A switch network traffic control method based on artificial intelligence

Abstract

The present invention relates to a switch network flow control method based on artificial intelligence, which collects the operation status data of the switch time series in a large historical period, including network flow and network congestion status; uses the network flow data of the switch time series to build a recurrent neural network model; uses the network flow data as an explanatory variable and the network congestion status as a response variable to build a Bayesian network model; obtains the network flow data of the current time series of the switch to be monitored, and inputs it into the recurrent neural network model to predict the network flow data of the next time series; inputs the network flow data of the next time series into the Bayesian network model to predict the probability of network congestion at the next time, and controls the classification and convergence of network flow according to the probability of network congestion. The present invention solves the problem that the prior art is difficult to warn and handle network congestion accidents in advance, resulting in poor network flow control effect.

Description

Technical Field

[0001] The present invention belongs to the technical field of network flow control and relates to a switch network flow control method based on artificial intelligence. Background Art

[0002] In today's society, computer networks play a vital role in all walks of life. As one of the key devices in computer networks, switches play an important role in packet forwarding and flow control. However, with the continuous expansion of network scale and the popularization of Internet applications, network traffic has also shown an explosive growth trend. If excessive network traffic is not effectively controlled, it is easy to cause network congestion accidents or even failures of switches. This congestion phenomenon will lead to an increase in packet delays, losses, and transmission errors, affecting the performance and reliability of the network. Therefore, strengthening network flow control has become a necessary means to improve the working efficiency of switches.

[0003] Currently, most of the control of network traffic is achieved by monitoring the operating status of the switch (such as network traffic, port status and CPU utilization) to achieve real-time prediction of network congestion. However, network congestion accidents often occur instantly, and it is difficult to provide early warning and handle network congestion accidents by monitoring the current network traffic status, resulting in poor network traffic control effects. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a switch network traffic control method based on artificial intelligence.

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] The present application provides a switch network traffic control method based on artificial intelligence, comprising the following steps:

[0007] S1. Collecting the operation status data of the switch time series in a large-scale historical period, wherein the operation status data includes network traffic and network congestion status, wherein the network congestion status includes network congestion and no network congestion;

[0008] S2, construct a recurrent neural network model through the network traffic data of the switch time series;

[0009] S3, using the network traffic data of the switch time series as the explanatory variable and the network congestion status as the response variable, construct a Bayesian network model;

[0010] S4, obtaining the network traffic data of the current time series of the switch to be monitored;

[0011] S5, inputting the network traffic data of the current time series into a recurrent neural network model to predict the network traffic data of the next time series;

[0012] S6, inputting the network traffic data of the next time series into the Bayesian network model to predict the probability of network congestion occurring at the next time;

[0013] S7. When the probability of network congestion occurring at the next time is greater than the first threshold, control the switch to divert the network traffic; when the probability of network congestion occurring at the next time is less than the second threshold, control the switch to aggregate the network traffic.

[0014] Furthermore, in step S2, the recurrent neural network model includes the following construction steps:

[0015] S21. Data preparation: Prepare a network traffic dataset containing R time series data samples on the switch;

[0016] S22, determine training data: arbitrarily select a data subset containing K data samples in the network traffic data set, use the first M data samples in the data subset as the input sequence, and the last K-M data samples as the target sequence, and obtain R-K training data pairs in total;

[0017] S23, determining a model architecture: determining neurons of the model, wherein the neurons include an input layer, a recurrent layer, and an output layer, wherein each recurrent layer includes a hidden layer;

[0018] S24, forward propagation: For each input sequence of the input layer, the model performs forward propagation calculations and outputs the corresponding predicted values ​​through the recurrent layer processing;

[0019] S25, calculate loss: compare the predicted value of the model with the target sequence of the output layer, and calculate the loss through the loss function;

[0020] S26, back propagation: adjusting the model parameters according to the loss, wherein the model parameters are weights and biases;

[0021] S27, training completed: repeat steps S24-S26 until the loss is lower than the set loss threshold, and the model training is completed.

[0022] Furthermore, in step S23, the input layer is the input sequence, the output layer is the target sequence, the circulation layer is three layers, an activation function is set for each layer, and the model training is started using randomly initialized weights and biases.

[0023] Furthermore, in step S25, the loss function is configured as a mean square error function or a cross entropy loss function.

