Large-scale self-organizing network control system and method based on software-defined network

Abstract

The present invention discloses a large-scale ad hoc network control system and method based on software-defined networking. The system's data plane layer consists of large-scale nodes; the control plane layer consists of a link monitoring module, a data flow management module, a routing planning module, and a topology control module; and the application plane layer consists of a routing planning application and a topology control application. The link monitoring module is responsible for receiving the ad hoc network link status information reported in real time by the data plane layer and generating a traffic matrix; the data flow management module is responsible for managing all data flows in the ad hoc network; the routing planning module performs data flow routing planning based on the ad hoc network traffic matrix and information about all data flows; and the topology control module is responsible for developing the ad hoc network initial topology and reconstructing the topology when the inter-node connectivity status changes. The present invention improves on the standard SDN architecture by adding functional modules, utilizing node community information to control the topology, and optimizing the genetic algorithm to increase the convergence speed of routing solutions.

Description

Technical Field

[0001] The present invention relates to a self-organizing network control technology based on software-defined networks, and in particular to a large-scale self-organizing network control system and method based on software-defined networks. Background Art

[0002] Before the advent of SDN technology, ad hoc networks typically used traditional routing protocols such as OLSR and DSDV for communication. Because they used distributed routing planning methods, they had slow convergence speeds and were slow to perceive network changes, making it difficult to meet the Quality of Service (QoS) requirements of network applications.

[0003] SDN is a new network architecture that obtains global network information through a southbound interface (SBI) and provides data support to upper-layer applications through a northbound interface (NBI). The core concept of SDN is to achieve global network resource scheduling by separating control and forwarding. Distributed local routing planning methods in traditional network architectures suffer from slow convergence, long response times, and delayed perception of network changes. SDN, on the other hand, leverages its real-time access to global network link status for global routing planning, resulting in fast convergence, short response times, and sensitivity to network changes.

[0004] Currently, routing algorithms within the SDN network architecture are primarily categorized into two types: heuristic algorithms and machine learning algorithms. Heuristic routing algorithms are characterized by their support for dynamically changing network topologies and communication requirements. However, they are slow to solve in large-scale networks and struggle to meet the real-time requirements of routing planning. Machine learning routing algorithms are characterized by their ability to support model training for large-scale networks. However, the network topology and communication requirements must be predefined prior to model training, making them difficult to support dynamically changing network topologies and communication requirements. Therefore, existing SDN routing algorithms cannot be directly applied to large-scale ad hoc networks. There is an urgent need to develop efficient routing planning algorithms that can adapt to these characteristics, improve the communication quality of ad hoc networks, and provide communication support for future information warfare.

[0005] In the standard SDN architecture, nodes initiate data flow routing requests through packetIn messages. The SDN service plans routes for the data flow based on current link information and QOS weight preferences. This is a passive routing planning method that cannot comprehensively consider the QOS performance and network load balancing of all data flows.

[0006] Standard SDN networks typically use TCP as the transport layer protocol for data flow transmission. Because TCP's congestion control mechanism aims to maximize available network bandwidth, it's difficult to rapidly plan routes using bandwidth reservation methods used in deterministic networks. Network applications in ad hoc networks all have specific QoS performance requirements, such as low latency for real-time communication, high bandwidth for file transfer, and low packet loss for audio and video. Routing algorithms within SDN architectures typically assign QoS weights to data flows of various service types to indicate their preference for latency, bandwidth, and packet loss. Link latency and packet loss are relatively constant. As long as the topology and network node locations remain unchanged, the latency and packet loss rates, and the QoS path for the data flow, are also assumed to remain unchanged. However, bandwidth is a special case. In ad hoc networks, TCP is typically used as the transport layer protocol for data flows, and TCP's congestion control strategy aims to maximize the throughput of the data flow by occupying as much available bandwidth as possible. Therefore, even if the link bandwidth remains unchanged, the path of the data flow may be replanned. Therefore, it is impossible to use the resource reservation method in the deterministic network to quickly plan the paths of all data flows centrally based on mathematical calculations.

[0007] Large-scale ad hoc networks are characterized by a massive number of nodes. In standard SDN architectures, the control plane lacks a topology control module. Failure to control the ad hoc network topology results in complex network topology, a massive number of links, and high routing complexity. Furthermore, heuristic topology reconstruction algorithms are inefficient, necessitating the urgent need to research efficient topology reconstruction algorithms within the SDN architecture. Summary of the Invention

[0008] The purpose of the present invention is to provide a large-scale self-organizing network control system and method based on software-defined networking, which performs topology control, routing planning and topology reconstruction on the self-organizing network.

[0009] The technical solution to achieve the objectives of the present invention is as follows: In the first aspect, the present invention provides a large-scale self-organizing network control system based on software-defined networking, wherein the data plane layer of the system is composed of large-scale nodes, the control plane layer is composed of a link monitoring module, a data flow management module, a routing planning module, and a topology control module, and the application plane layer is composed of a routing planning application and a topology control application;

[0010] The link monitoring module is used to receive link status information uploaded in real time by the data plane, including latency, bandwidth, and packet loss rate, and generate an ad hoc network traffic matrix;

[0011] The data flow management module defines bandwidth-limited QOS data flows dedicated to ad hoc networks, monitors data flow requests from nodes in the data plane, receives data flow parameters, and manages the data flow;

[0012] The routing planning module is used to call the routing planning application to perform centralized routing planning for all data flows based on the real-time traffic matrix of the ad hoc network and the information of all data flows. The routing planning application uses a genetic routing planning algorithm based on community constraints for routing planning.

