A network traffic optimization method under optical network convergence environment

By deploying intelligent computing nodes and machine learning models in optoelectronic hybrid routers and dynamically adjusting network resources, the problem that traditional network management methods cannot respond quickly to traffic changes is solved, network resource utilization and performance are improved, and user experience is improved.

CN118890571BActive Publication Date: 2024-11-26CHENGDU SANWANGXUN TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411387984.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-08
Publication Date
2024-11-26
Estimated Expiration
2044-10-08

AI Technical Summary

Technical Problem

Traditional network management methods cannot respond quickly to traffic changes, resulting in low resource utilization, which may cause network congestion and service quality decline.

Method used

Deploy multiple intelligent computing nodes in optoelectronic hybrid routers to collect traffic data and network status data in real time, use machine learning models to predict traffic, and dynamically adjust router configuration, wavelength and bandwidth to respond to traffic changes.

Benefits of technology

It improves the utilization rate of network resources, alleviates network congestion, reduces latency, improves network performance, and enhances the computing and processing capabilities of the system, optimizes resource allocation and user experience.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118890571B_ABST
    Figure CN118890571B_ABST
Patent Text Reader

Abstract

The present application belongs to the technical field of network optimization, and more specifically, relates to a network traffic optimization method in an optical network convergence environment; the present invention arranges multiple intelligent computing nodes to collect local traffic data and network status data; and then uses a machine learning algorithm to predict future traffic changes, so that network resources can be dynamically adjusted according to actual needs, avoiding idleness or overload of resources, thereby greatly improving the utilization rate of network resources; and dynamically adjusting the configuration, wavelength and bandwidth of routers through prediction results can effectively alleviate network congestion, reduce delays, and improve the overall performance of the network; and the present invention deploys multiple intelligent computing nodes under an optoelectronic hybrid router, which not only improves the computing and processing capabilities of the system, but also enhances reliability, flexibility and scalability, optimizes resource allocation and utilization, and improves user experience.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the technical field of network optimization, and more specifically, to a network traffic optimization method in an optical network convergence environment. Background Art

[0002] With the rapid development of the Internet and communication technology, network traffic has shown an explosive growth trend; traditional circuit switching networks have gradually been unable to meet the growing bandwidth demand and low latency requirements; optical networks have become the mainstream of modern communication networks due to their advantages of high bandwidth, low loss and long-distance transmission; however, pure optical networks also face some challenges, such as insufficient flexibility and complex dynamic resource management; therefore, optoelectronic hybrid networks came into being, which combine the large capacity of optical networks and the flexible control capabilities of electrical networks, and can better meet the needs of modern communication networks.

[0003] In the optoelectronic hybrid network environment, traffic prediction and optimal resource allocation become the key to ensuring network performance; traditional network management methods mainly rely on static configuration and predefined strategies, which cannot respond quickly to traffic changes, resulting in low resource utilization and may even cause technical problems such as network congestion and reduced service quality. Summary of the invention

[0004] The present invention provides a network traffic optimization method in an optical network convergence environment, which is intended to solve the current technical problem of being unable to quickly respond to traffic changes, resulting in low resource utilization, and even possibly causing network congestion and reduced service quality.

[0005] A method for optimizing network traffic in an optical network convergence environment comprises the following steps:

[0006] Step 1: Deploy multiple intelligent computing nodes under an optoelectronic hybrid router; and collect local traffic data and network status data in real time through the data acquisition module;

[0007] Step 2: Use the machine learning model in the intelligent computing node to perform traffic forecasting based on the local traffic data and network status data collected in real time to predict future traffic changes;

[0008] Step 3: Based on the prediction results of the intelligent computing node, the collaborative control platform dynamically adjusts the configuration of the optoelectronic hybrid router;

[0009] Step 4: In the optical network, wavelength and bandwidth are dynamically allocated based on the predicted traffic demand;

[0010] Step 5: Monitor the network status and traffic conditions in real time. When there is a deviation between the actual traffic and the predicted results, feedback is given to the corresponding intelligent computing node, and the routing configuration and resource allocation are adjusted through the control platform.

