Intelligent prediction and dynamic optimization distribution method for computer network traffic

Abstract

The invention provides an intelligent prediction and dynamic optimization distribution method for computer network traffic, which relates to the technical field of computer networks and comprises the following steps of: acquiring and preprocessing network traffic data; constructing a multi-model fusion network traffic prediction model; generating a dynamic optimization distribution strategy based on a prediction result; and allocation strategy execution and dynamic feedback adjustment: executing the allocation strategy through the SDN controller, monitoring the network state in real time and feeding back the network state. A multi-model fusion strategy is adopted, the time sequence capturing capability of the improved LSTM, the feature fitting capability of the XGBoost and the long dependence processing capability of the time sequence attention Transform are combined, and the weight is optimized through PSO, so that compared with a single model, the prediction error is reduced, the burst flow and periodic flow features can be accurately captured, and the problem that in the flow distribution link, the flow distribution efficiency is greatly improved is solved. Most methods only aim at maximizing the bandwidth utilization rate, and ignore the problem of QoS demand difference of different services.

Description

Technical Field

[0001] This invention relates to the field of computer network technology, and in particular to a method for intelligent prediction and dynamic optimization allocation of computer network traffic. Background Technology

[0002] With the rapid development of technologies such as 5G, IoT, and cloud computing, computer networks are carrying increasingly diverse types of services, and network traffic is showing characteristics such as explosive growth, strong randomness, and uneven spatial and temporal distribution.

[0003] Some existing network traffic prediction methods use a single time series model for traffic prediction. However, a single model is difficult to capture the complex nonlinear characteristics and burst traffic patterns in network traffic, resulting in low prediction accuracy. At the same time, in the traffic allocation stage, most methods only aim to maximize bandwidth utilization, ignoring the differences in QoS requirements of different services, and thus failing to achieve refined and differentiated allocation of network resources. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for intelligent prediction and dynamic optimization allocation of computer network traffic, thereby solving the problems mentioned in the background section.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to a method for intelligent prediction and dynamic optimization allocation of computer network traffic, specifically including the following steps: Step 1: Network traffic data collection and preprocessing. Deploy distributed traffic collection nodes to collect multi-dimensional network data, clean, standardize and encode the data, and build a standardized dataset. Step 2: Construct a multi-model fusion network traffic prediction model, extract key traffic features, and build three basic prediction models: an improved LSTM model, an XGBoost model, and a temporal attention Transformer model. The weights are obtained by optimizing the model using the particle swarm optimization algorithm, and the three basic models are fused using a weighted voting method to output the prediction results. Step 3: Generate a dynamic optimization allocation strategy based on the prediction results, evaluate the network status in real time, model the service QoS requirements, construct and solve a multi-objective optimization model to obtain the optimal traffic allocation scheme. Step four: Execution of allocation strategy and dynamic feedback adjustment. The allocation strategy is executed through the SDN controller, the network status is monitored in real time and feedback is provided, and the prediction model and allocation strategy are dynamically adjusted to achieve closed-loop optimization.

[0006] Furthermore, the data collection in step one requires the deployment of 10 distributed traffic collection nodes connected to the access switch. The network core access switch is installed inside the positioning enclosure and collects traffic data of the network core link and access link in real time based on the NetFlow and sFlow protocols. The collection dimensions include: timestamp, source IP address, destination IP address, source port, destination port, protocol type, packet size, number of packets, and link bandwidth utilization. At the same time, network topology information and service attribute information are also collected.

[0007] Furthermore, the data preprocessing in step one cleans the collected raw data, removes missing and outlier values ​​caused by collection anomalies, and standardizes the cleaned data, mapping continuous data such as traffic data and bandwidth utilization to the [0,1] interval; discrete data such as service type and protocol type are converted using one-hot encoding, and finally a standardized network traffic dataset is constructed.

[0008] Furthermore, in step two, key traffic features are extracted from the preprocessed dataset, including time features, statistical features, and business-related features. The random forest algorithm is used to rank the features by importance, and key features with a cumulative importance ratio of more than 85% are selected as model inputs.