[0013] The topology control module is used to call the topology control application to generate the initial topology of the ad hoc network and reconstruct the topology when the communication status between nodes changes.

[0014] Furthermore, the data plane layer is a large-scale self-organizing network composed of multiple nodes.

[0015] Furthermore, the data flow management module defines a bandwidth-limited QOS data flow, which expresses the performance preference for the network based on the delay preference weight, bandwidth preference weight, and packet loss preference weight, and at the same time limits the bandwidth occupancy upper limit of the data flow; the data flow management module maintains the information of all data flows in the network.

[0016] Furthermore, the routing planning module performs centralized routing planning based on the real-time traffic matrix and information of all data flows, with the comprehensive QOS performance of all data flows and load balancing of network nodes as the goal; when the traffic matrix changes, the number of data flows changes, or the topology is reconstructed, the module will re-plan the routing.

[0017] Furthermore, the topology control module constrains the topology construction between nodes based on the node community relationship, optimizes the MANET topology, and facilitates the use of the community-constrained genetic routing planning algorithm on the MANET; formulates the initial topology of the MANET and reconstructs the topology when the communication status between nodes changes.

[0018] Furthermore, the routing planning application uses a community-constrained genetic routing planning algorithm for routing planning. This algorithm uses node community information to optimize the gene crossover process in the genetic algorithm. The community discovery steps of this algorithm are:

[0019] Step 6-1: Input the node communication relationship matrix communication_matrix (N×N 0 / 1 square matrix), which is used to represent the communication connection status of each node in the network. The matrix element communication_matrix[i][j]=1 indicates that node i has a direct communication relationship with node j.

[0020] Step 6-2: Create an empty list community_matrices to store the community matrix generated subsequently;

[0021] Step 6-3: Input the node communication relationship matrix communication_matrix into the deepwork community detection algorithm to obtain the node grouping result; the algorithm output is a two-dimensional list communities, where each sublist contains a set of node numbers belonging to the same community;

[0022] Step 6-4: For each detected community, perform the following operations: ① Create an all-zero basis matrix: Generate an N×N zero matrix community_mask with the same dimensions as the input matrix; ② Set community connection identifiers: Traverse all node pairs (i, j) in the current community and set community_mask[i, j] to 1; ③ Matrix storage: Add the generated community identifier matrix to the community_matrices list;

[0023] Step 6-5: Output a set of K community matrices, community_matrices, where each matrix fully represents the internal communication topology of the corresponding community;

[0024] The derivation process of the fitness function of the genetic routing algorithm is:

[0025] Step 6-6, use an undirected graph To define large-scale ad hoc networks, Represents a node in the network The collection of ; Represents a link in the network The topology between two nodes defines a link, i.e. ;

[0026] Steps 6-7, use Represents the real-time bandwidth matrix of large-scale ad hoc networks, and defines Source node To adjacent nodes The real-time bandwidth is

[0027]

[0028] in represents a positive real number;

[0029] Steps 6-8, use Represents the real-time delay matrix of large-scale ad hoc networks, and defines For Link The delay is

[0030]

[0031] Steps 6-9, use Represents the real-time packet loss rate matrix of large-scale ad hoc networks, and defines For Link The packet loss rate is

[0032]

[0033] Steps 6-10, use represents the data flow in a large-scale ad hoc network, represents a collection of data streams, then

[0034]

[0035] is the source node of the data stream, For the destination node, 、 、 are the delay preference weight, bandwidth preference weight and packet loss preference weight of the data flow respectively. The throughput limit of the data flow;

[0036] Steps 6-11, use Represents the weight matrix for centralized path planning of all data flows in a large-scale ad hoc network, where Represents an edge The weight of

[0037]

[0038] Steps 6-12, use Represents the forwarding path set of the data flow, which is the weight matrix The corresponding set of k shortest paths (k=3, 3 paths represent the paths of delay preference, bandwidth preference, and packet loss preference data flows from node s to node d) on the weighted undirected graph is:

[0039]

[0040] Steps 6-13, use Indicates the forwarding path, For data flow in the path The bandwidth that can be reserved on For Link The remaining bandwidth is

[0041]

[0042] Steps 6-14, Indicates the total path delay of the data flow, Indicates a link The delay is

[0043]

[0044] Steps 6-15, Indicates the path packet loss rate of the data flow, Indicates a link The packet loss rate is

[0045]

[0046] Steps 6-16, represents the QOS performance score of the data flow, then

[0047]

[0048] Steps 6-17, use represents the flow betweenness set of large-scale ad hoc network nodes, For nodes The flow betweenness of

[0049]

[0050] Steps 6-18, use represents the sum of the flow betweenness of all nodes in a large-scale ad hoc network, then

[0051]

[0052] Step 6-19, use The total number of nodes is The average node flow betweenness in a large-scale ad hoc network is

[0053]

[0054] Steps 6-20, The total number of nodes is The arithmetic mean square error of node flow betweenness in a large-scale ad hoc network is:

[0055]

[0056] The smaller it is, the more evenly the flow betweenness is distributed, and the better the network load balancing and routing invulnerability are.