[0011] The present invention arranges multiple intelligent computing nodes to collect local traffic data and network status data; then uses machine learning algorithms to predict future traffic changes, so that network resources can be dynamically adjusted according to actual needs, avoiding idle or overloaded resources, thereby greatly improving the utilization rate of network resources; and dynamically adjusts the configuration, wavelength and bandwidth of the router through the prediction results, which can effectively alleviate network congestion, reduce delays, and improve the overall performance of the network; and the present invention deploys multiple intelligent computing nodes under an optoelectronic hybrid router, which not only improves the computing and processing capabilities of the system, but also enhances reliability, flexibility and scalability, optimizes resource allocation and utilization, and improves user experience.

[0012] Preferably, in the step 1, the location of the intelligent computing node in deploying the intelligent computing node at the edge of the network is as follows:

[0013] Determine the K value: By calculating the total error sum of squares for different K values, plotting the relationship between the total error sum of squares and K value, select the K value at the inflection point where the total error sum of squares is significantly reduced as the optimal number of clusters, where the total error sum of squares is calculated as follows:

[0014] ;

[0015] Where: K represents the number of clusters; represents the i-th cluster; Representation Cluster A data point in represents the centroid of the i-th cluster; Represents the data point x to the centroid The square of the Euclidean distance;

[0016] Initialization: Initialize the centroids of K clusters;

[0017] Update the centroid: assign each data point to the cluster to which the nearest centroid belongs; recalculate the centroid of each cluster as the mean of all data points in the cluster; repeat the update of the centroid until the centroid no longer changes or the change is less than the preset threshold; the centroid update method is as follows:

[0018] ;

[0019] Where: represents the centroid of the i-th cluster; represents the number of data points in the i-th cluster; Representation Cluster A data point in

[0020] The determined optimal K value is used to perform K-means clustering to obtain each cluster and its centroid; for each cluster, the node with the most obvious traffic change characteristics is selected as the intelligent computing node.

[0021] Preferably, the machine learning model described in step 2 adopts an LSTM model; the training steps of the LSTM model are as follows:

[0022] Data collection: Obtain historical data of network traffic, which includes timestamps and traffic values;

[0023] Data cleaning: Process missing values ​​and outliers based on the acquired historical data to obtain a cleaned data set;

[0024] Time series feature extraction: Extract time series features based on cleaned data; add time features to the cleaned data set;

[0025] Periodic feature extraction: Based on the results of time series feature extraction, periodic feature extraction is performed to generate time series data with periodic features;

[0026] Time series analysis: Draw a flow time series graph and identify periodic components based on the flow change pattern in the flow time series graph;

[0027] Frequency domain analysis: Identify the main periodic components in the flow data based on Fourier transform and the identified periodic components to obtain periodic characteristics;

[0028] Data standardization: Split the data into training sets and test sets in chronological order, standardize the traffic data and periodic features in the training sets and test sets, and convert the data into a distribution with a mean of 0 and a standard deviation of 1;

[0029] Model training: The standardized traffic data and periodic features in the training set are used as the input of the LSTM model to train the LSTM model; the mean square error is used as the loss function, and the Adam optimizer is used for model training.

