Network flow dynamic optimization method based on time sequence analysis

Abstract

The invention discloses a network flow dynamic optimization method based on time sequence analysis, and relates to the technical field of cloud computing networks. Comprising the steps of 1, collecting data, 2, establishing a multi-scale time sequence model, and 3, training and optimizing the multi-scale time sequence model. 4, deploying the multi-scale time sequence model into the production environment, predicting the network flow in real time, and monitoring the performance data of the multi-scale time sequence model and the change of the network flow in real time; and 5, evaluating and adjusting the multi-scale time sequence model: according to the monitored performance data, evaluating the prediction performance of the multi-scale time sequence model by using MAE and RMSE indexes, evaluating the generalization ability of the multi-scale time sequence model by using cross validation, and adjusting the multi-scale time sequence model according to the evaluation result.

Description

Technical Field

[0001] The invention discloses a network traffic dynamic optimization method based on time series analysis, and relates to the technical field of cloud computing networks. Background Art

[0002] With the rapid development of cloud computing and big data technology, network traffic management is becoming increasingly important. Network traffic is not only related to the efficiency and stability of data transmission, but also a key factor in network security and service quality. However, network traffic is highly dynamic and uncertain, which brings great challenges to the optimization and management of network traffic. Static traffic management strategies in existing technologies are often static, and traffic control is performed based on preset rules and thresholds, which cannot adapt to the rapid changes and unpredictable peaks of network traffic. In addition, there is a lack of real-time monitoring and it is impossible to respond to fluctuations in network traffic in a timely manner, resulting in congestion and delays during peak traffic. Moreover, static traffic management is a single-scale analysis, and the traffic analysis method only considers a single time scale, which cannot fully capture the short-term fluctuations and long-term trends of network traffic. At the same time, static traffic management resources are unevenly allocated. In a multi-user and multi-task cloud computing environment, resource allocation often lacks flexibility and cannot dynamically adjust resource allocation according to real-time traffic. There is also the problem that traffic prediction models are usually not accurate enough, and cannot provide reliable decision support for network traffic management. Summary of the invention

[0003] In view of the problems of the prior art, the present invention provides a method for dynamic optimization of network traffic based on time series analysis, realizes real-time monitoring and historical data analysis of network traffic data, constructs a multi-scale time series model that captures characteristics of different time scales, and uses the multi-scale time series model to realize dynamic optimization of network traffic.

[0004] The specific scheme proposed by the present invention is:

[0005] The present invention provides a method for dynamic optimization of network traffic based on time series analysis, comprising:

[0006] Step 1: Collect data: Collect historical network traffic data and deploy network monitoring tools to collect real-time network traffic data. Network traffic data includes packet size, transmission rate, packet loss rate, and delay data.

[0007] Step 2: Establish a multi-scale time series model: Extract time series features from the collected data and create a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer.

[0008] The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer.

[0009] Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern;

[0010] Introducing the attention mechanism in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction;

[0011] Step 3: Train and tune the multi-scale time series model: Use time series features to train the multi-scale time series model;

[0012] Select the loss function for the regression problem, use the optimizer to optimize the parameters of the multi-scale time series model, apply the L1 or L2 regularization algorithm to adjust the multi-scale time series model and combine it with the early stopping method to prevent overfitting;

[0013] Step 4: Deploy the multi-scale time series model to the production environment to predict network traffic in real time, and monitor the performance data of the multi-scale time series model and changes in network traffic in real time;

[0014] Step 5: Evaluate and adjust the multi-scale time series model: Based on the monitored performance data, use the MAE and RMSE indicators to evaluate the prediction performance of the multi-scale time series model, use cross-validation to evaluate the generalization ability of the multi-scale time series model, and adjust the multi-scale time series model based on the evaluation results.

[0015] Furthermore, in step 1 of the method for dynamic optimization of network traffic based on time series analysis, the collected data is preprocessed:

[0016] Perform data cleaning: handle missing values, outliers and noisy data to ensure data quality.

