Cross-domain self-organizing network flow measurement method, device, medium and product

Through meta-learning and reinforcement learning models, network traffic measurement is optimized, and the minimum set of measurement nodes is determined, which solves the accuracy problem of cross-domain self-organized network traffic prediction, and realizes efficient traffic measurement to adapt to dynamic network changes.

CN119071804APending Publication Date: 2024-12-03NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411289481.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing network traffic measurement optimization methods cannot accurately predict traffic in cross-domain ad hoc networks, especially when node computing power and storage capacity are limited, it is difficult to handle complex network data flows.

Method used

Using meta-learning and reinforcement learning models, the traffic matrix and routing matrix are constructed through the training data set, the terminal node collection with the minimum number of measurements is determined, network traffic measurement is optimized, and data packets are crawled using Wireshark software to achieve cross-domain self-organized network traffic. accurate prediction.

Benefits of technology

It improves the accuracy and efficiency of network traffic measurement, adapts to network dynamic changes, reduces resource consumption, and is suitable for complex environments of cross-domain self-organizing networks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119071804A_ABST
    Figure CN119071804A_ABST
Patent Text Reader

Abstract

The invention discloses a cross-domain self-organizing network flow measurement method and device, a medium and a product, and relates to the field of network flow prediction, and the method comprises the steps: obtaining a network flow data packet of a to-be-predicted cross-domain self-organizing network within a preset time; the to-be-predicted cross-domain ad hoc network comprises a cluster head and a plurality of terminal nodes; determining a traffic matrix and a routing matrix of the to-be-predicted cross-domain ad hoc network according to the network traffic data packet; according to the traffic matrix and the routing matrix, utilizing a to-be-measured terminal node prediction model to determine a minimum measurement number terminal node set; the minimum measurement number terminal node set comprises a plurality of terminal nodes to be measured; wherein the to-be-measured terminal node prediction model is obtained by training a meta-learning and reinforcement learning model by using a training data set; and measuring the network flow of the to-be-predicted cross-domain ad hoc network based on the minimum measurement number of terminal node sets. According to the invention, the measurement precision and efficiency of the cross-domain self-organizing network flow are effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of network traffic prediction, and in particular to a cross-domain self-organizing network traffic measurement method, device, medium and product. Background Art

[0002] The cross-domain self-organizing network is a cross-media network that is decentralized, highly resistant to destruction, and automatically corrects routing after topology changes. Packet switching technology is used for data transmission between end nodes, and there are wireless communications of different frequencies between network domains. Due to the lack of necessary infrastructure outdoors, centralized networks are difficult to deploy. The multi-hop routing of the cross-domain self-organizing network enables nodes in different domains to communicate in a network, so that perceived sea and air information can be transmitted. The nodes in the cross-domain self-organizing network generate different data streams, making the entire network information flow more complicated. As time goes by, network nodes continue to enter and exit, and the topology of the network also changes dramatically. The application scenarios of this application are as follows: Figure 1 As shown in the figure. In order to better manage the cross-domain self-organizing network, it is necessary to measure the status of the overall network traffic. Direct measurement technology is an effective method to obtain the network operation status. However, the characteristics of the large number of nodes and large traffic values ​​in the network make the communication load and resource consumption of direct measurement large. In order to reduce the direct measurement resource consumption caused by network management, a network traffic measurement optimization method is needed to optimize the measurement mechanism.

[0003] There are generally two methods for optimizing network traffic measurement: one is to reduce the measurement cost by reducing the CPU, memory and energy utilization required for measurement, such as Ghode adding energy restrictions to the wireless ad hoc network protocol, and Salsano studying the network monitoring mechanism at the system kernel level. The other method only measures nodes with large network traffic, such as Nie using Q learning to optimize network traffic measurement. Although these methods can optimize network traffic measurement, they are not suitable for cross-domain self-organizing networks. Cross-domain self-organizing networks have the characteristics of low node computing power, low storage and high dynamics. Due to the changes in the mobile positions of cross-domain nodes, the overall network routing is also changing in real time, and the computing power and storage capacity of mobile nodes are limited, but they need to process large and more complex network data.