[0030] Preferably, adjusting the configuration of the optoelectronic hybrid router includes updating the routing table of the optoelectronic hybrid router and adjusting the queue management strategy;

[0031] The steps of updating the routing table of the optoelectronic hybrid router are as follows:

[0032] Obtain traffic forecast data from intelligent computing nodes, including traffic demand for a period of time in the future;

[0033] Smooth the predicted data to remove noise:

[0034] ;

[0035] Where: is the smoothed traffic forecast value; is the original traffic prediction value; n is the time window size;

[0036] Calculate the cost of each path based on the smoothed traffic forecast data:

[0037] ;

[0038] Where: is the delay of the path; and represents the weight coefficient; is the predicted flow from source node s to destination node d; is the bandwidth capacity of path p; represents the transmission cost from the source node s to the destination node d through the path p;

[0039] Select the path with the lowest cost and meeting the traffic demand to update the routing table;

[0040] The steps of adjusting the queue management strategy are as follows:

[0041] Assign queue weights based on smoothed traffic forecast data:

[0042] ;

[0043] Where: represents the weight of the i-th queue; is the flow demand of the i-th path; is the flow demand of the jth path;

[0044] The queue management policy is adjusted based on the calculated queue weight to ensure that paths with high traffic volume are given higher priority in the queue.

[0045] Preferably, step 4 comprises the following steps:

[0046] Step 4.1: Initialize the wavelength and bandwidth resources of all paths;

[0047] Step 4.2: Calculate the wavelength requirement of each source-destination pair based on the traffic prediction data;

[0048] Step 4.3: Allocate bandwidth according to wavelength requirements to ensure that the total bandwidth does not exceed the available bandwidth;

[0049] Step 4.4: Calculate the transmission cost of each path and select the path with the lowest cost for wavelength and bandwidth allocation.

[0050] Preferably, the wavelength requirement of each source-destination pair is obtained based on the following formula:

[0051] ;

[0052] Where: represents the number of wavelengths required to travel from source node s to destination node d in time t; represents the predicted value of the flow from the source node s to the destination node d at time t; Indicates the bandwidth capacity of each wavelength channel.

[0053] Preferably, the bandwidth allocation according to wavelength requirements is based on the following formula:

[0054] ;

[0055] Where: represents the bandwidth allocated from source node s to destination node d at time t; represents the total bandwidth available in the network at time t; represents the number of wavelengths required to travel from source node s to destination node d in time t; Indicates the bandwidth capacity of each wavelength channel.

[0056] Preferably, the transmission cost of each path is obtained based on the following formula:

[0057] ;

[0058] Where: is the delay of the path; and represents the weight coefficient; is the predicted flow from source node s to destination node d; is the bandwidth capacity of path p; Represents the transmission cost from source node s to destination node d through path p.

[0059] Preferably, step 5 comprises the following steps:

[0060] Compare the actual traffic data with the predicted traffic data of the intelligent computing node and calculate the prediction error:

[0061] ;

[0062] Where: represents the prediction error from source node s to destination node d at time t; represents the actual flow from source node s to destination node d at time t; represents the predicted flow from source node s to destination node d at time t;

[0063] Feedback to tune machine learning models:

[0064] ;

[0065] Where: represents the adjusted predicted flow from source node s to destination node d at time t; Represents the adjustment coefficient, which is used to control the adjustment range;

[0066] Steps 3 and 4 are performed based on the adjusted predicted traffic to obtain the adjusted configuration, wavelength, and bandwidth of the optoelectronic hybrid router.

[0067] The beneficial effects of the present invention include:

[0068] The present invention arranges multiple intelligent computing nodes to collect local traffic data and network status data; then uses machine learning algorithms to predict future traffic changes, so that network resources can be dynamically adjusted according to actual needs, avoiding idle or overloaded resources, thereby greatly improving the utilization rate of network resources; and dynamically adjusts the configuration, wavelength and bandwidth of the router through the prediction results, which can effectively alleviate network congestion, reduce delays, and improve the overall performance of the network; and the present invention deploys multiple intelligent computing nodes under an optoelectronic hybrid router, which not only improves the computing and processing capabilities of the system, but also enhances reliability, flexibility and scalability, optimizes resource allocation and utilization, and improves user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0070] Figure 1 An overall step block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] See also Figure 1 As shown, a network traffic optimization method in an optical network convergence environment includes the following steps:

[0073] Step 1: Deploy multiple intelligent computing nodes under an optoelectronic hybrid router; and collect local traffic data and network status data in real time through the data acquisition module;

[0074] The local traffic data includes:

[0075] Flow rate:

[0076] The total amount of data passing through the network in each time interval (such as the number of bits transmitted per second bps or the number of packets per second pps); record the changes in traffic rate at different time points.