[0017] Perform data normalization to meet the input requirements of the LSTM model.

[0018] Furthermore, in step 2 of the method for dynamic optimization of network traffic based on time series analysis, extracting time series features from the collected data includes:

[0019] Define different time scales based on the characteristics of network delay and business requirements.

[0020] At each defined time scale, the corresponding features are extracted:

[0021] Extract second-level features, including instantaneous value, rate of change, and peak value of delay.

[0022] Extract minute-level features, including average delay, delay volatility, and cumulative delay.

[0023] Extract hourly features, including latency trends, periodic changes, and peak and valley values ​​within the hour.

[0024] Extract daily features, including the average daily delay, the difference in delay between weekdays and weekends, and the seasonal variation of delay.

[0025] Furthermore, in step 2 of the method for dynamic optimization of network traffic based on time series analysis, the extracted time series features are transformed: logarithmic transformation, difference, and moving average transformation processing are performed on the time series features to enhance the model's ability to capture trends and periodicity;

[0026] Perform feature fusion on the extracted time series features: fuse features of different scales to form the input of the model. The fusion methods include splicing or weighted fusion, where the weights are dynamically adjusted according to the importance of the features;

[0027] Perform feature selection on the extracted time series features: Use correlation-based selection methods to remove redundant or unimportant features to improve the performance and interpretability of the model.

[0028] Furthermore, in step 3 of the network traffic dynamic optimization method based on time series analysis, a loss function for mean square error MSE or mean absolute error MAE is selected, and the multi-scale time series model is optimized using the Adam optimizer, wherein the network delay parameters are set using the Adam optimizer: the initial learning rate is set, β1 is set to 0.9, β2 is set to 0.999, and ε is a constant to ensure numerical stability.

[0029] The present invention provides a network traffic dynamic optimization device based on time series analysis, including a data collection module, a multi-scale time series model construction module, a training and tuning module, a model deployment and maintenance module, and an evaluation and testing module.

[0030] The data collection module collects data: collects historical network traffic data, and deploys network monitoring tools to collect real-time network traffic data. Network traffic data includes data packet size, transmission rate, packet loss rate, and delay data.

[0031] Multi-scale time series model building module builds a multi-scale time series model: extracts time series features from the collected data, and creates a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer.

[0032] The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer.

[0033] Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern;

[0034] Introducing the attention mechanism in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction;

[0035] Training and tuning module Training and tuning of multi-scale time series models: using time series features to train multi-scale time series models, selecting loss functions for regression problems, using optimizers to optimize parameters of multi-scale time series models, applying L1 or L2 regularization algorithms to adjust multi-scale time series models and combining early stopping methods to prevent overfitting;

[0036] The model deployment and maintenance module deploys the multi-scale time series model to the production environment, predicts network traffic in real time, and monitors the performance data of the multi-scale time series model and changes in network traffic in real time;

[0037] The evaluation and testing module evaluates and adjusts the multi-scale time series model: Based on the monitored performance data, the MAE and RMSE indicators are used to evaluate the prediction performance of the multi-scale time series model, and cross-validation is used to evaluate the generalization ability of the multi-scale time series model. The multi-scale time series model is adjusted according to the evaluation results.

[0038] Furthermore, the data collection module of the network traffic dynamic optimization device based on time series analysis preprocesses the collected data:

[0039] Perform data cleaning: handle missing values, outliers and noisy data to ensure data quality.

[0040] Perform data normalization to meet the input requirements of the LSTM model.

[0041] Furthermore, the multi-scale time series model construction module of the network traffic dynamic optimization device based on time series analysis extracts time series features from the collected data, including:

[0042] Define different time scales based on the characteristics of network delay and business requirements.

[0043] At each defined time scale, the corresponding features are extracted:

[0044] Extract second-level features, including instantaneous value, rate of change, and peak value of delay.

[0045] Extract minute-level features, including average delay, delay volatility, and cumulative delay.