[0004] In summary, current network traffic measurement optimization methods cannot accurately predict the traffic of cross-domain ad hoc networks. Summary of the invention

[0005] The purpose of this application is to provide a cross-domain self-organizing network traffic measurement method, device, medium and product to achieve accurate prediction of cross-domain self-organizing network traffic.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a cross-domain self-organizing network traffic measurement method, comprising:

[0008] Obtaining network traffic data packets within a preset time of a cross-domain self-organizing network to be predicted; the cross-domain self-organizing network to be predicted includes a cluster head and multiple terminal nodes;

[0009] Determine the traffic matrix and routing matrix of the cross-domain self-organizing network to be predicted according to the network traffic data packet;

[0010] According to the traffic matrix and the routing matrix, a terminal node prediction model to be measured is used to determine a terminal node set with a minimum number of measurements; the terminal node set with a minimum number of measurements includes a plurality of terminal nodes to be measured; wherein the terminal node prediction model to be measured is obtained by training a meta-learning and reinforcement learning model using a training data set; the training data set includes a traffic matrix and a routing matrix of a cross-domain self-organizing network for training and a corresponding terminal node set with a minimum number of measurements;

[0011] The network traffic of the cross-domain self-organizing network to be predicted is measured based on the minimum measurement number terminal node set.

[0012] Optionally, the meta-learning and reinforcement learning models are trained using the training dataset, including:

[0013] Modeling a Markov decision process based on the routing matrix and traffic matrix of the cross-domain self-organizing network in each unit time, and constructing multiple network environments;

[0014] Determine, in each of the network environments, a plurality of trajectory data of the terminal node interacting with the network and a loss function value of the meta-learning and reinforcement learning model on a task; the task is a traffic measurement task;

[0015] According to all the trajectory data and the loss function value, the network parameters of the meta-learning and reinforcement learning models are updated to obtain a prediction model for the terminal node to be measured.

[0016] Optionally, according to the traffic matrix and the routing matrix, using a prediction model for terminal nodes to be measured, determining a set of terminal nodes with a minimum number of measurements specifically includes:

[0017] Initializing the network parameters of the prediction model of the terminal node to be measured to obtain an initial prediction model of the terminal node to be measured;

[0018] In the current network environment, fine-tune the network parameters of the initial prediction model of the terminal node to be measured to obtain a fine-tuned prediction model of the terminal node to be measured;

[0019] Setting the terminal node i of the current measurement to 1, and inputting the states of all OD flows into the fine-tuned prediction model of the terminal node to be measured, to obtain a terminal node to be measured;

[0020] Calculate the total amount of network traffic currently measured;

[0021] Determine whether the total amount of network traffic is greater than a set total amount;

[0022] If so, the set of terminal nodes with the minimum number of measurements is obtained;

[0023] If not, the method returns to "inputting the states of all OD flows into the fine-tuned prediction model of the terminal node to be measured to obtain a terminal node to be measured" until the total amount of network traffic currently measured is greater than the set total amount.

[0024] Optionally, measuring the network traffic of the cross-domain self-organizing network to be predicted based on the minimum measured number of terminal nodes set specifically includes:

[0025] The cluster head node of the cross-domain self-organizing network to be predicted sends a measurement instruction to the terminal node to be measured;

[0026] The terminal node to be measured sends the captured network traffic data packets to the cluster head node, and other terminal nodes stop sending network traffic data packets to the cluster head node;

[0027] The network traffic of the cross-domain self-organizing network to be predicted is determined according to the network traffic data packets captured by the terminal node to be measured.

[0028] Optionally, Wireshark software is used on the terminal node to capture network traffic data packets.

[0029] In a second aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described cross-domain self-organizing network traffic measurement methods.

[0030] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described cross-domain self-organizing network traffic measurement methods.

[0031] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned cross-domain self-organizing network traffic measurement methods.

[0032] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0033] The present application provides a cross-domain self-organizing network traffic measurement method, device, medium and product, which determines the node set for performing network traffic measurement at the next moment according to the network operating environment at the previous moment, such as network topology, network end-to-end traffic size and other information. The cross-domain self-organizing network traffic measurement method of the present application can fit the trend of dynamic changes in the network and effectively improve the measurement accuracy and efficiency of network traffic. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments 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 work.