[0024] Furthermore, in step S3, the Bayesian network model includes the following construction steps:

[0025] S31, determining the model connection relationship: using the network traffic data of the switch time series as the parent node, the network congestion status as the child node, connecting the parent node and the child node to form a Bayesian network model structure;

[0026] S32, data parameterization processing: estimating parameters of network traffic data and network congestion conditions according to preset standards;

[0027] S33, constructing a conditional probability table: constructing a conditional probability table of the correlation between each network flow parameter and the network congestion state parameter according to the parameters obtained by the data parameterization processing.

[0028] Furthermore, in step S32, the parameters of the transmission performance data and the transmission quality data are estimated according to the preset standards, specifically dividing the network traffic data into three level parameters of "high", "medium" and "low" according to the preset network traffic data value range; and dividing the occurrence of network congestion and the absence of network congestion in the network congestion state into two logical parameters of "yes" and "no", respectively.

[0029] Further, in step S33, the parameters obtained by the data parameterization processing are used to construct a conditional probability table of the correlation between each network traffic parameter and the network congestion state parameter, specifically: for each variable parameter, given its parent node, the joint conditional probability of the variable at the parent node is calculated, and the formula is:

[0030]

[0031] Where P(U) is a set of nodes U = B1, B2, ..., B n The joint conditional probability of i ) is the node B in the Bayesian network i The parent node set of P(B i ) represents the target node B i The prior probability of .

[0032] Furthermore, in step S3, the Bayesian network model also uses residual analysis or model fit to evaluate the prediction performance of the model.

[0033] Beneficial effects of the present invention:

[0034] By collecting the operation status data of the switch time series in a large historical period, including network traffic and network congestion conditions; using the network traffic data of the switch time series to build a recurrent neural network model; using the network traffic data as the explanatory variable and the network congestion conditions as the response variable, building a Bayesian network model; obtaining the network traffic data of the current time series of the switch to be monitored; inputting the network traffic data of the current time series into the recurrent neural network model to predict the network traffic data of the next time series; inputting the network traffic data of the next time series into the Bayesian network model to predict the probability of network congestion at the next time, and controlling the classification and convergence of network traffic according to the probability of network congestion. The present invention solves the problem that the prior art is difficult to warn and handle network congestion accidents in advance, resulting in poor network traffic control effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0036] Figure 1 The present invention is a flowchart of a method for controlling network traffic of a switch based on artificial intelligence. DETAILED DESCRIPTION

[0037] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0038] See also Figure 1 , the present application provides a switch network traffic control method based on artificial intelligence, comprising the following steps:

[0039] S1. Collecting the operation status data of the switch time series in a large-scale historical period, wherein the operation status data includes network traffic and network congestion status, wherein the network congestion status includes network congestion and no network congestion;

[0040] In this embodiment, a network traffic control method capable of early warning of network congestion is constructed based on historical big data. In this process, professional network monitoring tools, such as SNMP (Simple Network Management Protocol) or NetFlow, can be used to obtain network traffic data in real time. These tools can provide traffic statistics for each time period, such as bandwidth utilization, number of data packets, and traffic distribution. The network congestion status is often based on the switch log: the switch usually generates an operation log to record various events and status information of the switch. By analyzing these logs, data on the network congestion status can be obtained, such as connection timeout, number of packets waiting in queue, error log, etc.

[0041] Collecting large-scale historical switch operation status data is a long-term and continuous task that requires full consideration of the accuracy and completeness of the data. By using appropriate tools and methods, combined with real-time monitoring and offline analysis, valuable data on network traffic and congestion can be obtained, providing effective support for network optimization and traffic control.

[0042] S2, construct a recurrent neural network model through the network traffic data of the switch time series;

[0043] In daily work, network traffic has certain time-varying patterns. When we have enough sample data, we can use the network traffic of the previous time series to predict the network traffic of the next time series. This means that we can predict the changing trend of future network traffic by analyzing the trends and patterns in historical data. By modeling and analyzing time series data, we can find the traffic patterns and periodicity of different time periods. For example, some networks may have peaks during working hours during the day, but relatively low at night or on weekends. These regular changes can serve as a basis for predicting future network traffic.

[0044] This embodiment obtains network traffic data of large-scale historical time series and uses a recurrent neural network model to quantify the time series variation of network traffic. Recurrent Neural Network (RNN) is a neural network structure commonly used in processing sequence data. Unlike traditional feedforward neural networks, recurrent neural networks introduce feedback connections in time, allowing the network to model and process sequence data.