[0046] Extract hourly features, including latency trends, periodic changes, and peak and valley values ​​within the hour.

[0047] Extract daily features, including the average daily delay, the difference in delay between weekdays and weekends, and the seasonal variation of delay.

[0048] Furthermore, the multi-scale time series model construction module of the network traffic dynamic optimization device based on time series analysis performs feature conversion on the extracted time series features: logarithmic conversion, difference, and moving average conversion processing are performed on the time series features to enhance the model's ability to capture trends and periodicity;

[0049] Perform feature fusion on the extracted time series features: fuse features of different scales to form the input of the model. The fusion methods include splicing or weighted fusion, where the weights are dynamically adjusted according to the importance of the features;

[0050] Perform feature selection on the extracted time series features: Use correlation-based selection methods to remove redundant or unimportant features to improve the performance and interpretability of the model.

[0051] Furthermore, the training and tuning module of the network traffic dynamic optimization device based on time series analysis selects a loss function for the mean square error MSE or the mean absolute error MAE, and uses the Adam optimizer to optimize the multi-scale time series model, wherein the Adam optimizer is used to set the network delay parameters: set the initial learning rate, set β1 to 0.9, β2 to 0.999, and ε to a constant to ensure numerical stability.

[0052] The benefits of the present invention are:

[0053] 1. Improve prediction accuracy: By constructing a multi-scale time series model based on the long short-term memory network (LSTM), the present invention can effectively capture the trend, seasonality, and periodicity characteristics of network traffic data. This multi-scale analysis method can simultaneously consider short-term and long-term time dependencies, significantly improving the accuracy of network traffic prediction.

[0054] 2. Enhance the adaptability and robustness of the system: The adaptive learning rate adjustment strategy and attention mechanism introduced enable the model to dynamically adjust according to real-time data, paying more attention to the time points that have a greater impact on the prediction results. In addition, the application of ensemble learning further enhances the robustness of the model, allowing the system to maintain stable operation in the face of uncertainty and fluctuations in network traffic.

[0055] 3. Optimize resource allocation: The present invention can intelligently adjust network resource allocation, such as bandwidth, storage and computing resources, based on prediction results through real-time monitoring and dynamic optimization methods. This not only improves the efficiency of resource use, but also helps reduce operating costs while ensuring the quality of service for different users and services.

[0056] 4. Improve network performance and user experience: Through accurate traffic prediction and timely resource adjustment, the present invention can effectively reduce network congestion and delay and improve the overall performance of the network. For cloud computing service providers, this means that they can provide users with a faster, more stable and reliable service experience and enhance user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a schematic diagram of the application framework of the device of the present invention. DETAILED DESCRIPTION

[0058] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0059] Example 1

[0060] The present invention provides a method for dynamic optimization of network traffic based on time series analysis, comprising:

[0061] Step 1: Collect data: Collect historical network traffic data and deploy network monitoring tools to collect real-time network traffic data. Network traffic data includes packet size, transmission rate, packet loss rate, and delay data.

[0062] In step 1, the collected data is preprocessed:

[0063] Perform data cleaning: handle missing values, outliers and noisy data to ensure data quality.

[0064] Perform data normalization to meet the input requirements of the LSTM model.

[0065] Step 2: Establish a multi-scale time series model: Extract time series features from the collected data and create a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer.

[0066] The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer.

[0067] Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern;

[0068] The attention mechanism is introduced in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction.

[0069] The time series features extracted from the collected data include:

[0070] Define different time scales based on the characteristics of network delay and business requirements.

[0071] At each defined time scale, the corresponding features are extracted:

[0072] Extract second-level features, including instantaneous value, rate of change, and peak value of delay.

[0073] Extract minute-level features, including average delay, delay volatility, and cumulative delay.

[0074] Extract hourly features, including latency trends, periodic changes, and peak and valley values ​​within the hour.

[0075] Extract daily features, including the average daily delay, the difference in delay between weekdays and weekends, and the seasonal variation of delay.