[0035] Figure 1 This is a schematic diagram of the application scenario of this application;

[0036] Figure 2 A flow chart of a cross-domain self-organizing network traffic measurement method provided in one embodiment of the present application;

[0037] Figure 3 This is a schematic diagram of network traffic measurement optimization based on meta-learning and reinforcement learning in an embodiment of the present application;

[0038] Figure 4 Schematic diagram of the reinforcement learning method for this application;

[0039] Figure 5 Schematic diagram of the flow measurement optimization method based on meta-learning and PPO adopted in this application;

[0040] Figure 6 This is a graph showing the comparison results of three methods for the proportion of fixed traffic collected in the Abilene network as applied in this application;

[0041] Figure 7 This application is used in Comparison results of three methods for the proportion of fixed network traffic collection;

[0042] Figure 8 This is a comparison result diagram of three methods for the proportion of fixed traffic collection in wireless ad hoc networks applied in this application;

[0043] Fig. 9 This is the comparison result of two methods for the number of fixed data collection network nodes applied to the Abilene network.

[0044] Fig.10 This application applies to Comparison results of two methods for collecting the number of network nodes in a fixed network

[0045] Fig.11 This is a comparison of the two methods used in this application to collect the number of fixed network nodes in a wireless self-organizing network.

[0046] Fig.12 A comparison chart of the results of meta-learning and normal distribution initial parameter training in wireless self-organizing networks applied in this application;

[0047] Fig.13 This is a comparison chart of the results of meta-learning and normal distribution initial parameter training in a wireless self-organizing network (time interval 10s) applied in this application;

[0048] Fig.14 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0051] In an exemplary embodiment, Figure 2 and Figure 3 As shown, a cross-domain self-organizing network traffic measurement method is provided, including:

[0052] S1: Obtain network traffic data packets within a preset time of a cross-domain self-organizing network to be predicted; the cross-domain self-organizing network to be predicted includes a cluster head and multiple terminal nodes.

[0053] In practical applications, a wireless self-organizing network environment with a cluster head and multiple terminal nodes is initialized, and each self-organizing network terminal node uses the BATMAN-adv protocol for self-organizing networking. After the self-organizing network is established, the end-to-end will send network traffic based on the application, such as voice, image and other information. This application only focuses on the size of network traffic packets, not the specific information of each network flow. Then, use Wireshark software on each terminal node to capture the network traffic packets flowing through this network card and send them to the cluster head node that aggregates the traffic.

[0054] S2: Determine the traffic matrix and routing matrix of the cross-domain self-organizing network to be predicted according to the network traffic data packet.

[0055] In practical applications, network traffic collection optimization requires network preprocessing. The cluster head node obtains the topology matrix, traffic matrix, and routing matrix of the overall network based on the network data packets collected within 10 seconds.

[0056] The network topology matrix, traffic matrix, and routing matrix are calculated through network data packets within 10 seconds, including:

[0057] Definition: Y represents the network link load, A represents the network routing matrix, and X represents the network traffic matrix. Each element of the routing matrix is ​​represented by A(l,o,d). If A(l,o,d) = 1, it means that a flow from terminal node o to terminal node d passes through link l, otherwise A(l,o,d) = 0. The network data packet can extract the source address, destination address, source MAC address, destination MAC address, link information, and link information of the data packet flow.

[0058] S3: According to the traffic matrix and the routing matrix, using the terminal node prediction model to be measured, determine the minimum number of terminal nodes to be measured set; the minimum number of terminal nodes to be measured set includes multiple terminal nodes to be measured; wherein, the terminal node prediction model to be measured is obtained by training the meta-learning and reinforcement learning models using the training data set; the training data set includes the traffic matrix and routing matrix of the cross-domain self-organizing network for training and the corresponding minimum number of terminal nodes to be measured set.

[0059] Among them, the reinforcement learning model is Figure 4 The model structure parameters are shown in Table 1. The meta-learning model is completely consistent with the reinforcement learning model architecture and is used to save the initial parameters of the reinforcement learning model with generalization ability.

[0060] Table 1 Meta-learning and reinforcement learning model structure parameters

[0061]

[0062] The training process of the network traffic measurement optimization method based on meta-learning and PPO is as follows: Figure 5 As shown in the figure, different network environments are first obtained based on the routing matrix and traffic matrix of each unit time of the network. A gradient is determined for each environment in the inner loop, and the sum of the gradients is the update direction of the outer loop neural network. The policy network and value network in the PPO method perform gradient descent respectively, and the trained neural network parameters are obtained after multiple rounds of training. The trained neural network parameters can adapt to a variety of network environments. Compared with the normal distribution parameters, their network traffic measurement accuracy is higher and has faster convergence.