[0045] The core idea of ​​​​the recurrent neural network is to use the output of the previous moment as the input of the current moment. This internal loop structure allows the network to maintain memory when processing sequence data. The basic unit of RNN is a module called a recurrent unit, which receives input and generates output at each time step, and stores it as an internal state (hidden state) as one of the inputs of the next time step. This internal state allows the network to capture important contextual information when processing long sequences. Recurrent neural networks have the following characteristics and advantages in processing sequence data:

[0046] 1) Processing variable length input sequences: RNN is able to process input sequences of varying lengths and does not require fixed length inputs. This makes it very useful in tasks such as natural language processing, speech recognition, and machine translation.

[0047] 2) Contextual information modeling: Through the cyclic structure and the transfer of internal states, RNN can capture the contextual information in the sequence data. This makes it perform well when processing continuous data and time series data.

[0048] 3) Parameter sharing: RNN uses the same parameters at different time steps, which reduces the complexity of the network and improves the efficiency of training and inference.

[0049] Furthermore, in step S2, the recurrent neural network model includes the following construction steps:

[0050] S21. Data preparation: Prepare a network traffic dataset containing R time series data samples on the switch;

[0051] S22, determine training data: arbitrarily select a data subset containing K data samples in the network traffic data set, use the first M data samples in the data subset as the input sequence, and the last K-M data samples as the target sequence, and obtain R-K training data pairs in total;

[0052] It should be noted that the values ​​of K and M can be determined in the optimal range by the AIC criterion to improve the accuracy of the model.

[0053] S23, determining a model architecture: determining neurons of the model, wherein the neurons include an input layer, a recurrent layer, and an output layer, wherein each recurrent layer includes a hidden layer;

[0054] In this embodiment, a recurrent neural network (RNN) is a neural network architecture specifically used to process sequence data. Its main feature is that a recurrent connection is introduced when processing sequence data, so that the network can maintain memory and consider context information.

[0055] The basic model architecture of RNN is as follows:

[0056] Input layer: Receives input of sequence data, such as text, audio, or time series data.

[0057] Recurrent layer: Modeling of sequence data is achieved through recurrent connections. Each time step receives the input of the current time step and the hidden state of the previous time step, and produces the output and the hidden state of the current time step.

[0058] Output layer: Depending on the requirements of the task, different output layer structures can be selected. For regression tasks, a linear layer can be used to output continuous values.

[0059] Recurrent Neural Networks (RNNs) are usually composed of multiple recurrent layers, each of which contains a hidden layer. At each time step, the hidden layer receives the input of the current time step and the hidden state of the previous time step as input, and uses an activation function to perform weighted summation and nonlinear transformation on these inputs to produce the hidden state of the current time step. The output of the hidden layer can be used for other calculations at the current time step, and will also serve as input for the next time step. Such recurrent connections and multiple hidden layers allow the network to model long-term dependencies in sequence data. Each hidden layer can be viewed as a somewhat abstract representation of the sequence data, and information can be transferred and passed between different hidden layers. This hierarchical structure helps the network learn and represent complex patterns in sequence data.

[0060] Furthermore, in step S23, the input layer is the input sequence, the output layer is the target sequence, the circulation layer is three layers, an activation function is set for each layer, and the model training is started using randomly initialized weights and biases.

[0061] S24, forward propagation: For each input sequence of the input layer, the model performs forward propagation calculations and outputs the corresponding predicted values ​​through the recurrent layer processing;

[0062] S25. Calculate loss: Compare the model's predicted value with the target sequence of the output layer and calculate the loss through the loss function.

[0063] Furthermore, in step S25, the loss function is configured as a mean square error function or a cross entropy loss function.

[0064] S26, back propagation: adjusting the model parameters according to the loss, wherein the model parameters are weights and biases;

[0065] S27, training completed: repeat steps S24-S26 until the loss is lower than the set loss threshold, and the model training is completed.

[0066] S3, using the network traffic data of the switch time series as the explanatory variable and the network congestion status as the response variable, construct a Bayesian network model;

[0067] In this embodiment, by using the Bayesian network model to quantify the relationship between the network traffic of the switch time series and the network congestion status, a deeper understanding of the important impact of network traffic changes on network congestion within a certain period of time can be obtained. A Bayesian network is a probabilistic graph model that can be used to represent the dependencies between variables and analyze these dependencies through probabilistic reasoning. In the relationship between network traffic and network congestion status, network traffic can be used as an input variable and network congestion status as an output variable, and the network congestion status can be predicted by analyzing the dependency between them.