[0076] Perform feature transformation on the extracted time series features: Perform logarithmic transformation, difference, and moving average transformation on the time series features to enhance the model's ability to capture trends and periodicity;

[0077] Perform feature fusion on the extracted time series features: fuse features of different scales to form the input of the model. The fusion methods include splicing or weighted fusion, where the weights are dynamically adjusted according to the importance of the features;

[0078] Perform feature selection on the extracted time series features: Use correlation-based selection methods to remove redundant or unimportant features to improve the performance and interpretability of the model.

[0079] Step 3: Train and tune the multi-scale time series model: Use time series features to train the multi-scale time series model;

[0080] Select the loss function for the regression problem, use the optimizer to optimize the parameters of the multi-scale time series model, apply the L1 or L2 regularization algorithm to adjust the multi-scale time series model and combine it with the early stopping method to prevent overfitting.

[0081] The loss function of mean square error (MSE) or mean absolute error (MAE) is selected, and the Adam optimizer is used to optimize the multi-scale time series model. The Adam optimizer is used to set the network delay parameters: set the initial learning rate, set β1 to 0.9, β2 to 0.999, and ε to a constant to ensure numerical stability.

[0082] Step 4: Deploy the multi-scale time series model to the production environment to predict network traffic in real time, and monitor the performance data of the multi-scale time series model and changes in network traffic in real time;

[0083] Step 5: Evaluate and adjust the multi-scale time series model: Based on the monitored performance data, use the MAE and RMSE indicators to evaluate the prediction performance of the multi-scale time series model, use cross-validation to evaluate the generalization ability of the multi-scale time series model, and adjust the multi-scale time series model based on the evaluation results.

[0084] Example 2

[0085] The present invention provides a network traffic dynamic optimization device based on time series analysis, including a data collection module, a multi-scale time series model construction module, a training and tuning module, a model deployment and maintenance module, and an evaluation and testing module.

[0086] The data collection module collects data: collects historical network traffic data, and deploys network monitoring tools to collect real-time network traffic data. Network traffic data includes data packet size, transmission rate, packet loss rate, and delay data.

[0087] Multi-scale time series model building module builds a multi-scale time series model: extracts time series features from the collected data, and creates a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer.

[0088] The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer.

[0089] Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern;

[0090] Introducing the attention mechanism in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction;

[0091] Training and tuning module Training and tuning of multi-scale time series models: using time series features to train multi-scale time series models, selecting loss functions for regression problems, using optimizers to optimize parameters of multi-scale time series models, applying L1 or L2 regularization algorithms to adjust multi-scale time series models and combining early stopping methods to prevent overfitting;

[0092] The model deployment and maintenance module deploys the multi-scale time series model to the production environment, predicts network traffic in real time, and monitors the performance data of the multi-scale time series model and changes in network traffic in real time;

[0093] The evaluation and testing module evaluates and adjusts the multi-scale time series model: Based on the monitored performance data, the MAE and RMSE indicators are used to evaluate the prediction performance of the multi-scale time series model, and cross-validation is used to evaluate the generalization ability of the multi-scale time series model. The multi-scale time series model is adjusted according to the evaluation results.

[0094] As the information interaction and execution process between the modules of the above-mentioned device are based on the same concept as the embodiment of the method of the present invention, the specific contents can be found in the description of the embodiment of the method of the present invention and will not be repeated here.

[0095] Likewise, the benefits of the device of the present invention are:

[0096] 1. Improve prediction accuracy: By constructing a multi-scale time series model based on the long short-term memory network (LSTM), the present invention can effectively capture the trend, seasonality, and periodicity characteristics of network traffic data. This multi-scale analysis method can simultaneously consider short-term and long-term time dependencies, significantly improving the accuracy of network traffic prediction.

[0097] 2. Enhance the adaptability and robustness of the system: The adaptive learning rate adjustment strategy and attention mechanism introduced enable the model to dynamically adjust according to real-time data, paying more attention to the time points that have a greater impact on the prediction results. In addition, the application of ensemble learning further enhances the robustness of the model, allowing the system to maintain stable operation in the face of uncertainty and fluctuations in network traffic.

