CN112399483B - Constraint route control system based on GNs graph predictor - Google Patents

Constraint route control system based on GNs graph predictor Download PDF

Info

Publication number
CN112399483B
CN112399483B CN202011211827.9A CN202011211827A CN112399483B CN 112399483 B CN112399483 B CN 112399483B CN 202011211827 A CN202011211827 A CN 202011211827A CN 112399483 B CN112399483 B CN 112399483B
Authority
CN
China
Prior art keywords
network
gns
control system
predictor
graph
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011211827.9A
Other languages
Chinese (zh)
Other versions
CN112399483A (en
Inventor
秦卫忠
杨晨
韩保华
王善忠
胥晓冬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiahuan Technology Co ltd
Original Assignee
Jiahuan Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiahuan Technology Co ltd filed Critical Jiahuan Technology Co ltd
Priority to CN202011211827.9A priority Critical patent/CN112399483B/en
Publication of CN112399483A publication Critical patent/CN112399483A/en
Application granted granted Critical
Publication of CN112399483B publication Critical patent/CN112399483B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/08—Load balancing or load distribution
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
    • H04W28/24—Negotiating SLA [Service Level Agreement]; Negotiating QoS [Quality of Service]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W40/00—Communication routing or communication path finding
    • H04W40/02—Communication route or path selection, e.g. power-based or shortest path routing
    • H04W40/12—Communication route or path selection, e.g. power-based or shortest path routing based on transmission quality or channel quality
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W40/00—Communication routing or communication path finding
    • H04W40/24—Connectivity information management, e.g. connectivity discovery or connectivity update
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W40/00—Communication routing or communication path finding
    • H04W40/34—Modification of an existing route

Abstract

The invention provides a constraint route control system based on a GNs graph predictor, which comprises a QoS route control system and the GNs graph predictor; the GNs graph predictor is used for simulating the dynamic change of the communication flow of the whole network when different network nodes have instantaneous peak loads and providing a prejudgment standard for the QoS route control system according to different communication flow changes; the GNs graph predictor performs offline calculation of network communication flow, and feeds an offline calculation result back to the QoS route control system, so that the implementation efficiency of the current QoS route is improved; the QoS routing control system regularly feeds back historical data of network communication flow to the GNs diagram predictor, and the GNs diagram predictor carries out off-line calculation again according to the received feedback data. According to the method, the node data propagation condition under the condition of local node instantaneous peak load is simulated and calculated through the GNs network, and effective local isolation and route change are constructed according to the condition, so that the instantaneous peak load is controlled in a small range, and the impact and the damage of the instantaneous peak load to the overall communication network are avoided.