[0077] Traffic type:

[0078] Application layer protocols (such as HTTP, FTP, VoIP, etc.) or service types (such as video streaming, file transfer, etc.); detailed packet analysis based on traffic classification.

[0079] Packet characteristics:

[0080] Packet size distribution; packet source and destination addresses; packet marking or priority (such as QoS tag).

[0081] Traffic pattern:

[0082] Traffic changes during peak and off-peak hours; Periodic traffic changes (such as differences between day and night).

[0083] The network status data includes:

[0084] Link status:

[0085] The bandwidth capacity of the link; link utilization (i.e. the proportion of bandwidth currently in use); link delay, jitter and bit error rate.

[0086] Node status:

[0087] CPU and memory usage of network devices such as routers and switches; port status of network devices (such as port utilization and port error rate);

[0088] Topology information:

[0089] The physical and logical topology of the network; the connection relationship and link status between nodes.

[0090] Load balancing information:

[0091] Current network load distribution; load balancing strategies and effects of each node and link.

[0092] Delay and packet loss rate:

[0093] The average delay and jitter of each node and link; the packet loss rate and retransmission situation.

[0094] The location of the intelligent computing node in the deployment of the intelligent computing node at the network edge in step 1 is as follows:

[0095] Determine the K value: By calculating the total error sum of squares for different K values, plotting the relationship between the total error sum of squares and K value, select the K value at the inflection point where the total error sum of squares is significantly reduced as the optimal number of clusters, where the total error sum of squares is calculated as follows:

[0096] ;

[0097] Where: K represents the number of clusters; represents the i-th cluster; Representation Cluster A data point in represents the centroid of the i-th cluster; Represents the data point x to the centroid The square of the Euclidean distance;

[0098] Initialization: Initialize the centroids of K clusters;

[0099] Update the centroid: assign each data point to the cluster to which the nearest centroid belongs; recalculate the centroid of each cluster as the mean of all data points in the cluster; repeat the update of the centroid until the centroid no longer changes or the change is less than the preset threshold; the centroid update method is as follows:

[0100] ;

[0101] Where: represents the centroid of the i-th cluster; represents the number of data points in the i-th cluster; Representation Cluster A data point in

[0102] The determined optimal K value is used to perform K-means clustering to obtain each cluster and its centroid; for each cluster, the node with the most obvious traffic change characteristics is selected as the intelligent computing node.

[0103] In this embodiment, by analyzing and aggregating data points with similar characteristics through a clustering algorithm, it is possible to accurately determine which nodes in the network have significant traffic change characteristics, thereby ensuring that the selected nodes can effectively represent the traffic changes in their clusters; and the clustering algorithm can reduce the number of nodes that require real-time calculation and prediction, thereby reducing the computing burden and resource consumption; secondly, the clustered intelligent computing nodes make predictions based on similar traffic patterns, and can provide more accurate traffic prediction results; because the traffic characteristics of these nodes are similar, the model can more accurately capture their changing trends.

[0104] Step 2: Use the machine learning model in the intelligent computing node to perform traffic forecasting based on the local traffic data and network status data collected in real time to predict future traffic changes;

[0105] The machine learning model described in step 2 adopts an LSTM model; the training steps of the LSTM model are as follows:

[0106] Data collection: Obtain historical data of network traffic, which includes timestamps and traffic values;

[0107] Data cleaning: Process missing values ​​and outliers based on the acquired historical data to obtain a cleaned data set;

[0108] Time series feature extraction: Extract time series features based on cleaned data; add time features to the cleaned data set;

