flow trend prediction is performed using an improved ARIMA model, and the predicted flow range is output. The filtered measured flow value is compared with the predicted flow range, and the flow scheduling adjustment is determined based on the comparison result.
[0016] 5) Identify the application scenario carried by the current link through the protocol parsing module, dynamically configure scheduling constraint parameters based on the type of application scenario, and execute traffic scheduling strategies that match the application scenario.
[0017] By comprehensively characterizing traffic fluctuations through a 6-dimensional traffic feature vector, and combining a lightweight CNN-LSTM hybrid model to accurately distinguish between noise and real traffic changes, and by overlaying a full-process collaboration of hierarchical filtering, time-series prediction and scenario-based scheduling, the problem of low accuracy and poor adaptability of traditional single processing methods is solved. By integrating prediction comparison into scheduling trigger judgment, the transformation from passive response to proactive prediction is realized, which significantly reduces the probability of invalid scheduling triggers and improves link stability.
[0018] In addition, the data processing method proposed above in this application may also have the following additional technical features:
[0019] As a further description of the above technical solution:
[0020] Step 1) extracts the 6-dimensional traffic feature vector from real-time traffic data, specifically including:
[0021] Acquire the duration and slope of real-time traffic data changes within a preset sliding time window;
[0022] Detect the traffic linkage between upstream and downstream related links of the target link and extract the collaborative features of related links;
[0023] Statistical analysis of the distribution characteristics of traffic pulse width and data packet size in real-time traffic data;
[0024] The port protocol type corresponding to the real-time traffic data is analyzed, and the above features are integrated to form a 6-dimensional traffic feature vector.
[0025] To ensure the completeness and standardization of feature collection, avoid the defects of missing key information in a single feature, provide high-quality data support for the accurate identification of subsequent AI models, and improve the matching degree between feature vectors and traffic fluctuation characteristics.
[0026] As a further description of the above technical solution:
[0027] Step 3) involves implementing a hierarchical filtering strategy based on noise probability, specifically including:
[0028] When the noise probability is ≥90%, the corresponding real-time traffic data is processed by a combination of low-weight weakening and labeling and discarding, and the real-time traffic data is given a low weight coefficient, and the data does not trigger traffic scheduling adjustment.
[0029] When the noise probability is 60% ≤ noise probability < 90%, the corresponding real-time traffic data is processed by a combination of dynamic weighted smoothing and temporary verification. The real-time traffic data is assigned a medium to low weight coefficient and is re-evaluated by trend verification in a subsequent sliding time window. If it is still determined to be noise, it is marked as discarded.
[0030] When the noise probability is less than 60%, the corresponding real-time traffic data is assigned a high weighting coefficient and incorporated into the calculation process of the smoothed traffic value to ensure that the true traffic trend is not lost.
[0031] By employing a tiered filtering strategy to process traffic data differently based on noise probability, high-probability noise is accurately filtered without interfering with scheduling, medium-probability suspected noise is temporarily stored for verification to avoid misjudgment, and low-probability real traffic is given high weight to retain trends, achieving a balance between filtering noise and retaining real trends, which is superior to the effect of traditional fixed-weight smoothing or one-size-fits-all filtering.
[0032] As a further description of the above technical solution:
[0033] Step 4) determines whether to trigger traffic scheduling adjustments based on the comparison results, specifically including:
[0034] If the measured flow rate is within the predicted flow rate range, the current flow rate change is determined to conform to the preset trend, and no flow scheduling adjustment is triggered.
[0035] If the measured flow rate exceeds the predicted flow rate range, and the duration of this excess reaches the preset number of sampling periods, it is determined to be a non-trend-based flow change, triggering the flow scheduling adjustment assessment process.
[0036] By employing a two-dimensional judgment logic of predictive interval comparison and duration verification, the system accurately distinguishes between trend fluctuations and non-trend mutations, thereby reducing frequent scheduling caused by small, high-frequency fluctuations at the source, further enhancing the stability of scheduling strategies, and ensuring service quality in highly sensitive scenarios.