[0063] As an optional implementation method, the meta-learning and reinforcement learning models are trained using the training data set, specifically including:

[0064] S31: Modeling a Markov decision process based on the routing matrix and traffic matrix of the training cross-domain self-organizing network in each unit time, and constructing multiple network environments.

[0065] In practical applications, we first define that each network port can be collected through the network measurement tool NetFlow. The present invention assumes that there are N terminal nodes in the wireless self-organizing network, and there are K (l) Links, r n,k Indicates whether the kth link in the nth terminal node is measured. When the link is measured, r n,k =1, otherwise r n,k = 0. Define d n =1 means that the packet capture tool NetFlow is enabled on the nth terminal node, otherwise its value is 0. The goal of network measurement optimization is to activate a small number of terminal nodes so that most of the network traffic can be captured. The proportion of network traffic is represented by γ, and all network traffic per unit time is represented by S. S n,k represents the traffic on the kth link in the nth terminal node. At this time, the network traffic measurement optimization problem can be modeled as:

[0066]

[0067] Input network routing matrix A, network traffic matrix X, internal learning rate α, external learning rate β, PPO (Reinforcement Learning, Proportional-to-Probability Optimization) cutoff range ε, the principle diagram of the PPO method is as follows Figure 4 As shown. According to the network routing matrix A per unit time t and the traffic matrix X t Model the Markov decision process and construct multiple network environments p(τ); randomly initialize the meta-learning PPO network parameters θ.

[0068] The reinforcement learning modeling of the network measurement optimization problem is as follows. State: Whether the OD flow (Origin-Destination flow, which refers to the network traffic from one terminal node to another) is measured is used as the reinforcement learning state. Assume that there are M OD flows in the network, that is, Z = {v0, v1..., v M-1}, if v i=1 means that the i-th OD flow has been detected, otherwise its value is 0. Action: The action space is {n|n∈N}, where n represents the number of the terminal node, which specifies the terminal node that needs to be captured and measured. Reward: The reward value is defined as the size of all traffic values ​​obtained by selecting terminal node n. Each terminal node may have multiple network ports, and each network port has multiple OD flows passing through. The OD flow that has not been measured in state X and the newly obtained OD flow after measuring terminal node n is the newly captured network traffic, and the sum of their traffic sizes is the current reward value. In order to prevent the fluctuation range of the reward value from being too large, this application first normalizes the reward value. Next state: The newly captured OD flow after taking action and the v in state Z i = 1 is the state Z′ at the next moment, that is, the newly measured OD flow is added after selecting action a. According to this definition, the four-tuple (Z, a, r, Z′) is saved to the experience pool, which is the process of constructing the current network environment p(τ).

[0069] S32: Determine multiple trajectory data of the terminal node interacting with the network in each of the network environments and the loss function value of the meta-learning and reinforcement learning models on the task; the task is a flow measurement task, that is, a flow measurement task that optimizes the entire network within a certain time slot. When a node with more flow is selected for measurement, the reinforcement learning model will give a larger reward value, thereby prompting the model to continuously improve.

[0070] In practice, we sample network traffic in each network environment to measure and optimize the batch task τ i ~p(τ); in task τ i Monte Carlo sampling is performed in , using the meta-learning model f θ Select a strategy and obtain K trajectory data of the terminal node interacting with the network Using trajectory data And the loss function value The evaluation concluded This loss represents the strategy (the meta-learning model f θ ) in task τ i The test loss function on , its update strategy based on the PPO method (reinforcement learning method) is:

[0071]

[0072] Among them, θ k+1 is the policy network parameter that needs to be updated in the k+1th iteration; T is the overall step size for interacting with the network environment; τ is a batch task of a sampled network traffic measurement optimization environment; π θ is the policy network function, that is, the policy currently being trained; a t is the action chosen at time t; st is the state at time t; is the policy network function of the k+1th iteration; is the advantage function, i.e. the cumulative reward value minus the average reward value; is the truncation function, and ε is the truncation ratio factor, which is used to limit the ratio of the new strategy to the old strategy.

[0073] Use gradient descent to calculate the adjusted internal PPO model (reinforcement learning model) parameters: Then in task τ i Monte Carlo sampling is performed in the neural network model f θ′ Make strategy selection and obtain trajectory data Calculate the neural network model f θ′ In the task τ i The loss value on Until the network environment of each time slot is traversed.