[0057] Step 6-21, represents the evaluation score of network load balancing and routing invulnerability in large-scale ad hoc networks, then

[0058]

[0059] Step 6-22, Objective Function of Centralized Routing Planning Algorithm for:

[0060]

[0061] Furthermore, the routing planning application uses a community-constrained genetic routing planning algorithm for routing planning. This algorithm uses node community information to optimize the gene crossover process in the genetic algorithm. The steps of the community-constrained genetic routing planning algorithm are given below:

[0062] Step 7-1: Input the ad hoc network topology, link parameter matrix, data flow information, node communication relationship matrix, population number, crossover rate, and mutation rate;

[0063] Step 7-2: Get node community information;

[0064] Step 7-3: Randomly generate the initial population (routing weight matrix) according to the population number;

[0065] Step 7-4: Calculate the population fitness value according to the fitness function;

[0066] Step 7-5: Determine whether the fitness function value converges. If so, output the routing weight matrix and exit. If not, continue with the following steps.

[0067] Step 7-6: Based on the crossover rate, use the tournament method to select individuals with fitness values ​​greater than the set threshold as parents;

[0068] Step 7-7: Perform gene crossover using node community information as a constraint to generate offspring;

[0069] Step 7-8: Mutate some individuals whose fitness values ​​are less than the set threshold according to the mutation rate;

[0070] Step 7-9: Generate a new population and go to step 7-4.

[0071] Furthermore, the topology control application uses a community-constrained topology control algorithm for topology construction. This algorithm constructs the initial topology of the self-organizing network based on the inter-node communication status and node community information. By merging the community matrices in the community matrix set and fusing the inter-node communication matrices, an initial fully connected graph is constructed. Then, redundant edges are randomly deleted to obtain an initial topology with a point connectivity of 2. This serves as the initial topology of the self-organizing network in the data plane layer. The following are the steps of the community-constrained topology control algorithm:

[0072] Step 8-1: Enter the community collection ,in is the community association matrix, N represents the number of nodes in the ad hoc network; the reachability matrix ,satisfy If and only if the node and Possess physical connection capability; network scale parameters Indicates the total number of nodes;

[0073] Step 8-2: Community integration matrix construction, defining the community joint matrix , generated by the following formula:

[0074]

[0075] Where ∨ represents the logical OR operation of each element;

[0076] Step 8-3: Initial topology generation, constructing the adjacency matrix corresponding to the candidate edge set:

[0077]

[0078] Where ⊙ is the Hadamard product operation; based on this, the initial fully connected graph is constructed:

[0079]

[0080] Among them, E0 is the set of edges of the constructed initial fully connected graph G0;

[0081] Step 8-4: Redundant edge pruning preprocessing, define the edge processing sequence as a set of randomly arranged edges:

[0082] is the random permutation function

[0083] Initialize a temporary topology:

[0084]

[0085] Step 8-5: Iterate the optimization process and perform connectivity check on each candidate edge:

[0086] ,implement

[0087] in, is any edge in the set of edges after all edges in the initial fully connected graph are randomly arranged; is the temporary topology G temp Delete Edge The resulting figure, For the picture The point connectivity of

[0088] Step 8-6: Output topology .

[0089] In a second aspect, the present invention provides a large-scale self-organizing network control method based on a software-defined network, based on the system implementation of the first aspect, the method comprises the following steps:

[0090] Step 1: The topology control module constructs a communication demand relationship matrix based on the communication demand relationship of the ad hoc network nodes in the data plane, uses the deepwork algorithm to mine node community information, and obtains a node community matrix set;

[0091] Step 2: The topology control module calls the topology construction application to randomly generate an ad hoc network topology with a point connectivity of 2 based on the inter-node connectivity matrix and the node community matrix set of the ad hoc network, as the initial topology of the ad hoc network;

[0092] Step 3: After the ad hoc network is activated, the link monitoring module receives link status information uploaded by the data plane in real time and builds a traffic matrix.

[0093] Step 4: The data flow management module receives the data flow request initiated by the self-organizing network node in real time, parses the parameters in the request, and uses the data flow initiating node, receiving node, QOS weight and bandwidth usage limit as data flow information to manage all data flow information;

[0094] Step 5: When the number of data flows changes, the link status changes, or the topology is reconfigured, the routing planning module calls the routing planning application to perform routing planning. The routing planning application uses the community-constrained genetic routing planning algorithm to solve the routing problem based on the data flow information provided by the data flow management module, the global link information provided by the link monitoring module, and the node community information.

[0095] Step 6: When the inter-node connectivity status changes, the topology control module calls the topology control application to reconfigure the topology. The topology control application uses a community-constrained topology control algorithm to reconfigure the topology based on the current inter-node connectivity matrix and the node community matrix set.

[0096] In a third aspect, the present invention provides an electronic device comprising an imaging device, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to the second aspect when executing the program.