[0068] Specifically, we can use historical data to train a Bayesian network model, which can learn the probability distribution between network traffic change patterns and network congestion. By learning these probability distributions, we can understand the probability of network congestion under different network traffic conditions, and thus quantify the relationship between them.

[0069] Once a well-trained Bayesian network model is obtained, it can be used to predict network congestion. By observing the changes in network traffic over a period of time, these observation data can be input into the Bayesian network model and probabilistic reasoning can be performed to obtain the probability distribution of network congestion conditions. These probability distributions can help us assess the risk of network congestion and take appropriate measures to prevent or resolve potential congestion problems.

[0070] Furthermore, in step S3, the Bayesian network model includes the following construction steps:

[0071] S31, determining the model connection relationship: using the network traffic data of the switch time series as the parent node, the network congestion status as the child node, connecting the parent node and the child node to form a Bayesian network model structure;

[0072] In the process of building a Bayesian network model, the explanatory variable is usually used as the parent node and the response variable as the child node to represent the influence relationship between the variables. In this embodiment, the network traffic data is used as the parent node and the network congestion status is used as the child node. The parent node and the child node are connected to form a Bayesian network model structure. In this structure, the network traffic data has a direct impact on the network congestion status.

[0073] S32, data parameterization processing: estimating parameters of network traffic data and network congestion conditions according to preset standards;

[0074] In this embodiment, the Bayesian network is a graphical model for modeling and inferring probability relationships. Parameterization is an important task in the Bayesian network, and its purpose is to estimate the parameters of the probability distribution in the network based on the observed data. These parameters are used to describe the conditional probabilities and marginal probabilities between variables. The values ​​of the test data samples obtained are high and low. In the process of parameterization, this embodiment uses preset standards to estimate the parameters of the transmission performance data and the transmission quality data:

[0075] Furthermore, in step S32, the parameters of the transmission performance data and the transmission quality data are estimated according to the preset standards, specifically dividing the network traffic data into three level parameters of "high", "medium" and "low" according to the preset network traffic data value range; and dividing the occurrence of network congestion and the absence of network congestion in the network congestion state into two logical parameters of "yes" and "no", respectively.

[0076] In this embodiment, the network traffic data is divided into three level parameters of "high", "medium" and "low" according to a preset numerical range, and the standard can be set according to the needs of different users; the network congestion status is divided into two logical parameters of "yes" and "no" which can be used to judge the probability of whether an event occurs. For example, for the network congestion status, "yes" indicates the probability of network congestion occurring, and "no" indicates the probability of no network congestion occurring.

[0077] S33, constructing a conditional probability table: constructing a conditional probability table of the correlation between each network flow parameter and the network congestion state parameter according to the parameters obtained by the data parameterization processing.

[0078] In this embodiment, the conditional probability table is a table used to represent the conditional probability relationship between variables, which shows the probability of the variables occurring under certain conditions. Assuming there are two discrete variables A and B, the conditional probability table will be presented in the form of a table, in which the rows represent the values ​​of variable A and the columns represent the values ​​of variable B. Each cell in the table represents the probability of a certain value of A occurring under the condition of a certain value of B. The purpose of the conditional probability table is to describe and measure the conditional probability relationship between variables, and it can be used to infer, predict and analyze the dependency between variables. By observing and analyzing the conditional probability table, the correlation between variables can be understood, and effective reasoning and decision-making can be made based on this information.

[0079] When it comes to conditional probabilities with multiple parent nodes, an extended form of the conditional probability table is often used, called a joint conditional probability table or joint probability table. A joint conditional probability table is a table used to describe the relationship between multiple variables. It takes into account all combinations of parent node values ​​and shows the probability of the child nodes given the values ​​of these parent nodes. In a joint conditional probability table, each column represents the value of a child node, and each combination of parent node values ​​corresponds to a row. Each cell represents the probability of a child node given a combination of parent node values. Joint conditional probability tables are very useful for modeling complex probabilistic relationships and making inferences. By observing and analyzing joint conditional probability tables, we can understand the dependencies between multiple variables and make effective probabilistic inferences and decisions.