Description

Constraint route control system based on GNs graph predictor
Technical Field
The invention relates to the technical field of network routing control, in particular to a network load balancing technology, and specifically relates to a constraint routing control system based on a GNs graph predictor.
Background
With the advent of the age 5G, there are many applications in Internet multimedia, such as video chat, live web broadcast, web talk, and distance education. Data transmission needs to meet different quality of service (QoS) requirements, and multi-QoS constrained routing becomes an effective way to solve QoS problems. Meanwhile, under the condition of a 5G network, if the communication flow of part of nodes is overhigh instantaneously, the large-scale data is diffused in the surrounding 5G nodes to cause the blockage of a local network and block other normal flows passing through the nodes, thereby affecting the stability of the whole network. In addition, in actual network maintenance and operation, the current network structure needs to be adjusted due to node debugging and the like, and the original QoS requirement needs to be guaranteed at the same time, which requires a higher requirement for network QoS-constrained routing, especially QoS-constrained routing control during local peak data transmission.
The QoS routing protocol comprises a multi-path QoS routing protocol and a node-controlled single-hop path QoS routing protocol. The algorithm can improve the network throughput and the network performance degradation, optimize the resource configuration, balance the network load, realize the optimization of the network global resource utilization rate and maximize the requirement of the network for accepting other QoS parameters no matter the multi-path Qos routing protocol or the QoS path of the single-hop path. Under normal network carrying, considering end-to-end reliability, time delay and energy consumption, the common QoS routing protocol can realize the functions of preprocessing the original data of the node, accelerating convergence of pruning rules and reducing the whole energy consumption of the network by a node energy perception awakening mechanism. However, the QoS protocol does not make specific research on local peak load conditions, and there are the following technical blind points or difficulties:
1) route optimization for local peak load. Under normal network operating conditions, it is generally difficult to encounter peak loads at local nodes. If some special situations occur, for example, some nodes are hacked, the local area nodes have a particularly high peak of network traffic, and if the local area nodes are not controlled, the traffic data can quickly affect the surrounding network nodes. Currently, there is no deeply studied and effective solution for such local peak load route optimization;
2) multiple elements and variations of network structure. In an actual engineering environment, a network structure is often changed due to multiple complex factors such as service expansion and technical maintenance. In some cases, when the network structure is changed, the peak traffic occurs for some reasons, and then the existing routing control system cannot adjust the changed network structure in time;
3) the large-scale occlusion model is computationally difficult. In a real environment, due to the adoption of multiple technical means, generally, the situation of large-scale network blocking is less. As such, people have less knowledge about the actual situation of large-scale network congestion, and the related research is insufficient. Especially, under the multi-node network environment, how to calculate the large-scale network blocking condition is still an engineering problem.
Therefore, the main purpose of network load scheduling and provisioning is to avoid network congestion, but there are many times when there is an especially high peak traffic at some nodes of the network. The peak traffic is often sudden and irregular, and the peak traffic can bring a large load impact to the network, resulting in overloading of a large area of network nodes. Existing route management systems are difficult to handle in the face of such peak traffic.
Disclosure of Invention
The invention aims to provide a constraint route control system based on a GNs graph predictor, aiming at the network transmission problem of sudden excess communication reception of local nodes under a wireless network, which adopts an independent GNs predictor and an offline calculation mode to simulate peak traffic propagation modes under multiple nodes of different routing strategies, selects a route control strategy combination with optimal QoS constraint from the peak traffic propagation modes, and integrates corresponding peak traffic characteristics and a strategy combination scheme into the existing network system, so that the route control strategies can be automatically switched when similar network peak modes are met, the optimal route control effect under the peak traffic of the local nodes is achieved, the problem of large-area network overload caused by local instantaneous peak traffic is avoided, and the overall network stability is improved.
In order to achieve the above object, a first aspect of the present invention provides a constraint route control system based on GNs graph predictor, comprising a QoS route control system and a GNs graph predictor;
the QoS route control system is used for carrying out route forwarding control on each node in the network;
the GNs graph predictor is used for simulating the dynamic change of the communication flow of the whole network when the single node or the multiple nodes in different network nodes have instantaneous peak load and providing a prejudgment standard for the QoS route control system according to different communication flow changes;
the GNs diagram predictor is independent of the QoS route control system, offline calculation of network communication flow is carried out, offline calculation results are fed back to the QoS route control system, and implementation efficiency of a current QoS route is improved;
and the QoS routing control system feeds the current network communication flow historical data back to the GNs diagram predictor in a set period, and the GNs diagram predictor carries out off-line calculation again according to the received feedback data.
Preferably, the QoS route control system is configured to deploy a routing protocol policy algorithm to each node so that each node performs route processing according to the configuration parameters of the node;