[0097] Compared with existing technologies, this invention offers the following significant advantages: It improves upon the traditional SDN architecture by incorporating a data flow management module and a topology control module. Compared to the traditional SDN passive routing planning method based on packetIn messages, which targets the QOS of a single flow, this SDN architecture's routing planning module uses a centralized, proactive routing planning algorithm that targets the combined QOS performance of all data flows and node load balancing. The "community-constrained genetic routing planning algorithm" converges faster than conventional genetic algorithm routing planning algorithms. The ad hoc network topology control algorithm utilizes node community relationships to control node networking, optimizing the topology to facilitate routing planning using the "community-constrained genetic routing planning algorithm." BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 This is a standard SDN system architecture diagram.

[0099] Figure 2 This is an architectural diagram of the SDN system of the present invention.

[0100] Figure 3 It is the community discovery process of nodes in ad hoc networks.

[0101] Figure 4 The process of initializing the SDN topology control module to the self-organizing network topology.

[0102] Figure 5 Determine the QOS data flow for the bandwidth in the ad hoc network defined by the flow management module of this SDN system.

[0103] Figure 6 Schematic diagram of gene crossover in the "Genetic Routing Planning Algorithm Based on Community Constraints".

[0104] Figure 7 Flowchart of the parallel genetic routing planning algorithm based on community attention mechanism. DETAILED DESCRIPTION

[0105] This invention proposes a large-scale self-organizing network control system and algorithm based on a software-defined network (SDN). The large-scale self-organizing network control system based on a software-defined network has a data plane layer composed of large-scale nodes, a control plane layer composed of a link monitoring module, a data flow management module, a routing planning module, and a topology control module, and an application plane layer composed of a routing planning application and a topology control application. The link monitoring module is responsible for receiving the self-organizing network link status information reported in real time by the data plane layer and generating a traffic matrix. The data flow management module is responsible for managing all data flows in the self-organizing network. The routing planning module performs data flow routing based on the self-organizing network traffic matrix and information on all data flows. The topology control module is responsible for developing the initial topology of the self-organizing network and reconstructing the topology when the connectivity status between nodes changes. This SDN system is designed with two algorithms: ① The routing planning application uses a "community-constrained genetic routing planning algorithm" for routing planning; ② The topology control application uses a "community-constrained topology control algorithm" for topology construction. The present invention fully considers the differences between large-scale self-organizing networks and ordinary self-organizing networks. To address the problems of large number of nodes, complex topology, and high time complexity of routing solution, the standard SDN architecture is first improved by adding functional modules. Secondly, the node community information is used to control the topology, and the genetic algorithm is optimized to improve the convergence speed of routing solution.

[0106] The data flow management module defines a bandwidth-determined QOS data flow, whose QOS preference weights include delay preference weight, bandwidth preference weight and packet loss preference weight, which limit the bandwidth occupancy upper limit of the data flow; the data flow management module maintains information on all data flows in the network, including data flow sending nodes, receiving nodes, QOS preferences and bandwidth occupancy upper limit.

[0107] The routing module calls the routing application to perform routing planning. Routing is re-planned when the traffic matrix changes, the number of data flows changes, or the topology is restructured. The routing application uses a "community-constrained routing algorithm" to perform centralized routing planning for all data flows.

[0108] The topology control module calls the topology control application to perform topology control. It constrains the topology construction between nodes based on node community relationships to reduce the number of links. It also establishes the initial topology for the ad hoc network and reconstructs the topology when the inter-node connectivity status changes. The topology control application uses the "Tactical Ad Hoc Network Topology Control Algorithm" to construct the ad hoc network topology.

[0109] Based on the above system, the present invention also proposes a large-scale self-organizing network control method based on software-defined networking, comprising the following steps:

[0110] Step 1: The topology control module constructs a communication demand relationship matrix based on the communication demand relationship of the ad hoc network nodes in the data plane, and uses the deepwork algorithm to mine the node community information to obtain the community matrix set.

[0111] Step 2: The topology control module calls the topology control application and uses the ad hoc network topology control algorithm to generate an ad hoc network topology with a point connectivity of 2 based on the community matrix set and the node inter-connectivity matrix as the initial topology of the ad hoc network.

[0112] Step 3: Open the ad hoc network. The link monitoring module receives the link status information uploaded by the data plane in real time and builds a traffic matrix.

[0113] Step 4: The data flow management module receives the data flow request initiated by the self-organizing network node in real time, parses the parameters in the request, and uses the data flow initiating node, receiving node, QOS weight and bandwidth usage limit as data flow information to manage all data flow information.

[0114] Step 5: When the number of data flows changes, the link status changes, or the topology is reconfigured, the routing module calls the routing application to perform routing. The routing application uses a community-constrained genetic routing algorithm to solve routing problems based on the data flow information provided by the data flow management module, the global link information provided by the link monitoring module, and node community information.

[0115] Step 6: When the connectivity between nodes changes, the topology control module calls the topology control application to reconfigure the topology. The topology control application uses a community-constrained topology control algorithm to reconfigure the topology.

[0116] The method of the present invention is described in detail below with reference to the accompanying drawings.