[0009] Furthermore, the three basic prediction models in step two include: an improved LSTM model, which introduces an attention mechanism into the hidden layer of the traditional LSTM, and focuses on the impact of traffic features at different time steps on the prediction results through an attention weight matrix; The XGBoost model takes the selected key features as input, learns by ensemble learning through the construction of multiple CART regression trees, and minimizes the prediction error by using gradient descent. The temporal attention Transformer model adopts an encoder-decoder structure. The encoder captures long-term temporal dependencies in traffic data through a self-attention mechanism, and the decoder outputs traffic prediction values ​​for future periods. The prediction results of three basic models are fused using a weighted voting method. The weights are obtained by optimizing the particle swarm optimization algorithm. The objective function is to minimize the mean absolute error between the predicted and actual values.

[0010] Furthermore, in step three, the network status is collected in real time based on the SNMP protocol to collect the current status parameters of the network, including the current bandwidth utilization, transmission delay, packet loss rate, and link load balancing. Step 3, in the QoS requirement modeling of services, requires classifying services into three levels: high priority, medium priority, and low priority, based on service attribute information.

[0011] Furthermore, the multi-objective optimization model construction in step three aims to maximize network resource utilization, maximize QoS satisfaction of high-priority services, and maximize link load balancing. A dynamic optimization allocation model is constructed with constraints including link bandwidth upper limit constraints and service QoS threshold constraints. The optimization model in step three uses an improved genetic algorithm to solve the multi-objective optimization model. The traffic allocation scheme is represented by real number encoding, the fitness function is designed as the objective function F, an elite retention strategy is introduced to improve the convergence speed, and the optimal traffic allocation scheme is obtained through iterative solution, including the traffic allocation ratio of each link and the service forwarding path.

[0012] Furthermore, in step four, the SDN controller converts the optimal allocation strategy into flow table rules and distributes them to the network core switches to achieve dynamic traffic forwarding and bandwidth allocation. In step four, real-time monitoring and feedback involves the traffic acquisition node monitoring network traffic data, link status parameters, and service QoS indicators after the strategy is implemented, and calculating the deviation between the actual and predicted values ​​and the QoS indicator compliance rate. In step four, the model and allocation strategy are dynamically adjusted. If the deviation exceeds the preset threshold or the QoS compliance rate of high-priority services is less than 95%, the adjustment mechanism is triggered. On the one hand, the training dataset of the traffic prediction model is updated and the model is retrained to optimize the prediction accuracy. On the other hand, the network state evaluation matrix is ​​corrected based on real-time monitoring data, and the optimized model is re-solved to generate the adjusted allocation strategy, thus realizing the closed-loop optimization of prediction-allocation-monitoring-adjustment.

[0013] Furthermore, in step one, the bottom of the positioning cover has multiple evenly distributed ventilation slots, and a ventilation cover is installed at the bottom of the positioning cover; a fixed side slot is provided on one side of the positioning cover, and a positioning slot is provided on the other side of the interior of the positioning cover; a clamping plate is slidably installed inside the fixed side slot; the inner end of the clamping plate is located inside the positioning slot; a pull-out sealing plate is slidably installed on the top of the ventilation cover; the pull-out sealing plate is located at the bottom of the positioning cover, and multiple evenly distributed guide slots are provided on the top of the pull-out sealing plate; a drive cylinder is installed at one end of the bottom of the positioning cover, and the output end of the drive cylinder is connected to one end of the pull-out sealing plate; the guide slots can communicate with the ventilation slots.

[0014] Furthermore, a base is installed at the bottom of the ventilation hood; an air intake hood is installed on one side of the top of the base; an air pump is installed on the top of the air intake hood, which is installed on the side of one of the ventilation hoods; a flow guide hood is also installed on the top of the ventilation hood; the flow guide hood is located between two ventilation hoods; and a network core switch, an access switch, and an SDN controller are respectively installed inside the three positioning hoods.