[0098] 3. Optimize resource allocation: The present invention can intelligently adjust network resource allocation, such as bandwidth, storage and computing resources, based on prediction results through real-time monitoring and dynamic optimization methods. This not only improves the efficiency of resource use, but also helps reduce operating costs while ensuring the quality of service for different users and services.

[0099] 4. Improve network performance and user experience: Through accurate traffic prediction and timely resource adjustment, the present invention can effectively reduce network congestion and delay and improve the overall performance of the network. For cloud computing service providers, this means that they can provide users with a faster, more stable and reliable service experience and enhance user satisfaction.

[0100] It should be noted that not all steps and modules in the above-mentioned processes and device structures are necessary, and some steps or modules can be ignored according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in the above-mentioned embodiments can be a physical structure or a logical structure, that is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or some components in multiple independent devices may be implemented together.

[0101] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. A method for dynamic optimization of network traffic based on time series analysis, characterized by include: Step 1: Collect data: Collect historical network traffic data and deploy network monitoring tools to collect real-time network traffic data. Network traffic data includes packet size, transmission rate, packet loss rate, and delay data. Step 2: Establish a multi-scale time series model: Extract time series features from the collected data and create a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer. The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer. Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern; Introducing the attention mechanism in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction; Step 3: Train and tune the multi-scale time series model: Use time series features to train the multi-scale time series model; Select the loss function for the regression problem, use the optimizer to optimize the parameters of the multi-scale time series model, apply the L1 or L2 regularization algorithm to adjust the multi-scale time series model and combine it with the early stopping method to prevent overfitting; Step 4: Deploy the multi-scale time series model to the production environment to predict network traffic in real time, and monitor the performance data of the multi-scale time series model and changes in network traffic in real time; Step 5: Evaluate and adjust the multi-scale time series model: Based on the monitored performance data, use the MAE and RMSE indicators to evaluate the prediction performance of the multi-scale time series model, use cross-validation to evaluate the generalization ability of the multi-scale time series model, and adjust the multi-scale time series model based on the evaluation results.

2. According to the method for dynamic optimization of network traffic based on time series analysis in claim 1, it is characterized by: In step 1, the collected data is preprocessed: Perform data cleaning: handle missing values, outliers and noisy data to ensure data quality. Perform data normalization to meet the input requirements of the LSTM model.

3. The method for dynamic optimization of network traffic based on time series analysis according to claim 1 is characterized by: In step 2, time series features are extracted from the collected data, including: Define different time scales based on the characteristics of network delay and business requirements. At each defined time scale, the corresponding features are extracted: Extract second-level features, including instantaneous value, rate of change, and peak value of delay. Extract minute-level features, including average delay, delay volatility, and cumulative delay. Extract hourly features, including latency trends, periodic changes, and peak and valley values ​​within the hour. Extract daily features, including the average daily delay, the difference in delay between weekdays and weekends, and the seasonal variation of delay.

4. A method for dynamic optimization of network traffic based on time series analysis according to claim 1 or 3, Its characteristic is that in step 2, the extracted time series features are transformed: logarithmic transformation, difference, and moving average transformation are performed on the time series features to enhance the model's ability to capture trends and periodicity; Perform feature fusion on the extracted time series features: fuse features of different scales to form the input of the model. The fusion methods include splicing or weighted fusion, where the weights are dynamically adjusted according to the importance of the features; Perform feature selection on the extracted time series features: Use correlation-based selection methods to remove redundant or unimportant features to improve the performance and interpretability of the model.

5. The method for dynamic optimization of network traffic based on time series analysis according to claim 1 is characterized in that In step 3, select the loss function for mean square error (MSE) or mean absolute error (MAE), and use the Adam optimizer to optimize the multi-scale time series model. The Adam optimizer is used to set the network delay parameters: set the initial learning rate, set β1 to 0.9, β2 to 0.999, and ε to a constant to ensure numerical stability.