[0117] Combine Figure 1A standard SDN system consists of a data plane, a control plane, and an application plane. The data and control planes communicate via southbound interfaces, while the control and application planes communicate via northbound interfaces. The control plane primarily consists of a packetIn message monitoring module, a link monitoring module, and a routing management module. The packetIn message monitoring module listens for packetIn requests initiated by nodes in the data plane. The link monitoring module collects link status information and generates a real-time traffic matrix, communicating with nodes in the data plane via the LLDP protocol. The application plane performs routing planning based on the source node, destination node, and QoS preference weight in the packetIn request and the real-time traffic matrix provided by the link monitoring module. The routing entries are then distributed to the routing management module in the control plane, which in turn distributes them to nodes in the data plane. Route planning in a standard SDN system is a passive, one-sided approach that only considers the QoS performance requirements of the current data flow. Standard SDN systems lack a data flow management module, preventing them from obtaining information about all data flows across the network. Routing planning is based on the overall QoS performance of all data flows and network load balancing. The standard SDN system does not have a topology control module and cannot control the topology of the ad hoc network. Therefore, any two nodes can build a topology through wireless communication, resulting in complex ad hoc network topology, a huge number of links, and high complexity of routing solution time.

[0118] Combine Figure 2 This SDN system improves upon the standard SDN architecture by adding a data flow management module and a topology control module to the control plane layer. The data flow management module's function is as follows: when a data flow transmission is initiated between nodes, the sender registers with the data flow management module, providing information about the data flow's initiating node, receiving node, QoS weight, and bandwidth usage limit. The data flow management module then manages all data flow information within the ad hoc network. Based on the data flow information provided by the data flow management module, the routing management module performs centralized routing planning, targeting overall QoS performance and network load balancing for all data flows. Routing is replanned upon changes in link status or topology. The topology control module controls the ad hoc network topology, reducing the number of links and optimizing the network topology. It creates an initial topology for the ad hoc network and reconstructs the topology when node connectivity changes. The application plane layer of this SDN system runs a routing planning application and a topology control application. The routing management module invokes the routing planning application to plan routes using a community-constrained genetic routing planning algorithm. The topology control module invokes the topology control application to create an initial topology for the ad hoc network and constructs the topology using a community-constrained topology control algorithm.

[0119] Combine Figure 3In ad hoc networks, a large number of community structures exist. A community is a collection of nodes with close communication needs within the network. In large-scale ad hoc networks, nodes are typically divided by level and clustered to complete operational tasks. For example, in a certain network, there are three types of communication nodes, Y, YL, and L, with predefined communication requirements: nodes of the same type can communicate with each other, and Y and L can communicate with YL, but there is no need for direct communication. The node communication requirement relationship graph can be further converted into a node communication relationship matrix, which can then be used to discover communities using traditional community discovery algorithms such as the random walk (deepwork) algorithm. This SDN system constructs the node communication relationship matrix matrixCom based on the node communication relationship graph and uses the random walk algorithm deepwork to obtain node community information nodeCommunityInfo.

[0120] Combine Figure 4 ,When the SDN system is initialized, the topology control module calls the ,topology control application and constructs the initial fully connected ,graph through the set of community relationship matrices in the nodeCommunityInfo and the ,inter-node accessibility matrix, and then randomly deletes redundant edges to obtain ,an initial topology with a point connectivity of 2, which serves as the ,initial topology of the self-organizing network in the data ,plane layer.

[0121] Combine Figure 5 To address the drawback of traditional QoS data flows, which cannot be centrally routed using resource reservation, the data flow management module of this SDN system defines a bandwidth-determined QoS data flow for use in ad hoc networks. Ad hoc networks primarily include three network applications: messaging, file transfer, and audio and video transmission, corresponding to latency-sensitive, bandwidth-sensitive, and packet loss-sensitive services, respectively. Unlike bandwidth-intensive network applications in data centers and other networks, these three types of network applications typically have throughput limits. Therefore, based on traditional QoS data flows, this SDN system draws on the concept of bandwidth limit determination for data flows in deterministic networks and designs a new bandwidth-determined QoS data flow for ad hoc networks. Specifically, this SDN system retains the QoS weight parameters of data flows, allowing the SDN system to still plan the most appropriate QoS paths for different types of data flows. At the same time, bandwidth limits are set for all three types of data flows, allowing the SDN system to leverage the concept of resource reservation in deterministic networks and perform centralized QoS path planning for all data flows based on mathematical calculations.

[0122] Combine Figure 6When link status changes, the number of data flows changes, or the topology is reconfigured, the routing module invokes the routing application and uses the "community-constrained genetic routing algorithm" to perform centralized routing for all data flows. Specifically, the genes in the genetic algorithm's solution space undergo parallel crossover evolution based on community relationship constraints. This significantly reduces meaningless gene crossovers and increases the speed of solution evolution, significantly accelerating the algorithm's convergence. The specific implementation steps of the "community-constrained genetic routing algorithm" are:

[0123] The community discovery steps of the genetic routing algorithm are:

[0124] Step 6-1: Input the node communication relationship matrix communication_matrix, which is an N×N 0 / 1 square matrix used to represent the communication connection status of each node in the network. The matrix element communication_matrix[i][j]=1 indicates that node i has a direct communication relationship with node j;

[0125] Step 6-2: Create an empty list community_matrices to store the community matrix generated subsequently;

[0126] Step 6-3: Input the node communication relationship matrix communication_matrix into the deepwork community detection algorithm to obtain the node grouping results. The algorithm outputs a two-dimensional list communities, where each sublist contains a set of node numbers belonging to the same community. For example, when two communities are detected, the output may be [[0,1], [2,3]], indicating that nodes 0-1 constitute the first community and nodes 2-3 constitute the second community.

