High-performance interface flow distribution method supporting intelligent scheduling

Abstract

The invention relates to the technical field of intelligent scheduling, in particular to a high-performance interface flow distribution method supporting intelligent scheduling. The method comprises the following steps: S1, analyzing the infrastructure condition of a current network, and determining the performance limit of network equipment according to historical fault records; s2, continuously monitoring and collecting indexes of a network interface, analyzing historical data, identifying a traffic mode and predicting a future trend of the traffic; s3, constructing a network flow model based on the collected index data, and simulating network behaviors and performance expressions under different distribution strategies; and S4, generating a flow distribution strategy by using an intelligent algorithm, performing a simulation test of the strategy in a simulation environment, and evaluating the effect of the strategy. According to the method, the flow path and the priority can be dynamically adjusted through a strategy for carrying out future trend prediction on the flow and generating an intelligent algorithm, it is ensured that key services and real-time applications obtain enough bandwidth resources, and therefore the system response speed and the overall service quality are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology, and more specifically, to a high-performance interface traffic allocation method that supports intelligent scheduling. Background Technology

[0002] With the widespread application of technologies such as cloud computing, big data, and the Internet of Things, network data traffic has experienced explosive growth, posing unprecedented challenges to data centers, network service providers, and the infrastructure of various internet applications. Traditional traffic allocation methods, often based on static rules or simple load balancing strategies, struggle to efficiently handle high-concurrency, high-traffic scenarios, easily leading to network congestion, increased response latency, and degraded service quality. Therefore, high-performance interface traffic allocation methods with intelligent scheduling have become a research and application hotspot. The core objective of this method is to dynamically and efficiently manage network data traffic through intelligent strategies and algorithms, ensuring optimal resource utilization and improving overall system performance and service quality. However, existing technologies suffer from long response times in interface traffic allocation, resulting in degraded service quality. Therefore, this paper proposes a high-performance interface traffic allocation method that supports intelligent scheduling. Summary of the Invention

[0003] The purpose of this invention is to provide a high-performance interface traffic allocation method that supports intelligent scheduling, so as to solve the problem of long response time in interface traffic allocation and resulting service quality degradation mentioned in the background art.

[0004] To achieve the above objectives, the present invention aims to provide a high-performance interface traffic allocation method supporting intelligent scheduling, comprising the following steps:

[0005] S1: Analyze the current network infrastructure status and determine the performance limits of network devices based on historical fault records;

[0006] S2: Continuously monitor and collect network interface metrics, analyze historical data, identify traffic patterns, and predict future traffic trends;

[0007] S3: Based on the collected indicator data, build a network traffic model to simulate network behavior and performance under different allocation strategies;

[0008] S4: Generate traffic allocation strategies using intelligent algorithms, conduct simulation tests of the strategies in a simulated environment, and evaluate their effectiveness.

[0009] As a further improvement to this technical solution, in S1, the infrastructure status includes the network topology, the bandwidth capacity of each link, historical traffic data, and known bottleneck locations.

[0010] As a further improvement to this technical solution, in S2, the collected network interface metrics include traffic data, link status, packet loss rate, and latency.

[0011] As a further improvement to this technical solution, the specific steps in S2 for analyzing historical data, identifying traffic patterns, and predicting future traffic trends are as follows:

[0012] S21: Identify the trend of traffic changes over time based on collected historical data;

[0013] S22: Use clustering algorithms to group traffic data and identify different types of traffic patterns;

[0014] S23: Based on the characteristics of traffic data and prediction needs, establish a prediction model and train and optimize the model;

[0015] S24: Apply the trained model to predict traffic flow over a future period and output the predicted traffic flow value.

[0016] As a further improvement to this technical solution, the calculation steps for identifying the trend of flow rate change over time in step S21 are as follows:

[0017] Y t =Y0·e kt

[0018] Among them, Y t Yt represents the flow rate at time t; Y0 represents the flow rate at the initial time; k represents the growth rate; and t represents time.

[0019] Determine the parameter values ​​for Y0 and k, and take the natural logarithm of both sides of the model:

[0020] ln(Y t )=ln(Y0)+kt

[0021] Let Z t =ln(Y t ),but:

[0022] Z t =ln(Y0)+kt

[0023] By minimizing the RSS, ln(Y0) and k are estimated, and the RSS is:

[0024]

[0025] The expression for minimizing RSS is:

[0026]

[0027] For ln(Y0) and k, take the partial derivatives and set them to zero:

[0028]

[0029]

[0030] Solving the two systems of equations will yield estimates of ln(Y0) and k.