[0037] As a further description of the above technical solution:
[0038] Step 5) describes application scenarios including low-latency sensitive scenarios, high-stability sensitive scenarios, and hybrid scenarios. The specific details of dynamically configuring scheduling constraint parameters based on the application scenario type include:
[0039] For low-latency sensitive scenarios, configure a first minimum adjustment interval and a first threshold for traffic change magnitude;
[0040] For highly stable and sensitive scenarios, a second minimum adjustment interval and a second traffic change magnitude threshold are configured, wherein the second minimum adjustment interval is greater than the first minimum adjustment interval and the second traffic change magnitude threshold is greater than the first traffic change magnitude threshold.
[0041] For mixed scenarios, a third minimum adjustment interval and a third traffic change magnitude threshold are configured, with the third minimum adjustment interval and the third traffic change magnitude threshold respectively falling between the corresponding first parameter and second parameter.
[0042] The above solution dynamically configures constraint parameters according to scenario type, accurately adapts to the differentiated needs of low latency, high stability, and mixed scenarios, solves the pain point that traditional fixed parameters cannot take into account the needs of multiple scenarios, and achieves the optimal balance between real-time performance and stability in different scenarios.
[0043] The second aspect of this application proposes a network management device for data processing, including a traffic acquisition unit for acquiring real-time traffic data of a target link and extracting a 6-dimensional traffic feature vector from the real-time traffic data.
[0044] The preprocessing unit has a built-in lightweight CNN-LSTM hybrid model, which is used to receive a 6-dimensional traffic feature vector and output the noise probability and the true change probability of the corresponding real-time traffic data. It executes a hierarchical filtering strategy to perform weighted smoothing on the real-time traffic data to obtain smoothed traffic values.
[0045] The scheduling optimization unit has a built-in improved ARIMA model, which is used to perform short-term traffic trend prediction based on smoothed traffic values, output predicted traffic ranges, and determine whether to trigger traffic scheduling adjustments by comparing the measured traffic values with the predicted traffic ranges.
[0046] The protocol parsing unit is used to identify the application scenario type carried by the current link and output the scenario identification result;
[0047] The control unit is connected to the preprocessing unit, the scheduling optimization unit, and the protocol parsing unit, respectively. It is used to dynamically configure scheduling constraint parameters based on the scene recognition results and execute traffic scheduling strategies that match the application scene.
[0048] The above solution ensures efficient integration of data processing and scheduling strategy execution, improves the feasibility and operational reliability of the solution, and adapts to the actual hardware architecture of the network management device.
[0049] In addition, the network management device for data processing proposed in this application may also have the following additional technical features:
[0050] As a further description of the above technical solution:
[0051] The preprocessing unit also integrates an AI acceleration unit, which is used to carry the computational logic of the lightweight CNN-LSTM hybrid model and improve the output efficiency of noise probability and true change probability.
[0052] By using an AI acceleration unit to specifically handle the hybrid model computation logic, the output efficiency of noise probability and true change probability is significantly improved, the computational load of the preprocessing unit's main processor is reduced, the real-time performance of data processing in high-traffic concurrent scenarios is ensured, and scheduling decisions are avoided due to model computation delays.
[0053] As a further description of the above technical solution:
[0054] The device also includes a collaborative communication module, which is connected to the control unit and is used to realize incremental data interaction between edge nodes and central management nodes. The incremental data includes link jitter data and parameter adjustment requirement data.
[0055] By adopting an incremental data interaction mode through the collaborative communication module, the communication bandwidth consumption is reduced while achieving edge-center collaboration. By adjusting the global and local parameters differently, the limitations of single-node decision-making in distributed scenarios are resolved, global policy conflicts or local over-filtering are avoided, and the adaptability of the solution in large-scale distributed networks is improved.
[0056] As a further description of the above technical solution:
[0057] The device also includes a scenario feature configuration interface, which is connected to the protocol parsing unit to support user-defined application scenario types carried by the link and scheduling constraint parameters corresponding to each application scenario.
[0058] The scenario feature configuration interface allows users to customize scenario types and constraint parameters, breaking through the limitations of fixed scenario adaptation and enhancing the flexibility and scalability of the solution, which can adapt to the personalized network management needs of different industries and businesses.