[0109] The time series feature extraction includes: extracting hour features and extracting day of the week features; that is, extracting the hour number from the timestamp as a feature; extracting the day of the week from the timestamp as a feature; the cleaned data set with the time feature added includes the hour and day of the week features;

[0110] Periodic feature extraction: Based on the results of time series feature extraction, periodic feature extraction is performed to generate time series data of periodic features; exemplary steps are as follows:

[0111] The hour feature is converted into sine and cosine representation using the following formula:

[0112] ;

[0113] ;

[0114] Where: Represents the hourly characteristic value after sine transformation, and the converted value range is between -1 and 1; Represents the hourly characteristic value after cosine transformation, and the converted value range is between -1 and 1; Indicates the hour of the day, ranging from 0 to 23.

[0115] The following formula is used to convert the day of the week feature into sine and cosine representation:

[0116] ;

[0117] ;

[0118] Where: Represents the characteristic value of the week after sine transformation, and the value range after conversion is between -1 and 1; Represents the week characteristic value after cosine transformation, and the value range after transformation is between -1 and 1; Indicates the day of the week, ranging from 0 to 6, where 0 represents Sunday and 6 represents Sunday;

[0119] Generate time series data with periodic features, including sine and cosine representations of hours and days;

[0120] Time series analysis: Draw a flow time series graph and identify periodic components based on the flow change pattern in the flow time series graph;

[0121] Frequency domain analysis: Identify the main periodic components in the flow data based on Fourier transform and the identified periodic components to obtain periodic characteristics;

[0122] Data standardization: Split the data into training sets and test sets in chronological order, standardize the traffic data and periodic features in the training sets and test sets, and convert the data into a distribution with a mean of 0 and a standard deviation of 1;

[0123] Model training: The standardized traffic data and periodic features in the training set are used as the input of the LSTM model to train the LSTM model; the mean square error is used as the loss function, and the Adam optimizer is used for model training.

[0124] Exemplarily, the specific structure of the LSTM model is as follows:

[0125] The specific structure of the LSTM model

[0126] Input Layer

[0127] Feature input: contains standardized traffic data and periodic features. The dimension of input data is (batch_size, time_steps, feature_num).

[0128] batch_size indicates the amount of data processed in each batch; time_steps indicates the step length of the time series, that is, the length of historical data considered by the model; feature_num indicates the number of features at each time point (such as traffic value, periodic features, etc.).

[0129] LSTM Layer 1

[0130] Units: 64;

[0131] Activation function: Use tanh as the activation function and sigmoid as the activation function for the gating mechanism.

[0132] Input shape: (time_steps, feature_num);

[0133] Return sequence: Set to True to indicate that the output of each time step in the output sequence will be used as the input of the next layer.

[0134] LSTM Layer 2

[0135] Number of units: 64;

[0136] Activation function: Use tanh as the activation function and sigmoid as the activation function for the gating mechanism.

[0137] Input shape: (time_steps, lstm_units) (output from the first layer of LSTM).

[0138] Return sequence: set to True;

[0139] LSTM Layer 3

[0140] Number of units: 64

[0141] Activation function: Use tanh as the activation function and sigmoid as the activation function for the gating mechanism.

[0142] Return sequence: Set to False to output only the output of the last time step.

[0143] Fully Connected Layer

[0144] Number of units: 1, used to output the predicted flow value.

[0145] Activation function: linear activation function (linear), that is, no activation function, output continuous value.

[0146] Output Layer

[0147] Output dimension: 1, indicating the predicted traffic value.

[0148] Activation function: linear activation function.