[0031] As a further improvement to this technical solution, the expression for S22 is:

[0032]

[0033] Where d(x) i C j ) represents the traffic data point x i To cluster center C j Distance; x ik For traffic data point x i The value of c on the k-th traffic pattern feature; jk Let be the value of the j-th cluster center on the k-th flow pattern feature; n is the total number of flow pattern features;

[0034] Each data point is assigned to the category corresponding to the nearest cluster center:

[0035]

[0036] Where K is the number of cluster centers; j * To make x i The cluster index that minimizes the distance to each cluster center;

[0037] Update cluster centers:

[0038]

[0039] Among them, |C j |For clustering C j The number of data points in the middle.

[0040] As a further improvement to this technical solution, the expression for S23 is as follows:

[0041]

[0042] in, The predicted flow rate at time point t+h; φ represents the constant term predicted by the model; p is the order of the autoregressive component; φ i These are the coefficients of the autoregressive component; Let be the model's predicted flow rate at time point t+hi; q is the order of the moving average component; θ is the value of the flow rate predicted by the model at time point t+hi. j The coefficients of the moving average component; This is an estimate of the prediction error of the model at time point t+hi.

[0043] As a further improvement to this technical solution, in step S3, the expression for the network traffic model is:

[0044] h t =LSTM(h t-1 [X] 1t X 2t X 3t Y t ];Θ)

[0045] y t =OutputLayer(h t ;W)

[0046] Among them, h t For LSTM, all historical information at time step t; h t-1 All historical information from the previous time step; [X] 1t X 2t X 3t Y t Let X be the eigenvector observed at time t. 1t Represents traffic data, X 2t Represents link status, X 3t Represents the packet loss rate; Θ is the parameter set in the LSTM model; y t is the network performance metric predicted by the model at time t; W is the weight matrix of the output layer.

[0047] As a further improvement to this technical solution, the specific steps for generating the traffic allocation strategy using an intelligent algorithm in step S4 are as follows:

[0048] S41: Define network state, possible traffic allocation decisions, and maximize throughput while minimizing latency;

[0049] S42: Encode the traffic allocation strategy as a gene, generate a new generation of strategies through crossover and mutation operations, and select based on simulation test results to gradually evolve a better strategy;

[0050] S43: Combining rule engines and machine learning to generate strategies that are both theoretically guided and adaptive.

[0051] As a further improvement to this technical solution, the expression for S41 is as follows:

[0052]

[0053] Where Q(s, a) is the action value function; r is the immediate reward; γ is the discount factor; s′ is the new state reached after taking action a; and a′ is all possible actions in state s′.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] This high-performance interface traffic allocation method that supports intelligent scheduling uses strategies for predicting future traffic trends and generating intelligent algorithms to dynamically adjust traffic paths and priorities, ensuring that critical services and real-time applications receive sufficient bandwidth resources, thereby improving system response speed and overall service quality. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating the overall method of the present invention. Detailed Implementation

[0057] 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 embodiments of the present invention, and not all embodiments. Based on the 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.

[0058] Example

[0059] Please see Figure 1 As shown, a high-performance interface traffic allocation method supporting intelligent scheduling is provided, including the following steps:

[0060] S1: Analyze the current network infrastructure status and determine the performance limits of network devices based on historical fault records;

[0061] S2: Continuously monitor and collect network interface metrics, analyze historical data, identify traffic patterns, and predict future traffic trends;

[0062] S3: Based on the collected indicator data, build a network traffic model to simulate network behavior and performance under different allocation strategies;

[0063] S4: Generate traffic allocation strategies using intelligent algorithms, conduct simulation tests of the strategies in a simulated environment, and evaluate their effectiveness.

[0064] In S1, the infrastructure status includes the network topology, bandwidth capacity of each link, historical traffic data, and known bottleneck locations.

[0065] In S2, the metrics collected for the network interface include traffic data, link status, packet loss rate, and latency.

[0066] In S2, the specific steps for analyzing historical data, identifying traffic patterns, and predicting future traffic trends are as follows:

[0067] S21: Identify the trend of traffic changes over time based on collected historical data;

[0068] S22: Use clustering algorithms to group traffic data and identify different types of traffic patterns;

[0069] S23: Based on the characteristics of traffic data and prediction needs, establish a prediction model and train and optimize the model;

[0070] S24: Apply the trained model to predict traffic flow over a future period and output the predicted traffic flow value.