[0074] S33: According to all the trajectory data and the loss function value, the network parameters of the meta-learning and reinforcement learning models are updated to obtain a prediction model of the terminal node to be measured.

[0075] In practical applications, use each and Update the outer meta-learning PPO network parameters The outer loop updates the meta-model network after multiple iterations, and outputs the parameters θ of the trained meta-learning and reinforcement learning models.

[0076] The input of the terminal node prediction model to be measured is a single network routing matrix A t , network traffic matrix X t , the trained meta-learning and PPO model parameters θ, the proportion of network traffic α. The output is the minimum number of network terminal nodes that need to be measured (the minimum number of terminal nodes to be measured) D.

[0077] As an optional implementation, S3 specifically includes:

[0078] Initialize the network parameters of the terminal node prediction model to be measured to obtain the initial prediction model of the terminal node to be measured. In practical applications, the parameters θ of the PPO network model (the terminal node prediction model to be measured) are initialized based on meta-learning and the PPO model.

[0079] In the current network environment, fine-tune the network parameters of the initial prediction model of the terminal node to be measured to obtain a fine-tuned prediction model of the terminal node to be measured. For the current network environment, fine-tune the network model θ to obtain θ′, and the fine-tuning strategy is the same as S31-S33.

[0080] Set the terminal node i=1 for the current measurement, and input the states of all OD flows into the fine-tuned prediction model of the terminal node to be measured to obtain a terminal node to be measured. Set the network terminal node number i=1 for the current measurement, and input the states of all OD flows (need to be collected or not collected) {v0,v1...,v M-1}, and obtain the first terminal node d1 that needs to be measured.

[0081] Calculates the total amount of network traffic currently being measured.

[0082] Determine whether the total amount of network traffic is greater than a set total amount.

[0083] If so, the set of terminal nodes with the minimum number of measurements is obtained.

[0084] If not, the method returns to "inputting the states of all OD flows into the fine-tuned prediction model of the terminal node to be measured to obtain a terminal node to be measured" until the total amount of network traffic currently measured is greater than the set total amount.

[0085] In practical applications, iterate through each measured terminal node. Calculate the total amount of network traffic currently measured. Collecting the same flow multiple times will not increase the reward value of the environment. If the proportion of collected traffic is greater than α, the number of selected terminal nodes is sufficient, the method ends, and the measurement set D is output. When the proportion of collected traffic is insufficient, let i = i + 1, and input the status of all OD flows (need to be collected or not) {v0,v1...,v M-1}, get the measured terminal node d i ; d i Add to the output set D.

[0086] S4: Measuring the network traffic of the cross-domain self-organizing network to be predicted based on the minimum measurement number terminal node set.

[0087] In actual applications, the cluster head node sends the measurement node information to the terminal nodes that need to be measured. The terminal nodes to be measured send the captured network data packets to the cluster head node where traffic is aggregated, and the terminal nodes that are not measured suspend the data packet aggregation process to the cluster head node.

[0088] As an optional implementation manner, measuring the network traffic of the cross-domain self-organizing network to be predicted based on the minimum measured number of terminal nodes set specifically includes:

[0089] The cluster head node of the cross-domain self-organizing network to be predicted sends a measurement instruction to the terminal node to be measured.

[0090] The terminal node to be measured sends the captured network traffic data packets to the cluster head node, and other terminal nodes suspend sending network traffic data packets to the cluster head node.

[0091] The network traffic of the cross-domain self-organizing network to be predicted is determined according to the network traffic data packets captured by the terminal node to be measured.

[0092] Three network datasets are used in the simulation of this application. The wired network includes the Abilene backbone network dataset and Dataset,The wireless network uses the NS3 network simulation tool for software simulation and data generation.,Hereinafter, the network simulated in NS3 is referred to as the wireless,ad hoc network dataset.

[0093] Example 1: The Abilene backbone network consists of 12 nodes and 54 links, which are divided into 30 links between internal nodes and 24 links connected to the external network.

[0094] Example 2, The dataset consists of 45 nodes and 142 links.

[0095] Example 3, the wireless self-organizing network data set is generated using the NS3 network simulation tool. The simulated scenario is that each wireless self-organizing node moves within a fixed range of a rectangular block, collects surrounding environment information, and sends the obtained information to other nodes.