6. A network traffic dynamic optimization device based on time series analysis, characterized in that It includes data collection module, multi-scale time series model building module, training and tuning module, model deployment and maintenance module, and evaluation and testing module. The data collection module collects data: collects historical network traffic data, and deploys network monitoring tools to collect real-time network traffic data. Network traffic data includes data packet size, transmission rate, packet loss rate, and delay data. Multi-scale time series model building module builds a multi-scale time series model: extracts time series features from the collected data, and creates a multi-scale LSTM model based on the long short-term memory network LSTM. The multi-scale LSTM model includes an input layer, a multi-scale LSTM layer, a fusion layer, and an output layer. The multi-scale LSTM layer includes short-scale LSTM, which is used to capture short-term network delay changes at the second or minute level; medium-scale LSTM, which is used to capture medium-term network delay changes at the hour or several hours level; and long-scale LSTM, which is used to capture long-term network delay trends across days or longer. Adding residual connections between multi-scale LSTM layers helps the model learn the delay variation pattern; Introducing the attention mechanism in the fusion layer to make the model focus on the time scale that has a greater impact on delay prediction; Training and tuning module Training and tuning of multi-scale time series models: using time series features to train multi-scale time series models, selecting loss functions for regression problems, using optimizers to optimize parameters of multi-scale time series models, applying L1 or L2 regularization algorithms to adjust multi-scale time series models and combining early stopping methods to prevent overfitting; The model deployment and maintenance module deploys the multi-scale time series model to the production environment, predicts network traffic in real time, and monitors the performance data of the multi-scale time series model and changes in network traffic in real time; The evaluation and testing module evaluates and adjusts the multi-scale time series model: Based on the monitored performance data, the MAE and RMSE indicators are used to evaluate the prediction performance of the multi-scale time series model, and cross-validation is used to evaluate the generalization ability of the multi-scale time series model. The multi-scale time series model is adjusted according to the evaluation results.

7. The network traffic dynamic optimization device based on time series analysis according to claim 6 is characterized by: The data collection module preprocesses the collected data: Perform data cleaning: handle missing values, outliers and noisy data to ensure data quality. Perform data normalization to meet the input requirements of the LSTM model.

8. The network traffic dynamic optimization device based on time series analysis according to claim 6 is characterized by: The multi-scale time series model building module extracts time series features from the collected data, including: Define different time scales based on the characteristics of network delay and business requirements. At each defined time scale, the corresponding features are extracted: Extract second-level features, including instantaneous value, rate of change, and peak value of delay. Extract minute-level features, including average delay, delay volatility, and cumulative delay. Extract hourly features, including latency trends, periodic changes, and peak and valley values ​​within the hour. Extract daily features, including the average daily delay, the difference in delay between weekdays and weekends, and the seasonal variation of delay.

9. A method for dynamic optimization of network traffic based on time series analysis according to claim 6 or 8, characterized in that The multi-scale time series model building module performs feature transformation on the extracted time series features: Perform logarithmic transformation, difference, and moving average transformation on time series characteristics to enhance the model's ability to capture trends and periodicity; Perform feature fusion on the extracted time series features: fuse features of different scales to form the input of the model. The fusion methods include splicing or weighted fusion, where the weights are dynamically adjusted according to the importance of the features; Perform feature selection on the extracted time series features: Use correlation-based selection methods to remove redundant or unimportant features to improve the performance and interpretability of the model.

10. The network traffic dynamic optimization device based on time series analysis according to claim 6 is characterized in that the training The tuning module selects the loss function for the mean square error (MSE) or the mean absolute error (MAE), and uses the Adam optimizer to optimize the multi-scale time series model. The Adam optimizer is used to set the network delay parameters: set the initial learning rate, set β1 to 0.9, β2 to 0.999, and ε to a constant to ensure numerical stability.

Images