[0071] In S21, the calculation steps for identifying the trend of traffic changes over time are as follows:

[0072] Y t =Y0·e kt

[0073] Among them, Y t Yt represents the flow rate at time t; Y0 represents the flow rate at the initial time; k represents the growth rate; and t represents time.

[0074] Determine the parameter values ​​for Y0 and k, and take the natural logarithm of both sides of the model:

[0075] ln(Y t )=ln(Y0)+kt

[0076] Let Z t =ln(Y t ),but:

[0077] Z t =ln(Y0)+kt

[0078] By minimizing the RSS, ln(Y0) and k are estimated, and the RSS is:

[0079]

[0080] The expression for minimizing RSS is:

[0081]

[0082] For ln(Y0) and k, take the partial derivatives and set them to zero:

[0083]

[0084]

[0085] Solving the two systems of equations will yield estimates of ln(Y0) and k.

[0086] The expression for S22 is:

[0087]

[0088] Where d(x) i C j ) represents the traffic data point x i To cluster center C j Distance; x ik For traffic data point x i The value of c on the k-th traffic pattern feature; jk Let be the value of the j-th cluster center on the k-th flow pattern feature; n is the total number of flow pattern features;

[0089] Each data point is assigned to the category corresponding to the nearest cluster center:

[0090]

[0091] Where K is the number of cluster centers; j * To make x i The cluster index that minimizes the distance to each cluster center;

[0092] Update cluster centers:

[0093]

[0094] Among them, |C j |For clustering C j The number of data points in the middle.

[0095] The expression for S23 is as follows:

[0096]

[0097] in, The predicted flow rate at time point t+h; φ represents the constant term predicted by the model; p is the order of the autoregressive component; φ i These are the coefficients of the autoregressive component; Let be the model's predicted flow rate at time point t+hi; q is the order of the moving average component; θ is the value of the flow rate predicted by the model at time point t+hi. j The coefficients of the moving average component; This is an estimate of the prediction error of the model at time point t+hi.

[0098] In S3, the expression for the network traffic model is:

[0099] h t =LSTM(h t-1 [X] 1t X 2t X 3t Y t ];Θ)

[0100] y t =OutputLayer(h t ;W)

[0101] Among them, h t For LSTM, all historical information at time step t; h t-1 All historical information from the previous time step; [X] 1t X 2t X 3t Y t Let X be the eigenvector observed at time t. 1t Represents traffic data, X 2t Represents link status, X 3t Represents the packet loss rate; Θ is the parameter set in the LSTM model; y t is the network performance metric predicted by the model at time t; W is the weight matrix of the output layer.

[0102] In S4, the specific steps for generating traffic allocation strategies using intelligent algorithms are as follows:

[0103] S41: Define network state, possible traffic allocation decisions, and maximize throughput while minimizing latency;

[0104] S42: Encode the traffic allocation strategy as a gene, generate a new generation of strategies through crossover and mutation operations, and select based on simulation test results to gradually evolve a better strategy;

[0105] S43: Combining rule engines and machine learning to generate strategies that are both theoretically guided and adaptive.

[0106] The expression for S41 is as follows:

[0107]

[0108] Where Q(s, a) is the action value function; r is the immediate reward; γ is the discount factor; s′ is the new state reached after taking action a; and a′ is all possible actions in state s′.

[0109] In this embodiment, S1: Analyze the current network infrastructure status and identify the performance limits of network devices based on historical fault records. Infrastructure status includes network topology, bandwidth capacity of each link, historical traffic data, and known bottleneck locations. By analyzing historical traffic data and known bottleneck locations, it is possible to identify which parts of the network are prone to becoming bottlenecks under high load, affecting data transmission speed and network service stability. This helps to optimize these aspects in a targeted manner, such as increasing bandwidth, upgrading hardware, or adjusting routing strategies. Combining historical fault records and current network usage allows for the development of more precise traffic management strategies. For example, adjusting traffic paths to avoid frequently congested areas, or pre-allocating resources when traffic peaks are predicted to prevent congestion.