[0059] As a further description of the above technical solution:
[0060] The scheduling optimization unit also includes a parameter calibration subunit. The parameter calibration subunit is used to calibrate the parameters of the improved ARIMA model according to a preset period based on the deviation between the actual traffic trend and the model's predicted traffic trend, so as to ensure that the model's prediction accuracy is adapted to the current network status.
[0061] The parameter calibration subunit dynamically calibrates the ARIMA model parameters at a preset cycle, ensuring that the prediction accuracy adapts dynamically to the network status, avoiding increased prediction deviation due to changes in the network environment, improving the reliability of short-term traffic trend prediction, and providing accurate basis for scheduling trigger determination.
[0062] Advantages of this invention:
[0063] According to the data processing method and network management device of this application, high-precision hierarchical filtering of traffic noise is achieved through 6-dimensional feature fusion and lightweight AI model, taking into account both noise filtering effect and preservation of real traffic trend.
[0064] By introducing a timing prediction mechanism and a two-dimensional scheduling trigger determination, the link jitter caused by frequent scheduling is reduced from the root; scenario-based dynamic constraint parameters and custom configuration interfaces improve the flexibility of the solution to adapt to multiple scenarios.
[0065] Edge-center collaborative communication and dynamic model calibration enhance global optimization capabilities and long-term robustness in distributed scenarios.
[0066] The modules work closely together, which not only solves the pain points of existing technologies such as low noise identification accuracy, poor scheduling stability, and insufficient scenario adaptability, but also ensures real-time performance through hardware unit optimization, significantly improving the traffic scheduling quality and operational reliability of the network management device.
[0067] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0068] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0069] Figure 1 This is a schematic flowchart of a data processing method and a network management device according to an embodiment of this application;
[0070] Figure 2 This is a schematic block diagram of a data processing method and network management device according to an embodiment of this application. Detailed Implementation
[0071] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.
[0072] The data processing method and network management device of the present application embodiments are described below with reference to the accompanying drawings.
[0073] like Figure 1-2 As shown in Embodiment 1 of this application, a distributed enterprise office network is used as the application scenario. The network includes one central management node and five edge management nodes, covering three types of scenarios: low latency sensitive scenarios (such as video conferencing), high stability sensitive scenarios (such as financial data transmission), and mixed scenarios (such as daily office work).
[0074] The traffic acquisition unit uses an Intel I350 Gigabit network card with a sampling period of 100ms. It supports the simultaneous acquisition of real-time traffic data from 5 links and the data transmission interface is PCIe 3.0.
[0075] The preprocessing unit uses the Xilinx Zynq UltraScale+ MPSoC as its core chip, which integrates an AI acceleration unit;
[0076] The scheduling optimization unit uses an Intel Core i7-12700 processor with a main frequency of 2.7GHz and has a built-in improved ARIMA model calculation module;
[0077] The protocol parsing unit supports parsing mainstream protocols such as TCP / IP, UDP, and SIP, with a scene recognition response time of ≤50ms;
[0078] The control unit uses an STM32H743 microcontroller, which is responsible for the coordinated control of each unit and the execution of scheduling strategies.
[0079] The collaborative communication module adopts a 5G industrial module (model Quectel AG550Q), which supports incremental data interaction between edge nodes and central nodes, with a communication bandwidth of ≥10Mbps;
[0080] The scene feature configuration interface adopts an RS485 bus interface, which allows users to customize scene parameters through a host computer.
[0081] The following are the specific steps for implementing AI-enhanced noise recognition and hierarchical filtering:
[0082] 1) Extraction of 6-dimensional flow feature vector:
[0083] After collecting real-time traffic data from each link through the traffic acquisition unit, the 6-dimensional features are extracted in the following manner:
[0084] Duration of change: The duration t of the statistical flow rate deviating from the mean of the last 5 sampling periods;
[0085] Slope of change: The rate of change k of the flow rate over time is calculated using the following formula: ;
[0086] in , They are respectively , The flow rate at any given moment;
[0087] Link coordination: Let the change in traffic on the target link be... The average change in traffic of upstream and downstream related links is Cooperative eigenvalues:
[0088] ;
[0089] like Then let =1;
[0090] Flow pulse width: The duration of a single flow fluctuation. t;
[0091] Data packet size distribution: Calculate the variance σ of the data packet size within the sampling period using the following formula:
[0092] ;
[0093] in This represents the average data packet size. This represents the number of data packets within the sampling period.