[0096] In order to more realistically restore the scenario of sending network data packets, the application layer traffic size and sending interval are analyzed using drone camera data. The mobile model adopts the Markov Gaussian mobile model, which is characterized by memory, and the next movement trajectory depends on the previous position and velocity vector. This application uses linear programming (LP) and Q learning methods as comparison methods. The linear programming method establishes a constrained optimization model based on the maximum and minimum traffic of the traffic node for solution. The Q learning method counts the OD flows passing through each link, and determines the reward value at the moment based on the relationship between the flows during the selection of each interface. The larger the reward value, the more new network OD flows are obtained by selecting the interface.

[0097] Figure 6 The figure shows the comparison of network node results of minimizing the measured traffic ratio in the Abilene network. The blue line shows the measurement results of linear programming, the red line shows the measurement results using Q learning, and the green line shows the measurement optimization results of PPO-meta. The horizontal axis shows the proportion of captured network traffic from 50% to 100%, and the vertical axis shows the number of nodes required to measure to capture the traffic ratio. Figure 6It can be seen that the PPO-meta method requires the least number of nodes to measure any fixed ratio of traffic, and achieves the best network traffic measurement optimization effect, which is averaged after 28 independent repeated experiments. The Q-learning measurement optimization effect is lower than the PPO-meta method and higher than the linear programming method. This is because PPO-meta considers the traffic size and network routing in the network, but the reward value of Q-learning only considers the network routing. Figure 7 Indicated in Measurement optimization results in the network, from Figure 7 It can be seen that when the measurement traffic ratio is between 50% and 72%, the PPO-meta measurement effect is slightly better than the Q-learning method. When the measurement ratio is 73%, the number of nodes required for Q-learning begins to increase. Figure 8 The measurement optimization results of wireless ad hoc networks are shown. The measurement optimization effects of PPO-meta are better than those of Q-learning and linear programming methods.

[0098] Fig. 9 The network traffic collected by the PPO-meta method and Q-learning in the Abilene network is shown. When the number of measured nodes is greater than 1, the PPO-meta method captures more traffic than the Q-learning method. Since the linear programming method is a constraint problem of minimizing the number of nodes under the condition of link traffic constraints, its method cannot maximize the captured network traffic under the condition of a fixed number of nodes. The PPO-meta method and Q-learning select one node each time to measure the amount of traffic passing through it, so the second method can be compared. Fig.10 Indicated in The optimization results measured in the network show that when the number of fixed nodes is small, the traffic captured by the Q-learning method is larger than the traffic measured by PPO-meta. The reason is that when Q-learning selects the first node to be measured, it selects the node with the largest covariance in the Q reward matrix. Fig.11 represents the measurement optimization result in the online self-organizing network, from Fig.11 It can be seen that when any number of nodes is fixed, the network traffic captured by the PPO-meta method is larger than that of the Q-learning method. Therefore, the PPO-meta method achieves better network traffic measurement optimization effect.

[0099] Fig.12 The changes in network traffic size captured by the PPO-meta method and the PPO-normal method in a wireless ad hoc network as the number of training rounds increases are shown. The number of fixed measurement nodes in the simulation is set to 15. Fig.12 The comparison of the training effect of the initial parameters obtained by the outer layer of meta-learning with a unit time of 100s and the normal distribution parameters is given in . Fig.12It can be seen that the network traffic measured by the PPO-meta method is higher than that of the PPO-normal method with normal distribution parameters, indicating that the meta-learning method has obtained better initial training parameters and can significantly improve the training iteration effect of the PPO network. Fig.13 The comparison between PPO-meta and PPO-normal methods is shown when the unit time is 10s. Fig.13 It can be seen that the effect of PPO-meta optimization of neural network iteration is lower than the improvement effect of 100s per unit time. The reason is that when the unit time is 10s, the network traffic fluctuates too violently, resulting in a large gap between the routing topology and the network traffic task per unit time. The initial parameter measurement optimization effect obtained by meta-learning is reduced, but it is still higher than the PPO-normal method with normal distribution parameters.

[0100] This paper proposes a network flow measurement optimization technology based on meta-learning and proximal strategy optimization, which determines the node set to perform network flow measurement at the next moment according to the network operating environment at the previous moment, such as network topology, network end-to-end flow size and other information. Through simulation verification, the proposed method can fit the trend of network dynamic changes and effectively improve the measurement efficiency of network flow.