[0110] S2: Continuously monitors and collects network interface metrics, analyzes historical data, identifies traffic patterns, and predicts future traffic trends. The collected network interface metrics include traffic data, link status, packet loss rate, and latency. By continuously monitoring these metrics, the working status and performance of the network interface can be understood in real time, problems can be detected and located in a timely manner, network faults can be responded to and resolved quickly, and service interruption time can be reduced.

[0111] In S2, the specific steps for analyzing historical data, identifying traffic patterns, and predicting future traffic trends are as follows:

[0112] S21: Identify the trend of traffic changes over time based on the collected historical data; the calculation steps for identifying the trend of traffic changes over time are as follows:

[0113] Y t =Y0·e kt

[0114] Among them, Y t Yt represents the flow rate at time t; Y0 represents the flow rate at the initial time; k represents the growth rate; and t represents time.

[0115] Determine the parameter values ​​for Y0 and k, and take the natural logarithm of both sides of the model:

[0116] ln(Y t )=ln(Y0)+kt

[0117] Let Z t =ln(Y t ),but:

[0118] Z t =ln(Y0)+kt

[0119] By minimizing the RSS, ln(Y0) and k are estimated, and the RSS is:

[0120]

[0121] The expression for minimizing RSS is:

[0122]

[0123] For ln(Y0) and k, take the partial derivatives and set them to zero:

[0124]

[0125]

[0126] Solving the two systems of equations will yield estimates of ln(Y0) and k.

[0127] By fitting an exponential growth model, the growth or decline trend of flow over time is analyzed. By estimating the model parameters ln(Y0) and k using logarithmic transformation and least squares method, the rate of change of flow over time and the initial flow level can be quantified, providing a foundation for understanding long-term flow behavior.

[0128] S22: Use clustering algorithms to group traffic data and identify different types of traffic patterns; the expression is:

[0129]

[0130] Where d(x) i C j ) represents the traffic data point x i To cluster center C j Distance; x ik For traffic data point x i The value of c on the k-th traffic pattern feature; jk Let be the value of the j-th cluster center on the k-th flow pattern feature; n is the total number of flow pattern features;

[0131] Each data point is assigned to the category corresponding to the nearest cluster center:

[0132]

[0133] Where K is the number of cluster centers; j * To make x i The cluster index that minimizes the distance to each cluster center;

[0134] Update cluster centers:

[0135]

[0136] Among them, |C j |For clustering C j The number of data points in the middle.

[0137] By using Euclidean distance and clustering algorithms, traffic data is divided into different groups or patterns. This allows for the identification of various different traffic behaviors or patterns within the network, such as traffic differences between weekdays and weekends, or day and night. The clustering results can be used to gain a more detailed understanding and predict traffic characteristics under specific time periods or conditions.

[0138] S23: Based on the characteristics of the traffic data and the prediction requirements, establish a prediction model and train and optimize the model; the expression is as follows:

[0139]

[0140] in, The symbol ^ represents the estimated value for the predicted flow rate at time point t+h, i.e., the future flow rate level that we want to predict. φ is the constant term in the model's predictions, representing the long-term average level in time series analysis without the influence of seasonality, trend, or other independent variables; p is the order of the autoregressive component, indicating how many past time points' flow values ​​are considered in the model to predict the current value. This affects how the model utilizes autocorrelation in historical data; i The coefficients of the autoregressive component are i = 1, 2, ..., p, which measure the influence of the flow values ​​at the first i time points on the future flow values. θ represents the model's predicted flow rate at time point t+hi; q is the order of the moving average component, indicating how many past error terms (prediction errors) are considered in the model to predict the current value; j The coefficients of the moving average component are denoted as j = 1, 2, ..., q. These coefficients determine the degree to which the past j error terms contribute to the future prediction error. This is an estimate of the prediction error of the model at time point t+hi.

[0141] S24: Apply the trained model to predict traffic flow over a future period and output the predicted traffic flow value.