[0080] Further, in step S33, the parameters obtained by the data parameterization processing are used to construct a conditional probability table of the correlation between each network traffic parameter and the network congestion state parameter, specifically: for each variable parameter, given its parent node, the joint conditional probability of the variable at the parent node is calculated, and the formula is:

[0081]

[0082] Where P(U) is a set of nodes U = B1, B2, ..., B n The joint conditional probability of i ) is the node B in the Bayesian network i The parent node set of P(B i ) represents the target node B i The prior probability of .

[0083] Furthermore, in step S3, the Bayesian network model also uses residual analysis or model fit to evaluate the prediction performance of the model.

[0084] In Bayesian network models, residual analysis and model fit assessment are often used to evaluate the predictive accuracy and fit of the model. These methods can help understand the performance and predictive ability of the model on the training data.

[0085] Residual analysis evaluates the accuracy of the model by calculating the difference between the actual observations and the model predictions. For Bayesian network models, the residuals (actual observations minus model predictions) for each observed variable can be calculated, and then the distribution and pattern of the residuals can be analyzed. If the residuals meet certain assumptions (for example, they follow a normal distribution) and have no obvious structure or trend, this may indicate that the model has a high predictive accuracy. On the contrary, if there are systematic patterns, trends, or outliers in the residuals, further improvements to the model or correction of the data may be needed.

[0086] Model fit evaluation is to evaluate the fit of the model based on the performance of the model on the training data. Common indicators include the log-likelihood function value (Log-Likelihood) and the Bayesian Information Criterion (BIC). The log-likelihood function value measures the fit of the model on the observed data. The higher the value, the better the fit. BIC is an indicator that comprehensively considers the complexity and fit of the model. A lower BIC value indicates that the model better balances complexity and fit.

[0087] In addition to residual analysis and model fit assessment, there are many other methods that can be used to evaluate the predictive accuracy of Bayesian network models, such as cross-validation and leave-one-out method.

[0088] S4, obtaining the network traffic data of the current time series of the switch to be monitored;

[0089] S5, inputting the network traffic data of the current time series into a recurrent neural network model to predict the network traffic data of the next time series;

[0090] S6, inputting the network traffic data of the next time series into the Bayesian network model to predict the probability of network congestion occurring at the next time;

[0091] S7. When the probability of network congestion occurring at the next time is greater than the first threshold, control the switch to divert the network traffic; when the probability of network congestion occurring at the next time is less than the second threshold, control the switch to aggregate the network traffic.

[0092] In this embodiment, when the Bayesian network model is used to predict the probability of network congestion occurring at the next time, the switch can be controlled to divert or aggregate network traffic according to the following strategy:

[0093] 1) When the probability of network congestion occurring at the next time is greater than a first threshold (e.g., 80%), the switch can be triggered to perform traffic diversion. Specifically, the network congestion prediction result can be used as a reference, and the switch routing strategy can be adjusted according to the prediction result to direct part of the traffic to an alternative path, solution or other feasible solution.

[0094] 2) When the probability of network congestion in the next time is less than the second threshold (for example, 20%), the switch can be triggered to perform traffic aggregation. This means that the current network status is good and congestion is unlikely to occur. The traffic can be concentrated on the trunk path to obtain higher bandwidth utilization.

[0095] It should be noted that it is important to determine the appropriate threshold size, which depends on the specific network environment, business requirements, and tolerance for network congestion. These thresholds can be adjusted and optimized according to the actual scenario. In addition, in actual applications, factors such as real-time and stability may also be considered. The sliding window method can be used to continuously update and predict the Bayesian network model, and generate corresponding control signals for the switch based on the latest prediction results. This can better cope with dynamic changes in the network and situations with high real-time requirements. In summary, by using the Bayesian network model to predict the probability of network congestion and setting thresholds to trigger the diversion or aggregation operations of the switch, dynamic scheduling of network traffic can be achieved to improve network performance and avoid congestion.

[0096] Beneficial effects of the present invention:

[0097] By collecting the operation status data of the switch time series in a large historical period, including network traffic and network congestion conditions; using the network traffic data of the switch time series to build a recurrent neural network model; using the network traffic data as the explanatory variable and the network congestion conditions as the response variable, building a Bayesian network model; obtaining the network traffic data of the current time series of the switch to be monitored; inputting the network traffic data of the current time series into the recurrent neural network model to predict the network traffic data of the next time series; inputting the network traffic data of the next time series into the Bayesian network model to predict the probability of network congestion at the next time, and controlling the classification and convergence of network traffic according to the probability of network congestion. The present invention solves the problem that the prior art is difficult to warn and handle network congestion accidents in advance, resulting in poor network traffic control effect.