the QoS route control system is set to record the original communication data of the nodes when the network is normal, the original communication data comprises the transmission time and the transmission energy consumption of effective data, and the original communication data is summarized and output to the GNs graph predictor periodically.
Preferably, the GNs graph predictor comprises an encoder, a processing unit, and a decoder;
the encoder is used for encoding the received original communication data from the QoS routing control system and carrying out normalization operation;
the processing unit comprises a plurality of serially connected graph neural network computing units, and each graph neural network computing unit is used as independent layer storage;
and the decoder is used for decoding the prediction result output by the processing unit and generating configuration strategy data used by the QoS route control system.
Preferably, the encoder is configured to perform normalization processing on the acquired raw communication data, and construct an input matrix for input to a subsequent processing unit, which is expressed as:
Figure GDA0003629697400000031
wherein, G represents an input matrix generated after encoding;
v denotes the original communication data actually received, Vi∈V；
E denotes a learning prediction vector, Ei,j∈E；
viRepresenting iteratively loaded learning functions, vi＝εv(ei)；
εvAnd εeThe vectors of the multilayer perceptrons with the same input matrix order;
thus, the input original communication data is normalized, and the original numerical values are replaced with the mean and the variance to generate an input matrix.
Preferably, the graph neural network computing unit in the arithmetic processor is configured to process graph structure data input, that is, the input matrix, wherein the graph neural network computing unit employs a space-based graph neural network computing unit, and the construction process includes:
firstly, a node set V and a link edge set E in an encoding output result G are constructed, wherein:
Figure GDA0003629697400000032
Figure GDA0003629697400000033
then, the weight between each node is:
Figure GDA0003629697400000034
obtaining the cumulative value of each node as:
Figure GDA0003629697400000035
then, the node prediction output is:
Figure GDA0003629697400000036
when a graph neural network computing unit is used for carrying out simulation prediction on a known system, namely a Qos routing control system, an effective iteration updating function needs to be constructed, and a model is converted into a model according to iteration training:
Figure GDA0003629697400000041
finally, the next state is updated according to the predicted acceleration using a semi-implicit Euler integral function, namely:
Figure GDA0003629697400000042
Figure GDA0003629697400000043
wherein Δ t is 1.
Preferably, the decoder is an inverse operation of the encoder, and is configured to convert a parameter matrix output by the GNs graph predictor into a form that can be read and configured by the Qos routing control system, which is specifically expressed as follows:
Figure GDA0003629697400000044
wherein, deltavRepresenting the real control variables in the y system, namely the QoS routing control system.
Therefore, the GNs graph predictor-based constraint routing control system according to the foregoing embodiment has significant advantages over the prior art in that:
1) the constraint routing control system based on the GNS graph predictor provided by the invention introduces the GNN artificial neural network, starts from a small-scale blocking mode, gradually expands the calculation level of the GNN artificial neural network, and realizes model research on multi-node large-scale blocking. Finding a propagation mode of local peak load in a theoretical infinite capacity node network from a data simulation level, and further constructing an optimal QoS constraint route calculation model based on the theoretical network propagation form;
2) a predictive computational model based on GNs graph simulators is proposed. For an infinite node network with infinite capacity, if a traditional SPH (smoothed Particle dynamics) algorithm is adopted, the calculation amount is very large, so that the invention introduces a GNs diagram predictor, the predictor model is used for simplifying the original simulation calculation process, the calculation efficiency is improved, and the data transmitted by different types of local peak loads in the network can be quickly simulated;
the peak load propagation condition of a multi-element heterogeneous network is further simulated on an ideal network, information transmission characteristics of different network structures in different load modes are obtained, and data basis is provided for predicting future flow change according to the current flow mode condition detected by an actual network node;
3) based on the simulation data, a prediction control flow is constructed, when network flow similar to the simulation prediction characteristic points appears on some nodes in the network, the original network routing control algorithm is immediately changed to a new network routing, so that the local peak load is controlled in a limited range, and the normal operation of a new-structure network formed by other nodes can be ensured.
Therefore, the invention simulates and calculates the node data transmission condition under the condition of local node instantaneous peak load through the GNs network, and constructs effective local isolation and route change according to the condition so as to control the instantaneous peak load in a small range and avoid the impact and the harm to the overall communication network.
It should be understood that all combinations of the foregoing concepts and additional concepts described in greater detail below can be considered as part of the inventive subject matter of this disclosure unless such concepts are mutually inconsistent. In addition, all combinations of claimed subject matter are considered a part of the presently disclosed subject matter.
The foregoing and other aspects, embodiments and features of the present teachings will be more fully understood from the following description taken in conjunction with the accompanying drawings. Additional aspects of the present invention, such as features and/or advantages of exemplary embodiments, will be apparent from the description which follows, or may be learned by practice of specific embodiments in accordance with the teachings of the present invention.