[0142] S3: Construct a network traffic model based on the collected indicator data to simulate network behavior and performance under different allocation strategies; the expression for the network traffic model is:

[0143] h t =LSTM(h t-1 [X] 1t X 2t X 3t Y t ];Θ)

[0144] y t=OutputLayer(h t ;W)

[0145] LSTM is a special type of recurrent neural network architecture, particularly adept at handling long-term dependencies in sequential data. In this context, it is used to capture complex patterns in network traffic evolution over time; h t This represents all historical information of the LSTM at time step t, also known as the hidden state at time step t. It indicates the internal state of the LSTM at time step t after integrating the input features of the current time step and information from all previous time steps. It contains key information about traffic behavior up to the current time point and is the foundation for model memory and prediction; h t-1 This contains all historical information from the previous time step, reflecting past network behavior, and is passed to the current time step to maintain the continuity of the time series; [X] 1t X 2t X 3t Y t Let X be the eigenvector observed at time t. 1t Represents traffic data, X 2t Represents link status, X 3t Θ represents the packet loss rate, reflecting the network's current state and historical performance; Θ is the set of parameters in the LSTM model, including the weights and biases of the input gate, forget gate, cell state update, and output gate. These parameters are adjusted during training to optimize the model's fit to data and predictive ability; the output layer is responsible for storing the hidden states h of the LSTM. t Converted into specific network performance metric prediction values ​​y t ;y t The network performance metric predicted by the model at time t is determined by the output layer function based on the hidden state h. t The calculation shows that W is the weight matrix of the output layer, which determines how the hidden state is mapped to the predicted output. By adjusting these weights, the model can learn the optimal mapping from historical traffic behavior and the current network state to the prediction performance metrics.

[0146] This formula integrates historical traffic data and current network status using an LSTM model to simulate the dynamic behavior of the network under different traffic allocation strategies, providing data-driven decision support for network management and optimization strategy formulation. Through continuous iterative learning, the model can gradually optimize its predictive capabilities, helping network administrators to anticipate and address potential performance bottlenecks and resource demands, thereby improving the overall performance and quality of service of the network.

[0147] S4: Generate a traffic allocation strategy using intelligent algorithms, conduct simulation tests on the strategy in a simulated environment, and evaluate its effectiveness. The specific steps for generating a traffic allocation strategy using intelligent algorithms are as follows:

[0148] S41: Defines the network state, possible traffic allocation decisions, and the process of maximizing throughput while minimizing latency; the expression is as follows:

[0149]

[0150] Where Q(s, a) is the action value function, representing the expected long-term benefit of taking action a in the current network state s. It is the main objective that reinforcement learning algorithms attempt to optimize, continuously updating through learning to find the optimal policy. r is the immediate reward, the feedback obtained immediately after taking action a, reflecting the direct impact of the decision on the current network performance, such as increasing throughput or reducing latency. In traffic allocation scenarios, the design of the reward function needs to comprehensively consider factors such as throughput, latency, and stability to reflect the optimization objective. γ is the discount factor, a value between 0 and 1, used to balance the importance of immediate rewards and future potential rewards. The smaller γ is, the more the algorithm focuses on immediate benefits; γ closer to 1, the more it emphasizes long-term benefits. In network management, a larger γ value helps ensure that the policy not only solves the current problem but also considers the positive impact on the future state of the network. s′ is the new state reached after taking action a; a′ is all possible actions in state s′; max a Q(s′, a′) represents the maximum expected reward among all possible next actions a′ that can be obtained in the new state s′ reached after taking action a. This term reflects the "foresight" of the algorithm, encouraging the selection of actions that can lead to more favorable subsequent states, even if the direct reward of the current action may be small.

[0151] This formula iteratively updates the Q-value, guiding traffic allocation decisions from initial random attempts to converge towards the optimal strategy. This method can automatically adapt and learn in complex network environments without explicitly programming strategies for every possible scenario, making it particularly suitable for dynamically changing problems that are difficult to define using traditional rules. Ultimately, this will improve network resource utilization efficiency, enhance user experience, and reduce operating costs.

[0152] S42: Encode the traffic allocation strategy as a gene, generate a new generation of strategies through crossover and mutation operations, and select based on simulation test results to gradually evolve a better strategy;

[0153] S43: Combining rule engines and machine learning to generate strategies that are both theoretically guided and adaptive.

[0154] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-performance interface traffic allocation method supporting intelligent scheduling, characterized in that: Includes the following steps: S1: Analyze the current network infrastructure status and determine the performance limits of network devices based on historical fault records; S2: Continuously monitor and collect network interface metrics, analyze historical data, identify traffic patterns, and predict future traffic trends; S3: Based on the collected indicator data, build a network traffic model to simulate network behavior and performance under different allocation strategies; S4: Generate traffic allocation strategies using intelligent algorithms, conduct simulation tests of the strategies in a simulated environment, and evaluate their effectiveness.