[0127] Step 6-4: For each detected community, perform the following operations: ① Create an all-zero basis matrix: Generate an N×N zero matrix community_mask with the same dimensions as the input matrix; ② Set community connection identifiers: Traverse all node pairs (i, j) in the current community and set community_mask[i, j] to 1; ③ Matrix storage: Add the generated community identifier matrix to the community_matrices list;

[0128] Step 6-5: Output a set of K community matrices, community_matrices, where each matrix fully represents the internal communication topology of the corresponding community;

[0129] The derivation process of the fitness function of the genetic routing algorithm is:

[0130] Step 6-6, use an undirected graph To define large-scale ad hoc networks, Represents a node in the network The collection of ; Represents a link in the network The topology between two nodes defines a link, i.e. .

[0131] Steps 6-7, use Represents the real-time bandwidth matrix of large-scale ad hoc networks, and defines Source node To adjacent nodes The real-time bandwidth is

[0132]

[0133] Steps 6-8, use Represents the real-time delay matrix of large-scale ad hoc networks, and defines For Link The delay is

[0134]

[0135] Steps 6-9, use Represents the real-time packet loss rate matrix of large-scale ad hoc networks, and defines For Link The packet loss rate is

[0136]

[0137] Steps 6-10, use represents the data flow in a large-scale ad hoc network, represents a collection of data streams, then

[0138]

[0139] is the source node of the data stream, For the destination node, 、 、 are the delay preference weight, bandwidth preference weight and packet loss preference weight of the data flow respectively. The throughput limit of the data flow.

[0140] Steps 6-11, use Represents the weight matrix for centralized path planning of all data flows in a large-scale ad hoc network, where Represents an edge The weight of

[0141]

[0142] Steps 6-12, use Represents the forwarding path of the data flow, which is the weight matrix The shortest path on the corresponding weighted undirected graph is

[0143]

[0144] Steps 6-13, For data flow in the path The bandwidth that can be reserved on For Link The remaining bandwidth is

[0145]

[0146] Steps 6-14, Indicates the total path delay of the data flow, Indicates a link The delay is

[0147]

[0148] Steps 6-15, Indicates the path packet loss rate of the data flow, Indicates a link The packet loss rate is

[0149]

[0150] Steps 6-16, represents the QOS performance score of the data flow, then

[0151]

[0152] Steps 6-17, use represents the flow betweenness set of large-scale ad hoc network nodes, For nodes The flow betweenness of

[0153]

[0154] Steps 6-18, use represents the sum of the flow betweenness of all nodes in a large-scale ad hoc network, then

[0155]

[0156] Step 6-19, use The total number of nodes is The average node flow betweenness in a large-scale ad hoc network is

[0157]

[0158] Steps 6-20, The total number of nodes is The arithmetic mean square error of node flow betweenness in a large-scale ad hoc network is:

[0159]

[0160] The smaller it is, the more evenly the flow betweenness is distributed, and the better the network load balancing and routing anti-destruction performance are.

[0161] Step 6-21, represents the evaluation score of network load balancing and routing invulnerability in large-scale ad hoc networks, then

[0162]

[0163] In step 6-22, the objective function of the centralized routing planning algorithm (i.e., the fitness function of the genetic algorithm) is:

[0164]

[0165] The population of the genetic algorithm is evolved using a gene crossover method based on community constraints to continuously generate better solutions until the fitness function value converges. The gene crossover steps based on community relationship constraints in the genetic routing planning algorithm are:

[0166] Step 6-23: Input parameters: Communities = {C1, C2, ..., C k}: community matrix set, containing K N×N 0 / 1 matrices, each matrix represents the internal connection structure of a community; W_parent1 ∈ [0,1]^{N×N}: parent routing weight matrix 1; W_parent2 ∈ [0,1]^{N×N}: parent routing weight matrix 2; G = (V, E): network topology graph data structure, V is the node set, E is the edge set;

[0167] Step 6-24: Initialize the child matrix and create an N×N zero matrix W_child as the basic container;

[0168] Step 6-25: Community traversal processing, perform the following operations on each community matrix C in the Communities set:

[0169] Generate a binary mask matrix M corresponding to the current community C, whose element definition rules are:

[0170] M[i][j] = 1 if and only if the following two conditions are met:

[0171] a) There is a physical connection between nodes i and j, that is, e i,j ∈ E;

[0172] b) Nodes i and j belong to the same community C, i.e., C[i][j] = 1;

[0173] Otherwise M[i][j] = 0

[0174] Step 6-26: Cross mask generation, create a random binary mask matrix X_mask, whose elements independently follow the Bernoulli distribution:

[0175] The probability of X_mask[i][j] = 1 is 50%, indicating that the gene of parent 1 is selected;

[0176] The probability of X_mask[i][j] = 0 is 50%, indicating that the gene of parent 2 is selected;

[0177] Step 6-27: Constrain the crossover operation and perform element-level fusion calculations, which specifically includes two parallel processing branches:

[0178] Parent 1 contribution: Hadamard product of W_parent1 elements and (M∧X_mask);

[0179] Parent 2 contribution: Hadamard product of W_parent2 elements and (M∧¬X_mask), where the ¬ symbol represents the logical NOT operator, which inverts the elements in the matrix (1 becomes 0, 0 becomes 1).