[0101] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.14 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the traffic matrix and routing matrix of the cross-domain self-organizing network. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a cross-domain self-organizing network traffic measurement method is implemented.

[0102] Those skilled in the art will understand that Fig.14The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned cross-domain self-organizing network traffic measurement method when executing the computer program.

[0103] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the above-mentioned cross-domain self-organizing network traffic measurement method when executed by a processor.

[0104] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the above-mentioned cross-domain self-organizing network traffic measurement method when executed by a processor.

[0105] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0106] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0107] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0108] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A cross-domain self-organizing network traffic measurement method, It is characterized in that include: Obtain network traffic data packets within a preset time of the cross-domain self-organizing network to be predicted; The cross-domain self-organizing network to be predicted includes a cluster head and multiple terminal nodes; Determine the traffic matrix and routing matrix of the cross-domain self-organizing network to be predicted according to the network traffic data packet; According to the traffic matrix and the routing matrix, using a prediction model for terminal nodes to be measured, determining a set of terminal nodes with a minimum number of measurements; The minimum number of terminal nodes to be measured set includes a plurality of terminal nodes to be measured; wherein the terminal node prediction model to be measured is obtained by training a meta-learning and reinforcement learning model using a training data set; the training data set includes a traffic matrix and a routing matrix of a cross-domain self-organizing network for training and a corresponding minimum number of terminal nodes to be measured set; The network traffic of the cross-domain self-organizing network to be predicted is measured based on the minimum measurement number terminal node set.

2. The cross-domain self-organizing network traffic measurement method according to claim 1, It is characterized in that Use the training dataset to train the meta-learning and reinforcement learning models, including: Modeling the Markov decision process based on the routing matrix and traffic matrix of the cross-domain self-organizing network in each unit time, and constructing multiple network environments; Determine, in each of the network environments, a plurality of trajectory data of the terminal node interacting with the network and the loss function value of the meta-learning and reinforcement learning model on the task; the task is a traffic measurement task; According to all the trajectory data and the loss function value, the network parameters of the meta-learning and reinforcement learning models are updated to obtain a prediction model for the terminal node to be measured.

3. The cross-domain self-organizing network traffic measurement method according to claim 1, It is characterized in that According to the traffic matrix and the routing matrix, using the terminal node prediction model to be measured, determining a minimum number of terminal nodes to be measured, specifically including: Initializing the network parameters of the prediction model of the terminal node to be measured to obtain an initial prediction model of the terminal node to be measured; In the current network environment, fine-tune the network parameters of the initial prediction model of the terminal node to be measured to obtain a fine-tuned prediction model of the terminal node to be measured; Setting the terminal node i of the current measurement to 1, and inputting the states of all OD flows into the fine-tuned prediction model of the terminal node to be measured, to obtain a terminal node to be measured; Calculate the total amount of network traffic currently measured; Determine whether the total amount of network traffic is greater than a set total amount; If so, the set of terminal nodes with the minimum number of measurements is obtained; If not, the method returns to "inputting the status of all OD flows in the fine-tuned prediction model of the terminal node to be measured to obtain a terminal node to be measured" until the total amount of network traffic currently measured is greater than the set total amount.

4. The cross-domain self-organizing network traffic measurement method according to claim 1, It is characterized in that Measuring the network traffic of the cross-domain self-organizing network to be predicted based on the minimum measurement number terminal node set specifically includes: The cluster head node of the cross-domain self-organizing network to be predicted sends a measurement instruction to the terminal node to be measured; The terminal node to be measured sends the captured network traffic data packets to the cluster head node, and other terminal nodes stop sending network traffic data packets to the cluster head node; The network traffic of the cross-domain self-organizing network to be predicted is determined according to the network traffic data packets captured by the terminal node to be measured.

5. The cross-domain self-organizing network traffic measurement method according to claim 1, It is characterized in that Wireshark software is used on the terminal node to capture network traffic data packets.

6. A computer device, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the cross-domain self-organizing network traffic measurement method described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the cross-domain self-organizing network traffic measurement method described in any one of claims 1 to 5 is implemented.

8. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the cross-domain self-organizing network traffic measurement method described in any one of claims 1 to 5 is implemented.