[0098] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A switch network traffic control method based on artificial intelligence, characterized in that: The following steps are involved: S1. Collecting the operation status data of the switch time series in a large-scale historical period, wherein the operation status data includes network traffic and network congestion status, wherein the network congestion status includes network congestion and no network congestion; S2, construct a recurrent neural network model through the network traffic data of the switch time series; S3, using the network traffic data of the switch time series as the explanatory variable and the network congestion status as the response variable, construct a Bayesian network model; S4, obtaining the network traffic data of the current time series of the switch to be monitored; S5, inputting the network traffic data of the current time series into a recurrent neural network model to predict the network traffic data of the next time series; S6, inputting the network traffic data of the next time series into the Bayesian network model to predict the probability of network congestion occurring at the next time; S7. When the probability of network congestion occurring at the next time is greater than the first threshold, control the switch to divert the network traffic; when the probability of network congestion occurring at the next time is less than the second threshold, control the switch to aggregate the network traffic; In step S2, the recurrent neural network model includes the following construction steps: S21. Data preparation: Prepare a network traffic dataset containing R time series data samples on the switch; S22. Determine training data: arbitrarily select a data subset containing K data samples in the network traffic data set, use the first M data samples in the data subset as the input sequence, and the last K-M data samples as the target sequence, and obtain a total of R-K training data pairs; the optimal value range of K and M is determined by the AIC criterion; S23, determining a model architecture: determining neurons of the model, wherein the neurons include an input layer, a recurrent layer, and an output layer, wherein each recurrent layer includes a hidden layer; S24, forward propagation: For each input sequence of the input layer, the model performs forward propagation calculations and outputs the corresponding predicted values ​​through the recurrent layer processing; S25, calculate loss: compare the predicted value of the model with the target sequence of the output layer, and calculate the loss through the loss function; S26, back propagation: adjusting the model parameters according to the loss, wherein the model parameters are weights and biases; S27, training completed: repeat steps S24-S26 until the loss is lower than the set loss threshold, and the model training is completed; In step S3, the Bayesian network model includes the following construction steps: S31, determining the model connection relationship: using the network traffic data of the switch time series as the parent node, the network congestion status as the child node, connecting the parent node and the child node to form a Bayesian network model structure; S32, data parameterization processing: estimating parameters of network traffic data and network congestion conditions according to preset standards; S33, constructing a conditional probability table: constructing a conditional probability table of the correlation between each network flow parameter and the network congestion state parameter according to the parameters obtained by the data parameterization processing.

2. The method for controlling switch network traffic based on artificial intelligence according to claim 1, characterized in that: In step S23, the input layer is the input sequence, the output layer is the target sequence, the circulation layer is three layers, an activation function is set for each layer, and the model training is started using randomly initialized weights and biases.

3. The method for controlling network traffic of a switch based on artificial intelligence according to claim 1, characterized in that: In step S25, the loss function is configured as a mean square error function or a cross entropy loss function.

4. The method for controlling switch network traffic based on artificial intelligence according to claim 1, characterized in that: In step S32, the parameters of the transmission performance data and the transmission quality data are estimated according to the preset standards, specifically, the network traffic data is divided into three level parameters of "high", "medium" and "low" according to the preset network traffic data value range; the occurrence of network congestion and the absence of network congestion in the network congestion state are divided into two logical parameters of "yes" and "no", respectively.

5. The method for controlling switch network traffic based on artificial intelligence according to claim 1, characterized in that: In step S33, the parameters obtained by the data parameterization process are used to construct a conditional probability table of the correlation between each network traffic parameter and the network congestion state parameter. Specifically, for each variable parameter, given its parent node, the joint conditional probability of the parameter at the parent node is calculated. The formula is: ， ， In the formula, P ( U ) is represented as a set of nodes U = B 1 , B 2 ,……, B n The joint conditional probability of Pa ( B i ) is a node in the Bayesian network B i The parent node set of P ( B i ) represents the target node B i The prior probability of .

6. The method for controlling switch network traffic based on artificial intelligence according to claim 1, characterized in that: In step S3, the Bayesian network model also uses residual analysis or model fit to evaluate the prediction performance of the model.

Images