[0180] Add the two result matrices and update the corresponding community area of ​​the W_child matrix;

[0181] Step 6-28: Output the result W_child ∈ [0,1]^{N×N}, which is the child routing generated by community constraint crossover

[0182] Weight matrix.

[0183] By executing the community-constrained genetic routing planning algorithm and outputting the routing weight matrix, the routes of all data flows in the ad hoc network can be obtained, which are then sent to the data plane by the routing management module.

[0184] Combine Figure 7 The movement of nodes in an ad hoc network will cause the communication status between nodes to change. At this time, the topology control module should call the topology control application to reconstruct the topology. The topology control application uses the "topology control algorithm based on community constraints" to reconstruct the topology of the ad hoc network.

[0185] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A large-scale self-organizing network control system based on software-defined networking, characterized in that: The system's data plane layer consists of large-scale nodes, the control plane layer consists of link monitoring modules, data flow management modules, routing planning modules, and topology control modules, and the application plane layer consists of routing planning applications and topology control applications. The link monitoring module is used to receive link status information uploaded in real time by the data plane, including latency, bandwidth, and packet loss rate, and generate an ad hoc network traffic matrix; The data flow management module defines bandwidth-limited QOS data flows dedicated to ad hoc networks, monitors data flow requests from nodes in the data plane, receives data flow parameters, and manages the data flow; The routing planning module is used to call the routing planning application to perform centralized routing planning for all data flows based on the real-time traffic matrix of the ad hoc network and the information of all data flows. The routing planning application uses a genetic routing planning algorithm based on community constraints for routing planning. The genetic routing planning algorithm uses node community information to optimize the gene crossover process in the genetic algorithm. The community discovery steps of the genetic routing planning algorithm are as follows: Step 6-1: Input the node communication relationship matrix communication_matrix, which is an N×N 0 / 1 square matrix used to represent the communication connection status of each node in the network. The matrix element communication_matrix[i][j]=1 indicates that node i has a direct communication relationship with node j; Step 6-2: Create an empty list community_matrices to store the community matrix generated subsequently; Step 6-3: Input the node communication relationship matrix communication_matrix into the deepwork community detection algorithm to obtain the node grouping result; the algorithm output is a two-dimensional list communities, where each sublist contains a set of node numbers belonging to the same community; Step 6-4: For each detected community, perform the following operations: ① Create an all-zero basis matrix: Generate an N×N zero matrix community_mask with the same dimensions as the input matrix; ② Set community connection identifiers: Traverse all node pairs (i, j) in the current community and set community_mask[i, j] to 1; ③ Matrix storage: Add the generated community identifier matrix to the community_matrices list; Step 6-5: Output a set of K community matrices, community_matrices, where each matrix fully represents the internal communication topology of the corresponding community; The derivation process of the fitness function of the genetic routing algorithm is: Step 6-6, use an undirected graph To define large-scale ad hoc networks, Represents a node in the network The collection of ; Represents a link in the network The topology between two nodes defines a link, i.e. ; Steps 6-7, use Represents the real-time bandwidth matrix of large-scale ad hoc networks, and defines is the source node To adjacent nodes The real-time bandwidth is ； in represents a positive real number; Steps 6-8, use Represents the real-time delay matrix of large-scale ad hoc networks, and defines For Link The delay is ； Steps 6-9, use Represents the real-time packet loss rate matrix of large-scale ad hoc networks, and defines For Link The packet loss rate is ； Steps 6-10, use represents the data flow in a large-scale ad hoc network, represents a collection of data streams, then ； is the source node of the data stream, For the destination node, 、 、 are the delay preference weight, bandwidth preference weight and packet loss preference weight of the data flow respectively. The throughput limit of the data flow; Steps 6-11, use Represents the weight matrix for centralized path planning of all data flows in a large-scale ad hoc network, where Represents an edge The weight of ； Steps 6-12, use Represents the forwarding path set of the data flow, which is the weight matrix The corresponding set of k shortest paths on the weighted undirected graph, k=3, the three paths represent the paths of the delay preference, bandwidth preference, and packet loss preference data flow from node s to node d, then there is ； Steps 6-13, use Indicates the forwarding path, For data flow in the path The bandwidth that can be reserved on For Link The remaining bandwidth is ； Steps 6-14, Indicates the total path delay of the data flow, Indicates a link The delay is ； Steps 6-15, Indicates the path packet loss rate of the data flow, Indicates a link The packet loss rate is ； Steps 6-16, represents the QOS performance score of the data flow, then ； Steps 6-17, use represents the flow betweenness set of large-scale ad hoc network nodes, For nodes The flow betweenness of ； Steps 6-18, use represents the sum of the flow betweenness of all nodes in a large-scale ad hoc network, then ； Step 6-19, use The total number of nodes is The average node flow betweenness in a large-scale ad hoc network is ； Steps 6-20, The total number of nodes is The arithmetic mean square error of node flow betweenness in a large-scale ad hoc network is: ； The smaller it is, the more evenly the flow betweenness is distributed, and the better the network load balancing and routing invulnerability are. Step 6-21, represents the evaluation score of network load balancing and routing invulnerability in large-scale ad hoc networks, then ； Step 6-22, Objective Function of Centralized Routing Planning Algorithm for: ； The topology control module is used to call the topology control application to generate the initial topology of the ad hoc network and reconstruct the topology when the communication status between nodes changes.