[0015] This invention provides a method for intelligent prediction and dynamic optimization allocation of computer network traffic, which has the following beneficial effects: When in use, this invention employs a multi-model fusion strategy, combining the improved temporal capture capability of LSTM, the feature fitting capability of XGBoost, and the long dependency processing capability of temporal attention Transformer. By optimizing weights through PSO, the prediction error is reduced by 15%-20% compared to a single model, and it can accurately capture the characteristics of burst traffic and periodic traffic.

[0016] Furthermore, based on a multi-objective optimization model, resources are dynamically allocated, prioritizing the QoS requirements of high-priority services, improving network resource utilization by 20%-30%, and link load balancing by more than 30%, effectively avoiding link congestion.

[0017] Furthermore, by using an SDN controller to rapidly distribute policies and combining a closed-loop feedback mechanism to dynamically adjust models and allocation strategies, it is possible to quickly respond to changes in network traffic and business needs, thereby improving the stability and reliability of network services. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments will be briefly described below.

[0019] The accompanying drawings described below are only related to some embodiments of the invention and are not intended to limit the invention.

[0020] In the attached diagram: Figure 1 A flowchart of the computer network traffic optimization and allocation method of the present invention is shown; Figure 2 A flowchart of the network traffic data acquisition and preprocessing method of the present invention is shown; Figure 3 A flowchart of the multi-model fusion method of the present invention is shown; Figure 4 A flowchart of the dynamic optimization allocation strategy method of the present invention is shown; Figure 5 A flowchart of the dynamic feedback method of the present invention is shown; Figure 6 A three-dimensional structural diagram of the positioning cover of the present invention is shown; Figure 7 A cross-sectional view of the base structure of the present invention is shown; Figure 8 A schematic diagram of the three-dimensional structure of the clamping plate of the present invention is shown; Figure 9 A three-dimensional structural diagram of the pull-out sealing plate of the present invention is shown.