[0094] Port protocol type: The port number and corresponding protocol are identified by the protocol parsing unit and encoded according to preset rules (e.g., TCP protocol is encoded as 1, UDP protocol is encoded as 2, and SIP protocol is encoded as 3).
[0095] 2) Lightweight CNN-LSTM hybrid model configuration:
[0096] Model structure: The CNN part contains 2 convolutional layers (3×3 kernel size, stride 1, padding=same) and 1 pooling layer (2×2 kernel size).
[0097] The LSTM part contains one hidden layer with 64 neurons, and the output layer has two neurons (outputting noise probability and true change probability respectively).
[0098] The total number of model parameters was compressed to 800,000. The Adam optimizer was used for training. The training dataset consisted of one month's worth of historical traffic data from the enterprise network (including 100,000 noisy data points and 200,000 real traffic change data points).
[0099] 3) Implementation of hierarchical filtering strategy:
[0100] High-probability noise (probability ≥ 90%): assigned a weight coefficient of 0.05, marked as noise data and discarded, not participating in scheduling calculations;
[0101] Medium probability suspected noise (60%≤probability<90%): Assign a weight coefficient of 0.2, temporarily store in the buffer, and wait for the next sampling period (100ms) to be judged again. If it is still suspected noise, it is discarded.
[0102] Low probability of actual change (probability < 60%): Assigned a weighting coefficient of 0.8, and participates in the calculation of smoothed flow value. The calculation formula is:
[0103] ;
[0104] in ᵢ is the first Weighting coefficients for each sampling period For the first The flow rate value for each sampling period.
[0105] Implementation of short-term time series forecasting:
[0106] Short-term flow trend prediction is performed using an improved ARIMA model (ARIMA(1,1,1)).
[0107] (1) Model input: Smoothed flow rate values over nearly 60 sampling periods (6s) ;
[0108] (2) Predicted output: Predicted flow range for the next two sampling periods (200ms) The prediction interval is calculated using a 95% confidence interval, and the formula is as follows:
[0109] , ;
[0110] in The mean flow rate predicted by the model. For the predicted standard deviation;
[0111] (3) Prediction-measurement collaborative judgment: If the current measured flow value ∈ If so, scheduling adjustments will not be triggered; if ∉ If the duration is ≥1 sampling period (100ms), then a scheduling adjustment assessment will be triggered.
[0112] (4) Model dynamic calibration: The deviation between the actual flow trend and the predicted trend is calculated every 5 minutes (3000 sampling periods).
[0113] ;
[0114] in The actual average flow rate during the calibration period. The average predicted flow rate during the calibration period;
[0115] when When the value is greater than 5%, adjust the autoregressive coefficient of the model. and moving average coefficient (Adjust the step size to 0.01) to ensure prediction accuracy.
[0116] Implementation of dynamic constraints based on different scenarios:
[0117] For low-latency sensitive scenarios (such as video conferencing), the minimum adjustment interval is 200ms, and the threshold for quantity change is 8%.
[0118] For highly stable and sensitive scenarios (such as financial data transmission), the minimum adjustment interval is 800ms, and the threshold for quantity change is 15%.
[0119] For mixed scenarios (such as daily office work), the minimum adjustment interval is 500ms, and the threshold for the magnitude of change is 12%.
[0120] When switching scenes, the control unit completes the parameter switching within one sampling period (100ms) to ensure that the scheduling strategy can quickly adapt to the scene requirements.
[0121] Edge-Center Collaborative Optimization Implementation:
[0122] (1) Local processing of edge nodes: Each edge management node completes noise filtering through the local preprocessing unit, collects link jitter data (latency jitter amplitude D, packet loss rate fluctuation value L), and uploads incremental data (containing only D, L and parameter adjustment requirement markers) to the central node every 200ms.