Drawings
The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component that is illustrated in various figures may be represented by a like numeral. For purposes of clarity, not every component may be labeled in every drawing. Embodiments of various aspects of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which:
FIG. 1 is a functional block diagram of a GNs graph predictor based constrained routing control system of the present invention.
FIG. 2 is a schematic diagram of the GNs graph predictor of the present invention.
Detailed Description
In order to better understand the technical content of the present invention, specific embodiments are described below with reference to the accompanying drawings.
In this disclosure, aspects of the present invention are described with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. Embodiments of the present disclosure are not necessarily intended to encompass all aspects of the invention. It should be appreciated that the various concepts and embodiments described above, as well as those described in greater detail below, may be implemented in any of numerous ways, as the disclosed concepts and embodiments are not limited to any one implementation. Additionally, some aspects of the present disclosure may be used alone, or in any suitable combination with other aspects of the present disclosure.
In the constraint routing control system based on the GNs graph predictor, a GNs graph predictor is added on the basis of the existing network routing system architecture, the GNs graph predictor adopts a GNN artificial neural network to construct a GNs graph prediction framework, peak traffic propagation modes under multiple nodes of different routing strategies are simulated through the GNs graph predictor, a routing control strategy combination with optimal QoS constraint is selected, and corresponding peak traffic characteristics and the strategy combination scheme are integrated into the network routing system, so that the routing control strategies can be automatically switched when similar network peak modes are met, network routing is dynamically adjusted, network load balance is realized, and the optimal routing control effect under local node peak traffic is achieved.
The exemplary GNs graph predictor-based constrained routing control system shown in connection with fig. 1 and 2 includes a QoS routing control system and a GNs graph predictor.
And the QoS route control system is used for carrying out route forwarding control on each node in the network. In a specific embodiment, the QoS route control system is a QoS route control system for each node in the network, which performs overall management on the route of each node.
The GNs graph predictor is used for simulating the dynamic change of the communication flow of the whole network when the single node or the multiple nodes in different network nodes have instantaneous peak load, and providing a prejudgment standard for the QoS route control system according to the change of different communication flows.
Preferably, the GNs prediction model may be calculated off-line, and the optimal policy is processed in the calculation result set in cooperation with the prediction of each grid node, and the policy is delivered to each node of the network.
The GNs graph predictor is independent of the QoS route control system, offline calculation of network communication flow is carried out, offline calculation results are fed back to the QoS route control system, and implementation efficiency of current QoS routes is improved.
And the QoS routing control system feeds the current network communication flow historical data back to the GNs diagram predictor in a set period, and the GNs diagram predictor carries out off-line calculation again according to the received feedback data.
Preferably, the QoS route control system is configured to deploy a routing protocol policy algorithm to each node so that each node performs route processing according to the configuration parameters of the node;
the QoS route control system is set to record the original communication data of the nodes when the network is normal, the original communication data comprises the transmission time and the transmission energy consumption of effective data, and the original communication data is summarized and output to the GNs graph predictor periodically.
At present, aiming at the fluctuating temperature caused by local information impact in a network, a local simulation mode is mainly adopted for adjustment in the prior art, but a global calculation mode is not adopted, because the influence of the global calculation is relatively large. Therefore, there are limitations on the influence of the propagation attribute and the channel bandwidth of each node in the calculation process, and these limitations cause a large deviation between the simulation data and the real data, which is determined by the simulation process and the result and cannot play a predictive effect in the actual operation effort.
In the implementation process, the graph network predictor units are introduced, a plurality of graph network predictor units are connected in series and grouped to construct, the larger calculation problem can be quickly adapted only by structural expansion, the calculation efficiency bottleneck is solved, and the structure of each single graph network predictor unit is simpler; and the network mode can be fully utilized, the flow can be parallelized, and only one time of encoding processing and prediction result decoding processing need to be carried out on the samples of the Qos routing control system.
It should be understood that GNs used in the present invention are abbreviations of Graph Network-based architectures (GNs), different Graph Network models can be constructed based on different Graph theories, and the present invention is directed to the Network transmission problem that a local node in the same wireless Network generates burst super traffic communication, and the node and connectivity are only traffic limited without other constraint requirements, so that the GNs are abbreviated as combining with the serial connection of a plurality of GN small units.
With reference to the prediction process examples shown in fig. 1 and 2, after iterative learning, the GNs graph predictor can quickly learn characteristics of network congestion in different modes, so as to make quick prediction according to the communication condition of the existing network, and select an effective policy to configure to the Qos routing control system.