[0021] List of reference numerals 1. Positioning cover; 101. Ventilation slot; 102. Ventilation cover; 103. Fixed side slot; 104. Positioning slot; 105. Clamping plate; 106. Pull-out sealing plate; 107. Flow guide channel; 2. Base; 201. Air intake shroud; 202. Air deflector; 3. Network core switch; 4. Connect to the switch; 5. SDN controller. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the described embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please refer to Figures 1 to 9 : Example 1: This invention proposes a method for intelligent prediction and dynamic optimization allocation of computer network traffic, including the following steps: Step 1, network traffic data collection and preprocessing: Deploy distributed traffic collection nodes to collect multi-dimensional network data, clean, standardize, and encode the data to construct a standardized dataset; Step 2, constructing a multi-model fusion network traffic prediction model: Extract key traffic features, construct three basic prediction models: an improved LSTM model, an XGBoost model, and a temporal attention Transformer model, optimize the weights using a particle swarm optimization algorithm, and fuse the three basic models using a weighted voting method to output the prediction results; Step 3, generating a dynamic optimization allocation strategy based on the prediction results: Real-time evaluation of network status, modeling of service QoS requirements, construction and solution of a multi-objective optimization model to obtain the optimal traffic allocation scheme; Step 4, execution of the allocation strategy and dynamic feedback adjustment: Execute the allocation strategy through an SDN controller 5, monitor the network status in real time and provide feedback, dynamically adjust the prediction model and allocation strategy to achieve closed-loop optimization; Step 1, data acquisition, requires the deployment of 10 distributed traffic acquisition nodes connected to access switch 4. Access switch 4, installed inside positioning housing 1, collects real-time traffic data from the network core and access links based on NetFlow and sFlow protocols. Acquisition dimensions include: timestamp, source IP address, destination IP address, source port, destination port, protocol type, packet size, number of packets, and link bandwidth utilization. Simultaneously, network topology information and service attribute information are also collected. Step 1, data preprocessing, cleans the collected raw data, removing missing and outlier values ​​caused by acquisition anomalies. The cleaned data is then standardized, mapping continuous data such as traffic data and bandwidth utilization to the [0,1] interval. The formula is: x'=(x-x_min) / (x_max-x_min). Where x represents the original data, x' represents the standardized data, and x_min and x_max represent the minimum and maximum values ​​of the data in this dimension, respectively. Discrete data such as business type and protocol type are transformed using one-hot encoding to finally construct a standardized network traffic dataset. Step two involves extracting key traffic features from the preprocessed dataset, including time features, statistical features, and business-related features. A random forest algorithm is used to rank the features by importance, selecting key features with a cumulative importance percentage exceeding 85% as model input. The three basic prediction models in step two include: an improved LSTM model, which introduces an attention mechanism into the hidden layers of the traditional LSTM. An attention weight matrix focuses on the impact of traffic features at different time steps on the prediction results. The attention weight calculation formula is: α_t = softmax(W_a·tanh(W_h·h_t+b_h)+b_a). Where α_t is the attention weight at time step t, h_t is the output of the LSTM hidden layer, W_a and W_h are weight matrices, and b_a and b_h are bias terms; The XGBoost model uses selected key features as input, constructs multiple CART regression trees for ensemble learning, and employs gradient descent to minimize prediction error. The tree depth is set to 6-10, and the learning rate is 0.01-0.1. The temporal attention Transformer model uses an encoder-decoder structure. The encoder captures long-term temporal dependencies in traffic data through a self-attention mechanism, and the decoder outputs traffic predictions for future periods. A weighted voting method is used to fuse the predictions from the three base models. The weights are optimized using particle swarm optimization. The objective function is to minimize the mean absolute error between the predicted and actual values. The formula for calculating the fused prediction value Y is: Y = w1·Y1 + w2·Y2 + w3·Y3 Where Y1, Y2, and Y3 are the prediction results of the improved LSTM, XGBoost, and Temporal Attention Transformer, respectively, w1, w2, and w3 are the optimized model weights, and w1+w2+w3=1; In step three, the network status is collected in real time based on the SNMP protocol, including current bandwidth utilization, transmission delay, packet loss rate, and link load balancing, to construct a network status evaluation matrix S: S=[s_ij]_(m×n) Where m is the number of links, n is the dimension of state parameters, and s_ij is the standardized value of the j-th state parameter of the i-th link; Step three, the service QoS requirement modeling, requires classifying services into three levels based on service attribute information: high priority (e.g., industrial control, real-time communication), medium priority (e.g., high-definition video, online office), and low priority (e.g., file download, web browsing). QoS constraint thresholds are set for each level: high priority services have latency ≤50ms and packet loss rate ≤0.1%; medium priority services have latency ≤100ms and packet loss rate ≤0.5%; and low priority services have latency ≤500ms and packet loss rate ≤1%. Step three also involves constructing a multi-objective optimization model with the objectives of maximizing network resource utilization, maximizing high-priority service QoS satisfaction, and maximizing link load balancing. A dynamic optimization allocation model is constructed, with constraints including link bandwidth upper limit constraints and service QoS threshold constraints. The objective function is as follows: maxF = λ1·U + λ2·Q + λ3·B st0≤f_ik≤C_i, Wherein, the link bandwidth constraint is defined as follows: f_ik represents the traffic allocated to the k-th type of service on the i-th link, and C_i represents the maximum bandwidth of the i-th link. D_k≤D_k^th, L_k≤L_k^th, Wherein, QoS constraints are defined as follows: D_k and L_k represent the actual latency and packet loss rate of the k-th type of service, and D_k^th and L_k^th represent the corresponding thresholds. Where U is network resource utilization, Q is QoS satisfaction of high-priority services, B is link load balancing, and λ1, λ2, and λ3 are weighting coefficients, with λ2>λ1>λ3. The optimization model in step three uses an improved genetic algorithm to solve the multi-objective optimization model. The traffic allocation scheme is represented by real number encoding, the fitness function is designed as the objective function F, an elite retention strategy is introduced to improve the convergence speed, and the optimal traffic allocation scheme is obtained through iterative solution, including the traffic allocation ratio of each link and the service forwarding path. In step four, the SDN controller 5 converts the optimal allocation strategy into flow table rules and distributes them to the network core switch 3 to achieve dynamic traffic forwarding and bandwidth allocation. In step four, real-time monitoring and feedback involve traffic acquisition nodes monitoring network traffic data, link status parameters, and service QoS indicators after the strategy is executed, calculating the deviation between the actual and predicted values, and the QoS indicator compliance rate. In step four, the model and allocation strategy are dynamically adjusted. If the deviation exceeds a preset threshold or the QoS compliance rate of high-priority services is lower than 95%, an adjustment mechanism is triggered. On the one hand, the training dataset of the traffic prediction model is updated, and the model is retrained to optimize prediction accuracy. On the other hand, the network status evaluation matrix is ​​corrected based on real-time monitoring data, and the optimized model is re-solved to generate the adjusted allocation strategy, thus achieving closed-loop optimization of prediction-allocation-monitoring-adjustment.