2. The high-performance interface traffic allocation method supporting intelligent scheduling according to claim 1, characterized in that: In S1, the infrastructure status includes the network topology, bandwidth capacity of each link, historical traffic data, and known bottleneck locations.

3. The high-performance interface traffic allocation method supporting intelligent scheduling according to claim 2, characterized in that: In S2, the metrics collected for the network interface include traffic data, link status, packet loss rate, and latency.

4. The high-performance interface traffic allocation method supporting intelligent scheduling according to claim 3, characterized in that: In S2, the specific steps for analyzing historical data, identifying traffic patterns, and predicting future traffic trends are as follows: S21: Identify the trend of traffic changes over time based on collected historical data; S22: Use clustering algorithms to group traffic data and identify different types of traffic patterns; S23: Based on the characteristics of traffic data and prediction needs, establish a prediction model and train and optimize the model; S24: Apply the trained model to predict traffic flow over a future period and output the predicted traffic flow value.

5. A high-performance interface traffic allocation method supporting intelligent scheduling according to claim 4, characterized in that: In S21, the calculation steps for identifying the trend of traffic changes over time are as follows: Y t =Y0·e kt Among them, Y t Yt represents the flow rate at time t; Y0 represents the flow rate at the initial time; k represents the growth rate; and t represents time. Determine the parameter values ​​for Y0 and k, and take the natural logarithm of both sides of the model: ln(Y t )=ln(Y0)+kt Let Z t =ln(Y t ),but: Z t =ln(Y0)+kt By minimizing the RSS, ln(Y0) and k are estimated, and the RSS is: The expression for minimizing RSS is: For ln(Y0) and k, take the partial derivatives and set them to zero: Solving the two systems of equations will yield estimates of ln(Y0) and k.

6. The high-performance interface traffic allocation method supporting intelligent scheduling according to claim 5, characterized in that: The expression for S22 is: Where d(x) i C j ) represents the traffic data point x i To cluster center C j Distance; x ik For traffic data point x i The value of c on the k-th traffic pattern feature; jk Let be the value of the j-th cluster center on the k-th flow pattern feature; n is the total number of flow pattern features; Each data point is assigned to the category corresponding to the nearest cluster center: Where K is the number of cluster centers; j * To make x i The cluster index that minimizes the distance to each cluster center; Update cluster centers: Among them, |C j |For clustering C j The number of data points in the middle.

7. A high-performance interface traffic allocation method supporting intelligent scheduling according to claim 6, characterized in that: The expression for S23 is as follows: in, The predicted flow rate at time point t+h; φ represents the constant term predicted by the model; p is the order of the autoregressive component; φ i These are the coefficients of the autoregressive component; Let be the model's predicted flow rate at time point t+hi; q is the order of the moving average component; θ is the value of the flow rate predicted by the model at time point t+hi. j The coefficients of the moving average component; This is an estimate of the prediction error of the model at time point t+hi.

8. A high-performance interface traffic allocation method supporting intelligent scheduling according to claim 7, characterized in that: In S3, the expression for the network traffic model is: h t =LSTM(h t-1 ,[X 1t ,X 2t ,X 3t ,Y t ];Θ) y t =OutputLayer(h t ;W) Among them, h t For LSTM, all historical information at time step t; h t-1 This contains all historical information from the previous time step; [X] 1t X 2t X 3t Y t Let X be the eigenvector observed at time t. 1t Represents traffic data, X 2t Represents link status, X 3t Represents the packet loss rate; Θ is the parameter set in the LSTM model; y t is the network performance metric predicted by the model at time t; W is the weight matrix of the output layer.

9. A high-performance interface traffic allocation method supporting intelligent scheduling according to claim 8, characterized in that: In S4, the specific steps for generating traffic allocation strategies using intelligent algorithms are as follows: S41: Define network state, possible traffic allocation decisions, and maximize throughput while minimizing latency; S42: Encode the traffic allocation strategy as a gene, generate a new generation of strategies through crossover and mutation operations, and select based on simulation test results to gradually evolve a better strategy; S43: Combining rule engines and machine learning to generate strategies that are both theoretically guided and adaptive.

10. A high-performance interface traffic allocation method supporting intelligent scheduling according to claim 9, characterized in that: The expression for S41 is as follows: Where Q(s, a) is the action value function; r is the immediate reward; γ is the discount factor; s′ is the new state reached after taking action a; and a′ is all possible actions in state s′.

Images