As shown in fig. 2, the GNs graph predictor includes an encoder, a processing unit, and a decoder; the encoder is used for encoding the received original communication data from the QoS routing control system and carrying out normalization operation; the processing unit comprises a plurality of serially connected graph neural network computing units, and each graph neural network computing unit is stored as an independent layer; and the decoder is used for decoding the prediction result output by the processing unit and generating configuration strategy data used by the QoS route control system.
Preferably, the encoder is configured to normalize the acquired raw communication data, and construct an input matrix for input to a subsequent processing unit, which is expressed as:
Figure GDA0003629697400000071
wherein G represents an input matrix generated after encoding;
v denotes the original communication data actually received, Vi∈V；
E denotes a learning prediction vector, Ei,j∈E；
viRepresenting iteratively loaded learning functions, vi＝εv(ei)；
εvAnd εeThe vectors of the multilayer perceptrons with the same input matrix order are input;
thus, the input original communication data is normalized, and the original numerical values are replaced with the mean and the variance to generate an input matrix.
Preferably, the graph neural network computing unit in the arithmetic processor is configured to process graph structure data input, that is, the input matrix, wherein the graph neural network computing unit employs a space-based graph neural network computing unit, and the construction process includes:
firstly, a node set V and a link edge combination E in an encoding output result G are constructed, wherein:
Figure GDA0003629697400000072
Figure GDA0003629697400000073
then, the weight between each node is:
Figure GDA0003629697400000074
obtaining the cumulative value of each node as:
Figure GDA0003629697400000081
then, the node prediction output is:
Figure GDA0003629697400000082
when a graph neural network computing unit is used for carrying out simulation prediction on a known system, namely a Qos routing control system, an effective iteration updating function needs to be constructed, and a model is converted into a model according to iteration training:
Figure GDA0003629697400000083
finally, the next state is updated according to the predicted acceleration using a semi-implicit Euler integral function, namely:
Figure GDA0003629697400000084
Figure GDA0003629697400000085
wherein Δ t is 1.
Preferably, the decoder is an inverse operation of the encoder, and is configured to convert a parameter matrix output by the GNs graph predictor into a form that can be read and configured by the Qos routing control system, which is specifically expressed as follows:
Figure GDA0003629697400000086
wherein, deltavRepresenting the real control variables in the y system, namely the Qos route control system.
In the process of implementing the invention, as shown in conjunction with figures 1 and 2
Figure GDA0003629697400000087
The location is the state of the traffic in the node during the actual routing of the network traffic. If K is defined as a time parameter, then
Figure GDA0003629697400000088
Defined as a sequence of undulating state trajectories.
Figure GDA0003629697400000091
Thus, a simulator s can be constructed:
Figure GDA0003629697400000092
prediction of future states is achieved by modeling the dynamics of past states.
It can give a calculation that can simulate a future trajectory model, expressed as:
Figure GDA0003629697400000093
the trajectory sequences of these dynamics are obtained by iterative operations, and
Figure GDA0003629697400000094
the result of one of the iterative operations is indicated.
Therefore, we use the change in system state as dθ：
Figure GDA0003629697400000095
This dynamic differential dust prevention is described. The graph predictor of the invention utilizes the self-learning capability of the graph neural network to continuously approach an ideal equation set, thereby realizing the simulation of the actual situation.
The self-learning process of the graph predictor can be expressed as:
Figure GDA0003629697400000096
wherein s isθIs a trained self-learning simulator, namely a small unit of a graph predictor in the invention;
Figure GDA0003629697400000097
is the kth training iteration update function;
dθthe system is an ideal equation set to be learned, and is an actual network communication parameter in a normal communication state in the invention;
theta is an internal parameter that can be adjusted by numerical training;
and ← represents one useful training update mechanism, in the present invention an euler integrator is used.
Based on the method, the spatial method proposed by Gori is adopted to construct a spatial-based graph neural network computing unit. Defining the Node Velocity of the input network as the average Velocity between the current time step and the last time step, and the flow difference is
Figure GDA0003629697400000098
(constant Δ t omitted for simplicity), also drop, the average acceleration calculated between the next and current time step is targeted:
Figure GDA0003629697400000099
thus, the target acceleration is obtained as:
Figure GDA00036296974000000910
with the above embodiments, the present invention adopts the structures of the input encoder and the output decoder, and the data in the neural network computing unit is normalized, so that the learning process is evaluated by using a dimensionless learning gradient in the present case, and a priority gradient computing formula is constructed:
Figure GDA0003629697400000101
wherein η is the learning rate of the network;
Figure GDA0003629697400000102
is the deviation between the predicted data and the actual data after the network training, and the invention uses the vector variance mean value to calculate.
The network learning rate in the model is an important factor for improving the learning efficiency of the network unit. If the learning rate is low, the convergence time of the objective function needs longer time, and if the learning rate is too low, the parameter change may oscillate, and the objective function may miss some minimum values and be less likely to converge. We therefore use the following formula for processing:
Figure GDA0003629697400000103
wherein j is the number of learning iterations;
η (j) is the learning rate of the optimization objective model parameter θ;
thus, the learning rate of the surface convergence condition ranges from 0.001 to 0.1.