[0024] In Example 2, based on Example 1, the positioning cover 1 in step 1 has a U-shaped structure, and the bottom of the positioning cover 1 has multiple evenly distributed ventilation slots 101, and a ventilation cover 102 is installed at the bottom of the positioning cover 1; a fixed side slot 103 is provided on one side of the positioning cover 1, and a positioning slot 104 is provided on the other side of the interior of the positioning cover 1; a clamping plate 105 is slidably installed inside the fixed side slot 103; the inner end of the clamping plate 105 is inside the positioning slot 104; a pull-out sealing plate 106 is slidably installed on the top of the ventilation cover 102; the pull-out sealing plate 106 is located at the bottom of the positioning cover 1, and the top of the pull-out sealing plate 106 has multiple evenly distributed guide slots 107; A drive cylinder is installed at the bottom of one end of the positioning cover 1, and the output end of the drive cylinder is connected to one end of the pull-out sealing plate 106; the guide groove 107 can be connected to the ventilation groove 101; a base 2 is installed at the bottom of the ventilation cover 102; an air intake cover 201 is installed on one side of the top of the base 2; an air pump is installed on the top of the air intake cover 201, and the air intake cover 201 is installed on the side of one of the ventilation covers 102; a guide cover 202 is also installed on the top of the ventilation cover 102; the guide cover 202 is located between the two ventilation covers 102; a network core switch 3, an access switch 4, and an SDN controller 5 are respectively installed inside the three positioning covers 1. When optimizing the allocation of computer network traffic, the positioning covers 1 Internally installed heat sensors are used to position the network core switch 3, access switch 4, and SDN controller 5 via three positioning covers 1. The inner ends of clamping plates 105 are inserted into the positioning slots 104 along the fixed side grooves 103, fixing the three sets of upper and lower clamping plates 105 to the upper and lower positions of the network core switch 3, access switch 4, and SDN controller 5 respectively. The base 2 fixes the positions of the ventilation hood 102, air intake hood 201, and air guide hood 202. After the heat sensors inside the positioning covers 1 detect the heat of the network core switch 3, access switch 4, and SDN controller 5, the airflow generated by the air pump passes through the air intake hood 201 and air guide hood 202. 2. Internal flow: The airflow moves within the ventilation hood 102. After the heat sensor detects heat, the drive cylinder at the bottom of one end of the positioning cover 1 at the corresponding position of the network core switch 3, access switch 4, and SDN controller 5 drives the pull-out sealing plate 106 to move. The pull-out sealing plate 106 slides on the top of the ventilation hood 102, connecting the guide groove 107 on the pull-out sealing plate 106 with the ventilation groove 101. This allows the cold airflow to flow upwards and cool down the network core switch 3, access switch 4, and SDN controller 5, reducing the high temperature generated by data processing and ensuring the efficient operation of the network core switch 3, access switch 4, and SDN controller 5.