[0149] Step 3: Based on the prediction results of the intelligent computing node, the collaborative control platform dynamically adjusts the configuration of the optoelectronic hybrid router;

[0150] The adjusting the configuration of the optoelectronic hybrid router includes updating the routing table of the optoelectronic hybrid router and adjusting the queue management strategy;

[0151] The steps of updating the routing table of the optoelectronic hybrid router are as follows:

[0152] Obtain traffic forecast data from intelligent computing nodes, including traffic demand for a period of time in the future;

[0153] Smooth the predicted data to remove noise:

[0154] ;

[0155] Where: is the smoothed traffic forecast value; is the original traffic prediction value; n is the time window size;

[0156] Calculate the cost of each path based on the smoothed traffic forecast data:

[0157] ;

[0158] Where: is the delay of the path; and represents the weight coefficient; is the predicted flow from source node s to destination node d; is the bandwidth capacity of path p; represents the transmission cost from the source node s to the destination node d through the path p;

[0159] Select the path with the lowest cost and meeting the traffic demand to update the routing table;

[0160] The steps of adjusting the queue management strategy are as follows:

[0161] Assign queue weights based on smoothed traffic forecast data:

[0162] ;

[0163] Where: represents the weight of the i-th queue; is the flow demand of the i-th path; is the flow demand of the jth path;

[0164] The queue management policy is adjusted based on the calculated queue weight to ensure that paths with high traffic volume are given higher priority in the queue.

[0165] Step 4: In the optical network, wavelength and bandwidth are dynamically allocated based on the predicted traffic demand;

[0166] The step 4 comprises the following steps:

[0167] Step 4.1: Initialize the wavelength and bandwidth resources of all paths;

[0168] Step 4.2: Calculate the wavelength requirement of each source-destination pair based on the traffic prediction data;

[0169] The wavelength requirement of each source-destination pair is obtained based on the following formula:

[0170] ;

[0171] Where: represents the number of wavelengths required to travel from source node s to destination node d in time t; represents the predicted value of the flow from the source node s to the destination node d at time t; Indicates the bandwidth capacity of each wavelength channel.

[0172] Step 4.3: Allocate bandwidth according to wavelength requirements to ensure that the total bandwidth does not exceed the available bandwidth;

[0173] The bandwidth allocation according to wavelength requirements is based on the following formula:

[0174] ;

[0175] Where: represents the bandwidth allocated from source node s to destination node d at time t; represents the total bandwidth available in the network at time t; represents the number of wavelengths required to travel from source node s to destination node d in time t; Indicates the bandwidth capacity of each wavelength channel.

[0176] Step 4.4: Calculate the transmission cost of each path and select the path with the lowest cost for wavelength and bandwidth allocation.

[0177] The transmission cost of each path is obtained based on the following formula:

[0178] ;

[0179] Where: is the delay of the path; and represents the weight coefficient; is the predicted flow from source node s to destination node d; is the bandwidth capacity of path p; Represents the transmission cost from source node s to destination node d through path p.

[0180] Step 5: Monitor the network status and traffic conditions in real time. When there is a deviation between the actual traffic and the predicted results, feedback is given to the corresponding intelligent computing node, and the routing configuration and resource allocation are adjusted through the control platform.

[0181] The step 5 comprises the following steps:

[0182] Compare the actual traffic data with the predicted traffic data of the intelligent computing node and calculate the prediction error:

[0183] ;

[0184] Where: represents the prediction error from source node s to destination node d at time t; represents the actual flow from source node s to destination node d at time t; represents the predicted flow from source node s to destination node d at time t;

[0185] Feedback to tune machine learning models:

[0186] ;

[0187] Where: represents the adjusted predicted flow from source node s to destination node d at time t; Represents the adjustment coefficient, which is used to control the adjustment range;

[0188] Steps 3 and 4 are performed based on the adjusted predicted traffic to obtain the adjusted configuration, wavelength, and bandwidth of the optoelectronic hybrid router.