In the implementation we use augmentationThe sample noise is added into training to improve the network learning level and the generalization level of the intensifier. During training, random walk noise is used
Figure GDA0003629697400000104
To disturb the original number of samples.
To maintain data consistency, guarantees are needed
Figure GDA0003629697400000105
From which we randomly sampled different sets and added random noise
Figure GDA0003629697400000106
The incoming interference training track is used for calculating the flow acceleration of the node
Figure GDA0003629697400000107
And calculating the predicted acceleration L of the flow of each node after the impact2I.e.:
Figure GDA0003629697400000108
in the following, in combination with actual operation of an engineering network, in a network including N nodes, V is SET as a source node of a peak load at a certain time, and when a routing policy is refreshed for the first time, the V node is added into an active SET ACTIV _ SET, and a data packet carrying an address and synchronization information is broadcast to its single-hop neighbor nodes (B, D, E, F) within a communication range R, and after receiving the data packet, the V node returns a data packet containing its own address and synchronization information, and after receiving the data packet, the V node returns an acknowledgement character packet to inform the B, D, E, F that the active node SET is not in the peak load at present, and the nodes can mutually and normally transmit information with each other, and the time delay between each node is fixed.
The above-described confirmation action is performed every m seconds.
When one of the V nodes exceeds the preset peak flow, the V node rapidly compares the flow with the prediction model from the local node, selects an action from the processing strategy and sends the action as a first forwarding command to the nodes B, D, E and F.
When the node B gets the instruction and enters an emergency processing mode immediately, the current flow model of the node B is compared, and a data packet transmitted by the node V is shielded, so that the information of the node V is prevented from impacting the node B.
And meanwhile, according to the previous prediction model, limiting the flow packets of the nodes D, E and F by a threshold value so as to avoid the influence of excessive data sets transmitted from other nodes in the local area on the transmission of the nodes.
This is also true for the other nodes in the ACTIVE SET ACTIVE _ SET. And simultaneously carrying out early warning report with the surrounding nodes. Then the nodes V, B, C, D, E and F are isolated from the large network, and the data packets in the nodes are quickly stored and encapsulated in the local nodes. After the nodes of the external network receive the early warning signal, the original QoS routing table is refreshed again according to the original prediction model, and the original route passing through the nodes with the problems is transferred to other nodes.
And when the data flow on the nodes V, B, C, D, E and F is reduced to a preset value, the load of the nodes falls back to the credible interval. And at the moment, a recovery instruction is still sent by the V node, and transmission is confirmed twice by nodes one by one according to V- > B- > C- > D- > E- > F- > V- > B- > C- > D- > E- > F. After all the nodes confirm twice, the instruction can confirm that the current load overload phenomenon is finished. At this moment, each node of the network sends a cancel instruction outwards again. The shielded nodes are connected back to the original network by the external network, and the QoS routing system is recovered to be normal.
When the load is abnormal, the direct downtime of the V node can occur. If any node of B, C, D, E and F fails to receive the normal polling message of the V node within the specified time, the peak overload mode is directly entered, and the operation is repeated until the accident situation is solved.
The system of the invention can be effectively applied to the Internet, in particular to a 5G wireless network. Because the independent GNs predictor and the offline computing mode are adopted, the method is easier to integrate into the existing network routing control system. The invention can provide more intelligent guarantee for the actual operation network system, avoid the problem of large-area network overload caused by local instantaneous peak flow and improve the overall network stability.
For example, in a large-scale online activity, a large number of users are rushed to send information to one or more wireless network nodes, and if the information is not blocked in time, the operation of the nodes in the area is affected. Simply disconnecting the local area network from the outside completely affects both network transmission at the active site and other network transmissions that would otherwise be required through the active site nodes. The system can effectively predict the influence range of the node peak flow, predict the influence effect of the flow on the external network, and automatically distinguish the active local area network from the external network in advance. The node routing in the external network is automatically adjusted, and the node of the local network is skipped, so that the overall network carrying is not influenced. The peak load in the local network of the activity runs in the local network of the activity, and the network requirement of the activity is not influenced. When the activity is finished, the system automatically connects the local network nodes back to the backbone network according to the load condition, and the original node routing of the network is recovered.
When a certain node in the network is attacked by a hacker, a large amount of information is generated, and an external instantaneous peak flow is formed. The system can determine the diffusion mode to the instantaneous peak flow according to the GNs prediction model, and quickly isolate the nodes associated with the diffusion mode from the backbone network. Meanwhile, a routing system of the network is refreshed, the node routing in the external network is automatically adjusted, the active local nodes are skipped, and the overall network carrying is not influenced. After the information load of the hacker attack is eliminated, the system automatically replies to the nodes of the hacker attack and refreshes the route.
Although the present invention has been described with reference to the preferred embodiments, it is not intended to be limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims (3)