[0025] The working principle of this embodiment is as follows: First, data acquisition and preprocessing are performed. Ten distributed traffic acquisition nodes are deployed and connected to access switch 4 to collect multi-dimensional network data, covering multiple indicators such as timestamps and source IP addresses, as well as network topology and service attribute information. After cleaning and removing outliers from the collected raw data, continuous data is standardized and mapped to the [0,1] interval, and discrete data is converted using one-hot encoding to construct a standardized dataset. Next, key traffic features are extracted. Time, statistical, and service-related features are extracted from the preprocessed dataset. The random forest algorithm is used to select key features with a cumulative importance ratio of over 85% as model input. Then, a multi-model fusion network traffic prediction model is constructed, including an improved LSTM model with an attention mechanism, an XGBoost model, and a temporal attention Transformer model. The weights are optimized using the particle swarm optimization algorithm, and the three basic models are fused using a weighted voting method to output the prediction results. A dynamic optimization allocation strategy is generated based on the prediction results. Network status parameters are collected in real time based on the SNMP protocol to construct an evaluation matrix. Service priorities are divided according to service attribute information, and QoS constraint thresholds are set, taking into account network resource utilization and QoS of high-priority services. A multi-objective optimization model is constructed with the goals of maximizing satisfaction and link load balancing. An improved genetic algorithm is used to solve for the optimal traffic allocation scheme. Finally, the optimal allocation strategy is converted into flow table rules by the SDN controller and distributed to the network core switch 3 to realize dynamic traffic forwarding and bandwidth allocation. The network traffic data, link status parameters and service QoS indicators are monitored in real time after the strategy is executed. If the deviation exceeds the preset threshold or the QoS compliance rate of high-priority services is lower than 95%, the adjustment mechanism is triggered. The traffic prediction model training dataset is updated and the model is retrained. The network status evaluation matrix is ​​corrected and the optimized model is resolved to generate the adjusted allocation strategy, thus realizing closed-loop optimization.

[0026] The following points should be noted in this article: 1. The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention; other structures can refer to general designs.

[0027] 2. Where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other to obtain new embodiments.

[0028] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent prediction and dynamic optimization allocation of computer network traffic, characterized in that, Includes the following steps: Step 1: Network traffic data collection and preprocessing. Deploy distributed traffic collection nodes to collect multi-dimensional network data, clean, standardize and encode the data, and build a standardized dataset. Step 2: Construct a multi-model fusion network traffic prediction model, extract key traffic features, and build three basic prediction models: an improved LSTM model, an XGBoost model, and a temporal attention Transformer model. The weights are obtained by optimizing the model using the particle swarm optimization algorithm, and the three basic models are fused using a weighted voting method to output the prediction results. Step 3: Generate a dynamic optimization allocation strategy based on the prediction results, evaluate the network status in real time, model the service QoS requirements, construct and solve a multi-objective optimization model to obtain the optimal traffic allocation scheme. Step 4: Execution of allocation strategy and dynamic feedback adjustment. The allocation strategy is executed through the SDN controller (5), the network status is monitored in real time and feedback is provided, and the prediction model and allocation strategy are dynamically adjusted to achieve closed-loop optimization.

2. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 1, characterized in that, In step one, data collection requires the deployment of 10 distributed traffic collection nodes connected to the access switch (4). The network core access switch (4) is installed inside the positioning cover (1). Based on the NetFlow and sFlow protocols, it collects traffic data of the network core link and access link in real time. The collection dimensions include: timestamp, source IP address, destination IP address, source port, destination port, protocol type, data packet size, number of data packets, and link bandwidth utilization. At the same time, it collects network topology information and service attribute information.

3. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 2, characterized in that, The data preprocessing in step one cleans the collected raw data, removes missing and outlier values ​​caused by collection anomalies, and standardizes the cleaned data, mapping continuous data such as traffic data and bandwidth utilization to the [0,1] interval; discrete data such as service type and protocol type are converted using one-hot encoding, and finally a standardized network traffic dataset is constructed.

4. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 3, characterized in that, Step two involves extracting key traffic features from the preprocessed dataset, including time features, statistical features, and business-related features. The random forest algorithm is used to rank the features by importance, and key features with a cumulative importance ratio of over 85% are selected as model inputs.

5. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 4, characterized in that, The three basic prediction models in step two include: an improved LSTM model, which introduces an attention mechanism into the hidden layer of the traditional LSTM, and focuses on the impact of traffic features at different time steps on the prediction results through an attention weight matrix; The XGBoost model takes the selected key features as input, learns by ensemble learning through the construction of multiple CART regression trees, and minimizes the prediction error by using gradient descent. The temporal attention Transformer model adopts an encoder-decoder structure. The encoder captures long-term temporal dependencies in traffic data through a self-attention mechanism, and the decoder outputs traffic prediction values ​​for future periods. The prediction results of three basic models are fused using a weighted voting method. The weights are obtained by optimizing the particle swarm optimization algorithm. The objective function is to minimize the mean absolute error between the predicted and actual values.

6. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 5, characterized in that, In step three, the network status is collected in real time based on the SNMP protocol to collect the current status parameters of the network, including the current bandwidth utilization, transmission delay, packet loss rate, and link load balancing. Step 3, in the QoS requirement modeling of services, requires classifying services into three levels: high priority, medium priority, and low priority, based on service attribute information.

7. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 6, characterized in that, The multi-objective optimization model construction in step three aims to maximize network resource utilization, maximize QoS satisfaction of high-priority services, and maximize link load balancing. A dynamic optimization allocation model is constructed with constraints including link bandwidth upper limit constraints and service QoS threshold constraints. The optimization model in step three uses an improved genetic algorithm to solve the multi-objective optimization model. The traffic allocation scheme is represented by real number encoding, the fitness function is designed as the objective function F, an elite retention strategy is introduced to improve the convergence speed, and the optimal traffic allocation scheme is obtained through iterative solution, including the traffic allocation ratio of each link and the service forwarding path.

8. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 7, characterized in that, In step four, the SDN controller (5) converts the optimal allocation strategy into flow table rules and sends them to the network core switch (3) to realize dynamic traffic forwarding and bandwidth allocation. In step four, real-time monitoring and feedback involves the traffic acquisition node monitoring network traffic data, link status parameters, and service QoS indicators after the strategy is implemented, and calculating the deviation between the actual and predicted values ​​and the QoS indicator compliance rate. In step four, the model and allocation strategy are dynamically adjusted. If the deviation exceeds the preset threshold or the QoS compliance rate of high-priority services is less than 95%, the adjustment mechanism is triggered. On the one hand, the training dataset of the traffic prediction model is updated and the model is retrained to optimize the prediction accuracy. On the other hand, the network state evaluation matrix is ​​corrected based on real-time monitoring data, and the optimized model is re-solved to generate the adjusted allocation strategy, thus realizing the closed-loop optimization of prediction-allocation-monitoring-adjustment.

9. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 8, characterized in that, In step one, the bottom of the positioning cover (1) has multiple evenly distributed ventilation slots (101), and the bottom of the positioning cover (1) is equipped with a ventilation cover (102); a fixed side slot (103) is provided on one side of the positioning cover (1), and a positioning slot (104) is provided on the other side of the interior of the positioning cover (1); a clamping plate (105) is slidably installed inside the fixed side slot (103); the inner end of the clamping plate (105) is inside the positioning slot (104); a pull-out sealing plate (106) is slidably installed on the top of the ventilation cover (102); the pull-out sealing plate (106) is located at the bottom of the positioning cover (1), and the top of the pull-out sealing plate (106) has multiple evenly distributed guide slots (107); a drive cylinder is installed at the bottom of one end of the positioning cover (1), and the output end of the drive cylinder is connected to one end of the pull-out sealing plate (106); the guide slot (107) can communicate with the ventilation slot (101).

10. The intelligent prediction and dynamic optimization allocation method for computer network traffic according to claim 9, characterized in that, The bottom of the ventilation hood (102) is equipped with a base (2); an air intake hood (201) is installed on one side of the top of the base (2); an air pump is installed on the top of the air intake hood (201), and the air intake hood (201) is installed on the side of a ventilation hood (102); a flow guide hood (202) is also installed on the top of the ventilation hood (102); the flow guide hood (202) is located between two ventilation hoods (102); a network core switch (3) and an access switch (4) and an SDN controller (5) are respectively installed inside the three positioning hoods (1).

Images