[0123] (2) Global optimization of the central node: The central node summarizes the data of the five edge nodes. If ≥3 edge nodes have D>20ms and L>1%, it is judged as global network fluctuation. The sliding window length of each edge node is uniformly adjusted from 500ms to 800ms, and the noise variance judgment threshold is adjusted from 0.1 to 0.15. If only a single edge node has the above situation, only the sliding window length of that node is adjusted to 600ms, and the parameters of other nodes remain unchanged.
[0124] (3) Collaborative feedback: The central node sends the optimized parameters to each edge node through the collaborative communication module, and the edge node completes the parameter update within one sampling period.
[0125] According to the above implementation method, the complete implementation process is as follows:
[0126] 1. Start the network management device and preset the constraint parameters and initial parameters of the model for three types of application scenarios through the scenario feature configuration interface;
[0127] 2. The traffic acquisition unit collects real-time traffic data for each link at a 100ms cycle and extracts a 6-dimensional traffic feature vector;
[0128] 3. The preprocessing unit outputs noise probability and true change probability through a lightweight CNN-LSTM hybrid model, and performs hierarchical filtering to obtain smoothed flow values;
[0129] 4. The scheduling optimization unit predicts the traffic range for the next 200ms based on the smoothed traffic value using an improved ARIMA model, compares the measured traffic value with the predicted range, and determines whether to trigger scheduling adjustments.
[0130] 5. The protocol parsing unit identifies the current application scenario, and the control unit configures the corresponding constraint parameters based on the scenario type and executes the traffic scheduling strategy;
[0131] 6. Edge nodes collect link jitter data and upload it to the central node. The central node then optimizes global parameters and issues updates.
[0132] Repeat steps 2-6 to achieve continuous optimization of traffic data processing and scheduling.
[0133] In summary, the data processing method and network management device according to the embodiments of this application achieve a noise identification accuracy of ≥98%, reduce the frequency of scheduling strategy adjustment by 75%, and reduce link jitter by ≤10ms. The service quality in each scenario meets the enterprise network requirements.
[0134] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0135] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0136] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.
Claims
1. A data processing method, characterized in that, Includes the following steps: 1) Collect real-time traffic data of the target link and extract a 6-dimensional traffic feature vector from the real-time traffic data. The 6-dimensional traffic feature vector specifically includes the duration of change, the slope of change, the coordination of associated links, the traffic pulse width, the data packet size distribution and the port protocol type. 2) Input the 6-dimensional traffic feature vector into a pre-set lightweight CNN-LSTM hybrid model, and output the noise probability and the true change probability of the corresponding real-time traffic data through the hybrid model; 3) Implement a hierarchical filtering strategy based on noise probability to perform weighted smoothing on real-time traffic data and obtain smoothed traffic values; 4) Based on the smoothed flow value, short-term flow trend prediction is performed using an improved ARIMA model, and the predicted flow range is output. The filtered measured flow value is compared with the predicted flow range, and the flow scheduling adjustment is determined based on the comparison result. 5) Identify the application scenario carried by the current link through the protocol parsing module, dynamically configure scheduling constraint parameters based on the type of application scenario, and execute traffic scheduling strategies that match the application scenario.
2. The data processing method according to claim 1, characterized in that, Step 1) extracts the 6-dimensional traffic feature vector from real-time traffic data, specifically including: Acquire the duration and slope of real-time traffic data changes within a preset sliding time window; Detect the traffic linkage between upstream and downstream related links of the target link and extract the collaborative features of related links; Statistical analysis of the distribution characteristics of traffic pulse width and data packet size in real-time traffic data; The port protocol type corresponding to the real-time traffic data is analyzed, and the above features are integrated to form a 6-dimensional traffic feature vector.