1. A constraint route control system based on a GNs graph predictor is characterized by comprising a QoS route control system and the GNs graph predictor;
the QoS route control system is used for carrying out route forwarding control on each node in the network;
the GNs graph predictor is used for simulating the dynamic change of the communication flow of the whole network when a single node or a plurality of nodes in different network nodes have instantaneous peak load and providing a prejudgment standard for the QoS route control system according to different communication flow changes;
the GNs graph predictor is independent of a QoS route control system, offline calculation of network communication flow is carried out, offline calculation results are fed back to the QoS route control system, and implementation efficiency of current QoS routes is improved;
and the QoS routing control system feeds the current network communication flow historical data back to the GNs diagram predictor in a set period, and the GNs diagram predictor carries out off-line calculation again according to the received feedback data.
2. A GNs graph predictor-based constraint routing control system according to claim 1, wherein the QoS routing control system is configured to deploy routing protocol policy algorithms to the respective nodes to cause each node to route according to the node's configuration parameters;
the QoS route control system is set to record the original communication data of the nodes when the network is normal, the original communication data comprises the transmission time and the transmission energy consumption of effective data, and the original communication data is summarized and output to the GNs graph predictor periodically.
3. The GNs-graph-predictor-based constrained routing control system of claim 1 or 2, wherein the GNs-graph predictor comprises an encoder, a processing unit, and a decoder;
the encoder is used for encoding the received original communication data from the QoS routing control system and carrying out normalization operation;
the processing unit comprises a plurality of serially connected graph neural network computing units, and each graph neural network computing unit is used as independent layer storage;
and the decoder is used for decoding the prediction result output by the processing unit and generating configuration strategy data used by the QoS route control system.
CN202011211827.9A 2020-11-03 2020-11-03 Constraint route control system based on GNs graph predictor Active CN112399483B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011211827.9A CN112399483B (en) 2020-11-03 2020-11-03 Constraint route control system based on GNs graph predictor