[0189] The present invention arranges multiple intelligent computing nodes to collect local traffic data and network status data; then uses machine learning algorithms to predict future traffic changes, so that network resources can be dynamically adjusted according to actual needs, avoiding idle or overloaded resources, thereby greatly improving the utilization rate of network resources; and dynamically adjusts the configuration, wavelength and bandwidth of the router through the prediction results, which can effectively alleviate network congestion, reduce delays, and improve the overall performance of the network; and the present invention deploys multiple intelligent computing nodes under an optoelectronic hybrid router, which not only improves the computing and processing capabilities of the system, but also enhances reliability, flexibility and scalability, optimizes resource allocation and utilization, and improves user experience.

[0190] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A network traffic optimization method in an optical network fusion environment, characterized in that: The following steps are involved: Step 1: Deploy multiple intelligent computing nodes under an optoelectronic hybrid router; And collect local traffic data and network status data in real time through the data collection module; The location of the intelligent computing node in the deployment of the intelligent computing node at the network edge in step 1 is as follows: Determine the K value: By calculating the total error sum of squares for different K values, plotting the relationship between the total error sum of squares and K value, select the K value at the inflection point where the total error sum of squares is significantly reduced as the optimal number of clusters, where the total error sum of squares is calculated as follows: Where: K represents the number of clusters; C i represents the i-th cluster; x represents cluster C i A data point in i represents the centroid of the i-th cluster; ||x-μ i || 2 Represents the data point x to the centroid μ i The square of the Euclidean distance; Initialization: Initialize the centroids of K clusters; Update centroids: Assign each data point to the cluster to which the nearest centroid belongs; The centroid of each cluster is recalculated as the mean of all data points in the cluster. The centroid is updated repeatedly until it no longer changes or the change is less than a preset threshold. The centroid is updated as follows: Where: μ i represents the centroid of the i-th cluster; |C i | represents the number of data points in the i-th cluster; x represents cluster C i A data point in Perform K-means clustering using the determined optimal K value to obtain each cluster and its centroid; For each cluster, the node with the most obvious traffic change characteristics is selected as the intelligent computing node; Step 2: Use the machine learning model in the intelligent computing node to perform traffic forecasting based on the local traffic data and network status data collected in real time to predict future traffic changes; The machine learning model described in step 2 adopts an LSTM model; the training steps of the LSTM model are as follows: Data collection: Obtain historical data of network traffic, which includes timestamps and traffic values; Data cleaning: Process missing values ​​and outliers based on the acquired historical data to obtain a cleaned data set; Time series feature extraction: Extract time series features based on cleaned data; add time features to the cleaned data set; Periodic feature extraction: Based on the results of time series feature extraction, periodic feature extraction is performed to generate time series data with periodic features; Time series analysis: Draw a flow time series graph and identify periodic components based on the flow change pattern in the flow time series graph; Frequency domain analysis: Identify the main periodic components in the flow data based on Fourier transform and the identified periodic components to obtain periodic characteristics; Data standardization: Split the data into training sets and test sets in chronological order, standardize the traffic data and periodic features in the training sets and test sets, and convert the data into a distribution with a mean of 0 and a standard deviation of 1; Model training: The standardized traffic data and periodic features in the training set are used as the input of the LSTM model to train the LSTM model; mean square error is used as the loss function, and the Adam optimizer is used for model training; Step 3: Based on the prediction results of the intelligent computing node, the collaborative control platform dynamically adjusts the configuration of the optoelectronic hybrid router; The adjusting the configuration of the optoelectronic hybrid router includes updating the routing table of the optoelectronic hybrid router and adjusting the queue management strategy; The steps of updating the routing table of the optoelectronic hybrid router are as follows: Obtain traffic forecast data from intelligent computing nodes, including traffic demand for a period of time in the future; Smooth the predicted data to remove noise: Where: is the smoothed traffic forecast value; t-i is the original traffic prediction value; n is the time window size; Calculate the cost of each path based on the smoothed traffic forecast data: Where: delay(p) is the delay of the path; α and β are weight coefficients; traffic(s,d) is the predicted traffic from source node s to destination node d; capacity(p) is the bandwidth capacity of path p; cost(s,d,p) is the transmission cost from source node s to destination node d through path p; Select the path with the lowest cost and meeting the traffic demand to update the routing table; The steps of adjusting the queue management strategy are as follows: Assign queue weights based on smoothed traffic forecast data: Where: W i represents the weight of the i-th queue; D i is the flow demand of the ith path; D j is the flow demand of the jth path; Adjust the queue management policy based on the calculated queue weight to ensure that paths with high traffic volume are given higher priority in the queue. Step 4: In the optical network, wavelength and bandwidth are dynamically allocated based on the predicted traffic demand; Step 5: Monitor the network status and traffic conditions in real time. When there is a deviation between the actual traffic and the predicted results, feedback is given to the corresponding intelligent computing node, and the routing configuration and resource allocation are adjusted through the control platform.