3. The data processing method according to claim 1, characterized in that, Step 3) involves implementing a hierarchical filtering strategy based on noise probability, specifically including: When the noise probability is ≥90%, the corresponding real-time traffic data is processed by a combination of low-weight weakening and labeling and discarding, and the real-time traffic data is given a low weight coefficient, and the data does not trigger traffic scheduling adjustment. When the noise probability is 60% ≤ noise probability < 90%, the corresponding real-time traffic data is processed by a combination of dynamic weighted smoothing and temporary verification. The real-time traffic data is assigned a medium to low weight coefficient and is re-evaluated by trend verification in a subsequent sliding time window. If it is still determined to be noise, it is marked as discarded. When the noise probability is less than 60%, the corresponding real-time traffic data is assigned a high weighting coefficient and incorporated into the calculation process of the smoothed traffic value to ensure that the true traffic trend is not lost.
4. The data processing method according to claim 1, characterized in that, Step 4) determines whether to trigger traffic scheduling adjustments based on the comparison results, specifically including: If the measured flow rate is within the predicted flow rate range, the current flow rate change is determined to conform to the preset trend, and no flow scheduling adjustment is triggered. If the measured flow rate exceeds the predicted flow rate range, and the duration of this excess reaches the preset number of sampling periods, it is determined to be a non-trend-based flow change, triggering the flow scheduling adjustment assessment process.
5. The data processing method according to claim 1, characterized in that, Step 5) describes application scenarios including low-latency sensitive scenarios, high-stability sensitive scenarios, and hybrid scenarios. The specific details of dynamically configuring scheduling constraint parameters based on the application scenario type include: For low-latency sensitive scenarios, configure a first minimum adjustment interval and a first threshold for traffic change magnitude; For highly stable and sensitive scenarios, a second minimum adjustment interval and a second traffic change magnitude threshold are configured, wherein the second minimum adjustment interval is greater than the first minimum adjustment interval and the second traffic change magnitude threshold is greater than the first traffic change magnitude threshold. For mixed scenarios, a third minimum adjustment interval and a third traffic change magnitude threshold are configured, with the third minimum adjustment interval and the third traffic change magnitude threshold respectively falling between the corresponding first parameter and second parameter.
6. The network management device for data processing according to any one of claims 1-5, characterized in that, include: The traffic acquisition unit is used to collect real-time traffic data of the target link and extract a 6-dimensional traffic feature vector from the real-time traffic data. The preprocessing unit has a built-in lightweight CNN-LSTM hybrid model, which is used to receive a 6-dimensional traffic feature vector and output the noise probability and the true change probability of the corresponding real-time traffic data. It executes a hierarchical filtering strategy to perform weighted smoothing on the real-time traffic data to obtain smoothed traffic values. The scheduling optimization unit has a built-in improved ARIMA model, which is used to perform short-term traffic trend prediction based on smoothed traffic values, output predicted traffic ranges, and determine whether to trigger traffic scheduling adjustments by comparing the measured traffic values with the predicted traffic ranges. The protocol parsing unit is used to identify the application scenario type carried by the current link and output the scenario identification result; The control unit is connected to the preprocessing unit, the scheduling optimization unit, and the protocol parsing unit, respectively. It is used to dynamically configure scheduling constraint parameters based on the scene recognition results and execute traffic scheduling strategies that match the application scene.
7. The network management device for data processing according to claim 6, characterized in that, The preprocessing unit also integrates an AI acceleration unit, which is used to carry the computational logic of the lightweight CNN-LSTM hybrid model and improve the output efficiency of noise probability and true change probability.
8. The network management device for data processing according to claim 6, characterized in that, The device also includes a collaborative communication module, which is connected to the control unit and is used to realize incremental data interaction between edge nodes and central management nodes. The incremental data includes link jitter data and parameter adjustment requirement data.
9. The network management device for data processing according to claim 6, characterized in that, The device also includes a scenario feature configuration interface, which is connected to the protocol parsing unit to support user-defined application scenario types carried by the link and scheduling constraint parameters corresponding to each application scenario.
10. The network management device according to claim 6, characterized in that, The scheduling optimization unit also includes a parameter calibration subunit. The parameter calibration subunit is used to calibrate the parameters of the improved ARIMA model according to a preset period based on the deviation between the actual traffic trend and the model's predicted traffic trend, so as to ensure that the model's prediction accuracy is adapted to the current network status.