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011211827.9A CN112399483B (en) 2020-11-03 2020-11-03 Constraint route control system based on GNs graph predictor

Publications (2)

Publication Number Publication Date
CN112399483A CN112399483A (en) 2021-02-23
CN112399483B true CN112399483B (en) 2022-07-15

Family

ID=74598109

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011211827.9A Active CN112399483B (en) 2020-11-03 2020-11-03 Constraint route control system based on GNs graph predictor

Country Status (1)

Country Link
CN (1) CN112399483B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114580252A (en) * 2022-05-09 2022-06-03 山东捷瑞数字科技股份有限公司 Graph neural network simulation method and system for fluid simulation

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109688056A (en) * 2018-12-07 2019-04-26 南京理工大学 Intelligent Network Control System and method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7864751B2 (en) * 2007-08-14 2011-01-04 Greenberg Albert G Traffic engineering method, system and computer program product for managing traffic over dynamic networks during both normal and unexpected traffic scenarios

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109688056A (en) * 2018-12-07 2019-04-26 南京理工大学 Intelligent Network Control System and method

Also Published As

Publication number Publication date
CN112399483A (en) 2021-02-23

Similar Documents

Publication Publication Date Title
Zhou et al. SDN-based application framework for wireless sensor and actor networks
Zhang et al. Double-edge intelligent integrated satellite terrestrial networks
CN114567598B (en) Load balancing method and device based on deep learning and cross-domain cooperation
Maalej et al. QoS and energy aware cooperative routing protocol for wildfire monitoring wireless sensor networks
Zhang et al. Blockchain-based multi-access edge computing for future vehicular networks: A deep compressed neural network approach
Luo et al. Deep reinforcement learning based computation offloading and trajectory planning for multi-UAV cooperative target search
Moharamkhani et al. Multiobjective fuzzy knowledge‐based bacterial foraging optimization for congestion control in clustered wireless sensor networks
Tang et al. Blockchain-based trusted traffic offloading in space-air-ground integrated networks (sagin): A federated reinforcement learning approach
Ayub et al. Intelligent hello dissemination model for FANET routing protocols
CN114710437B (en) Internet of things edge network routing system combining block chains
CN114221691A (en) Software-defined air-space-ground integrated network route optimization method based on deep reinforcement learning
CN112399483B (en) Constraint route control system based on GNs graph predictor
Han et al. Time-varying topology model for dynamic routing in LEO satellite constellation networks
Kaur et al. A green hybrid congestion management scheme for IoT-enabled WSNs
Mahajan et al. Adaptive routing in wireless mesh networks using hybrid reinforcement learning algorithm
Anitha et al. DEQLFER—A Deep Extreme Q-Learning Firefly Energy Efficient and high performance routing protocol for underwater communication
Meng et al. Intelligent routing orchestration for ultra-low latency transport networks
Bhajantri et al. Fuzzy logic based cluster head selection and data processing in distributed sensor networks
CN112600827B (en) Virtual service migration method and system based on incremental minimum cost maximum flow
Liu et al. A routing model based on multiple-user requirements and the optimal solution
Li et al. Multi-Agent Federated DRL Enabled Resource Allocation for Air-Ground Integrated IoV Network
Suriya Praba et al. Energy measure semigraph-based connected edge domination routing algorithm in wireless sensor networks
Pradeep et al. An efficient energy consumption and delay aware autonomous data gathering routing protocol scheme using a deep learning mobile edge model and beetle antennae search algorithm for underwater wireless sensor network
Chen et al. A self-adaptive wireless sensor network coverage method for intrusion tolerance based on particle swarm optimization and cuckoo search
Kuze et al. Hierarchical optimal control method for controlling self-organized networks with light-weight cost

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB02 Change of applicant information
CB02 Change of applicant information

Address after: Building 10, No.19, ningshuang Road, Yuhuatai District, Nanjing City, Jiangsu Province, 210000

Applicant after: Jiahuan Technology Co.,Ltd.

Address before: Building 10, No.19, ningshuang Road, Yuhuatai District, Nanjing City, Jiangsu Province, 210000

Applicant before: Nanjing Jiahuan Technology Co.,Ltd.

GR01 Patent grant
GR01 Patent grant