2. The method for optimizing network traffic in an optical network convergence environment according to claim 1, characterized in that: The step 4 comprises the following steps: Step 4.1: Initialize the wavelength and bandwidth resources of all paths; Step 4.2: Calculate the wavelength requirement of each source-destination pair based on the traffic prediction data; Step 4.3: Allocate bandwidth according to wavelength requirements to ensure that the total bandwidth does not exceed the available bandwidth; Step 4.4: Calculate the transmission cost of each path and select the path with the lowest cost for wavelength and bandwidth allocation.

3. The method for optimizing network traffic in an optical network convergence environment according to claim 2, characterized in that: The wavelength requirement of each source-destination pair is obtained based on the following formula: Where: Wavelength Demand (s, d, t) represents the number of wavelengths required to travel from source node s to destination node d in time t; Traffic Prediction (s, d, t) represents the traffic prediction value from source node s to destination node d at time t; Channel Capacity Indicates the bandwidth capacity of each wavelength channel.

4. The method for optimizing network traffic in an optical network convergence environment according to claim 2, characterized in that: The bandwidth allocation according to wavelength requirements is based on the following formula: Bandwidth Allocation (s,d,t)=min(Wavelength Demand (s,d,t)×Channel Capacity ,Available Bandwidth ); Where: Bandwidth Allocation (s, d, t) represents the bandwidth allocated to the source node s from the destination node d at time t; Available Bandwidth Indicates the total bandwidth available in the network at time t; Wavelength Demand (s, d, t) represents the number of wavelengths required from source node s to destination node d in time t; Channel Capacity Indicates the bandwidth capacity of each wavelength channel.

5. The method for optimizing network traffic in an optical network convergence environment according to claim 2, characterized in that: The transmission cost of each path is obtained based on the following formula: Where: delay(p) is the delay of the path; α and β are weight coefficients; traffic(s,d) is the predicted traffic from source node s to destination node d; capacity(p) is the bandwidth capacity of path p; cost(s,d,p) is the transmission cost from source node s to destination node d through path p.

6. The method for optimizing network traffic in an optical network convergence environment according to claim 1, characterized in that: The step 5 comprises the following steps: Compare the actual traffic data with the predicted traffic data of the intelligent computing node and calculate the prediction error: Error(s,d,t)=Actual Traffic (s,d,t)-Predicted Traffic (s,d,t); Where: Error(s,d,t) represents the prediction error from source node s to destination node d at time t; Actual Traffic (s, d, t) represents the actual flow from source node s to destination node d at time t; Predicted Traffic (s, d, t) represents the predicted flow from source node s to destination node d at time t; Feedback to tune machine learning models: Adjusted Prediction (s,d,t)=Predicted Traffic (s,d,t)+αError(s,d,t); Where: Adjusted Prediction (s, d, t) represents the adjusted predicted flow from source node s to destination node d at time t; α represents the adjustment coefficient, which is used to control the adjustment range; Steps 3 and 4 are performed based on the adjusted predicted traffic to obtain the adjusted configuration, wavelength, and bandwidth of the optoelectronic hybrid router.