2. The large-scale self-organizing network control system based on software-defined networking according to claim 1 is characterized in that: The data plane layer is a large-scale self-organizing network composed of multiple nodes.

3. The large-scale self-organizing network control system based on software-defined networking according to claim 1, characterized in that: The data flow management module defines a bandwidth-limited QOS data flow, which expresses the performance preference for the network based on the delay preference weight, bandwidth preference weight, and packet loss preference weight, and also sets the upper limit of the bandwidth usage of the data flow; The data flow management module maintains the information of all data flows in the network.

4. The large-scale self-organizing network control system based on software-defined networking according to claim 1, characterized in that: The routing planning module performs centralized routing planning based on the real-time traffic matrix and information about all data flows, with the goal of comprehensive QOS performance of all data flows and load balancing of network nodes. When the traffic matrix changes, the number of data flows changes, or the topology is reconstructed, the module will re-plan the routing.

5. The large-scale self-organizing network control system based on software-defined networking according to claim 1, characterized in that: The topology control module constrains the topology construction between nodes based on the node community relationship and optimizes the MANET topology to facilitate the use of the community-constrained genetic routing planning algorithm on the MANET; formulates the initial topology of the MANET and reconstructs the topology when the communication status between nodes changes.

6. The large-scale self-organizing network control system based on software-defined networking according to claim 1, characterized in that: Steps of genetic routing planning algorithm based on community constraints: Step 7-1: Input the ad hoc network topology, link parameter matrix, data flow information, node communication relationship matrix, population number, crossover rate, and mutation rate; Step 7-2: Get node community information; Step 7-3: Randomly generate the initial population according to the population size; Step 7-4: Calculate the population fitness value according to the fitness function; Step 7-5: Determine whether the fitness function value converges. If so, output the routing weight matrix and exit; if not, continue with the subsequent steps. Step 7-6: Based on the crossover rate, use the tournament method to select individuals with higher fitness values ​​as parents; Step 7-7: Perform gene crossover using node community information as a constraint to generate offspring; Step 7-8: Mutate some individuals with lower fitness values ​​according to the mutation rate; Step 7-9: Generate a new population and go to step 7-4.

7. The large-scale self-organizing network control system based on software-defined networking according to claim 1, characterized in that: The topology control application uses a community constraint-based topology control algorithm to construct the topology; The algorithm constructs the initial topology of the ad hoc network based on the inter-node communication status and node community information. By merging the community matrices in the community matrix set and fusing the inter-node communication matrices, an initial fully connected graph is constructed. Then, redundant edges are randomly deleted to obtain an initial topology with a point connectivity of 2, which serves as the initial topology of the ad hoc network in the data plane layer.

8. A large-scale self-organizing network control method based on software-defined networking, characterized in that: Based on the system implementation of claim 1, the method comprises the following steps: Step 1: The topology control module constructs a communication demand relationship matrix based on the communication demand relationship of the ad hoc network nodes in the data plane, uses the deepwork algorithm to mine node community information, and obtains a node community matrix set; Step 2: The topology control module calls the topology construction application to randomly generate an ad hoc network topology with a point connectivity of 2 based on the inter-node connectivity matrix and the node community matrix set of the ad hoc network, as the initial topology of the ad hoc network; Step 3: After the ad hoc network is activated, the link monitoring module receives link status information uploaded by the data plane in real time and builds a traffic matrix. Step 4: The data flow management module receives the data flow request initiated by the self-organizing network node in real time, parses the parameters in the request, and uses the data flow initiating node, receiving node, QOS weight and bandwidth usage limit as data flow information to manage all data flow information; Step 5: When the number of data flows changes, the link status changes, or the topology is reconfigured, the routing planning module calls the routing planning application to perform routing planning. The routing planning application uses the community-constrained genetic routing planning algorithm to solve the routing problem based on the data flow information provided by the data flow management module, the global link information provided by the link monitoring module, and the node community information. Step 6: When the inter-node connectivity status changes, the topology control module calls the topology control application to reconfigure the topology. The topology control application uses a community-constrained topology control algorithm to reconfigure the topology based on the current inter-node connectivity matrix and the node community matrix set.

9. An electronic device comprising an imaging device, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to claim 8 are implemented.

Images