Intelligent networking method for high-dynamic large-scale mobile ad hoc network
Through the intelligent networking method, routing decision-making and security issues in high-dynamic large-scale mobile ad hoc network are solved, efficient management and security authentication of network nodes are realized, and suitable for high-dynamic large-scale new network application scenarios.
Patent Information
- Application Number
- CN202411206895.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-30
AI Technical Summary
In high-dynamic large-scale mobile ad hoc network, instantaneous changes in network topology, frequent connection interruptions, and rapid domain switching lead to difficulty in routing decision-making, and security threats such as node forgery and identity forgery, affecting network security and resource management.
Adopt intelligent networking method for high dynamic large-scale mobile ad hoc networking, and use mathematical optimization and causality by establishing networking models, building intelligent routing decision-making methods and designing security authentication strategies, including first-responsive, reactive and opportunity routing decisions, and using mathematical optimization and causality Inference and reinforcement learning to realize secure authentication of node organization and routing transmission.
It realizes organizational management and security guarantee for high-dynamic large-scale network nodes, improves the rapid organization and aggregation of network resources, and enhances the security and reliability of network transmission.
Smart Images

Figure CN118945756A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent mobile ad hoc networks, and in particular to an intelligent networking method for highly dynamic large-scale mobile ad hoc networks. Background Art
[0002] In today's network technology field, mobile ad hoc networks have become one of the most active research and application directions in network technology due to their advantages such as independence from infrastructure, flexible deployment, and strong anti-destruction capabilities. With the increasing popularity of smart portable devices, they have penetrated explosively into various industries such as detection and inspection, emergency rescue, battlefield communications, and aerospace. However, with the popularization of new intelligent nodes such as drones and high-speed vehicles, the demand for new high-dynamic large-scale mobile networking has emerged.
[0003] For traditional mobile networking, problems such as instantaneous changes in network topology, frequent connection interruptions, rapid switching of network domains caused by high dynamics, and whether there are available fixed base stations, mobile base stations, satellite base stations and other infrastructure can easily have a disruptive impact on the routing decisions of self-organizing networks; large-scale network nodes cause a sharp increase in network data volume and increased calculation or decision-making time, and the network transmission requirements caused by specific target tasks of mobile self-organizing networks are sudden, variable and homogeneous, which makes it difficult to quickly organize and aggregate network resources according to application requirements; in addition, there are various external threats to specific mobile self-organizing networks, such as node forgery, identity forgery, address forgery, etc., which will cause a high possibility of privacy leakage and poor controllability of connection security.
[0004] Therefore, the intelligent networking method has become a bottleneck for highly dynamic and large-scale new network applications, and there is an urgent need for an intelligent mobile ad hoc network suitable for highly dynamic and large-scale new network applications. Summary of the invention
[0005] The purpose of the present invention is to provide an intelligent networking method for highly dynamic large-scale mobile ad hoc networks, which realizes the organization and management of large-scale network nodes, solves the routing decision problems in various situations, ensures the security of network node access and transmission, and is suitable for highly dynamic large-scale new network application scenarios.
[0006] To achieve the above object, the present invention provides an intelligent networking method for a highly dynamic large-scale mobile ad hoc network, comprising the following steps:
[0007] S1. Establish a networking model for highly dynamic large-scale mobile ad hoc networks: establish a security node and functional node organization model from a logical perspective, and establish an intelligent network domain division model from a physical perspective;
[0008] S2. According to the communication requirements, an intelligent routing decision method for highly dynamic large-scale mobile ad hoc networks is constructed, which is specifically divided into three cases: proactive intelligent routing decision, reactive intelligent routing decision and opportunistic routing decision;
[0009] S3. Design security authentication strategies for node organization and routing transmission, including node security access authentication and transmission connection security authentication.
[0010] Preferably, in step S1, in establishing a networking model, network nodes are divided into security nodes and functional nodes according to their functions; and divided into four functional modules, namely, node organization engine, network domain division engine, routing decision engine and security authentication engine, according to the networking functions to be realized;
[0011] The network assurance node deploys all engine modules and completes the functions of each engine through distributed collaboration; the network function node deploys the routing decision engine module and the security authentication engine module.
[0012] Preferably, the specific steps of establishing the self-organizing network architecture model in step S1 are as follows:
[0013] S111. Using mathematical optimization, we construct the generalized network utility maximization problem of the ad hoc network architecture as follows:
[0014]
[0015] subject to Rx≤c
[0016] in, represents the guarantee node vector; represents a function node vector; represents the sending rate of the source node q in the network domain o; Identify the network utility function;
[0017] S112. Using mathematical optimization, causal reasoning and reinforcement learning, we model the Markov decision process and solve the generalized network utility maximization problem as follows:
[0018]
[0019] in, represents the state space; represents the action space; represents the state transition probability; represents the profit function;
[0020] S113, obtain the best optimization algorithm through reinforcement learning, and solve the best routing strategy R * and transport policy x * .
[0021] Preferably, the specific steps of node organization in step S1 are as follows:
[0022] S121, node virtual organization;
[0023] The security nodes are virtually organized according to logical connectivity to form a virtual mapping of security nodes;
[0024] According to different types of functional nodes, virtual mapping of multiple functional nodes is established;
[0025] Based on physical or logical relationships, the virtual mapping of the functional nodes is connected with the virtual mapping of the guarantee nodes to form a distributed virtual mapping organization of network nodes;
[0026] S122, ensure node optimization deployment;
[0027] First, the distributed structural causal model is established to ensure that node i is dynamically adjusted according to changes in the network environment;
[0028] Then, according to the conversion relationship between the causal model and the Markov decision process, the structural causal model is characterized as a multi-agent interference Markov decision process; the optimal deployment strategy π is found by maximizing the long-term cumulative utility function of the network. * ; Obtain the best security node deployment strategy to achieve efficient coverage of the target area by security nodes.
[0029] Preferably, in step S1, the specific process of the network domain division method is as follows:
[0030] S131, scene recognition;
[0031] Utilize existing network, environment perception technology and prior knowledge to establish a scene recognition model based on multimodal meta-transfer learning; perceive the network scene in the entire network domain through interaction between nodes;
[0032] S132, network domain division;
[0033] Based on the network scenario recognition results, a mathematical optimization model for network domain division is established; based on distributed causal reinforcement learning, the optimization problem is automatically solved to obtain the optimal routing strategy. and transmission strategy Implement an optimized network management model.
[0034] Preferably, the specific process of constructing the proactive intelligent routing decision method in step S2 is as follows:
[0035] S211. Based on the node organization model and the network domain division model, a graph network model is constructed as shown below:
[0036] G=(V,E,x V )
[0037] Where V represents a node set; represents the set of links between nodes; x V Represents the characteristic information of the node;
[0038] S212. Using the conditional random field model, a joint distribution is established to predict connectivity between nodes, as shown below:
[0039] p φ (y V |x V , E)
[0040] Where φ represents the model parameter; for directly maximizing log p φ (y L |x V , the lower bound of E) is optimized as follows:
[0041]
[0042] Among them, y L Represents some nodes The connectivity, q θ (y U |x V ) represents the connectivity distribution of the node set U;
[0043] S213, using the variational expectation maximization framework EM to optimize the lower bound in step S212 to obtain the model q θ :
[0044] In the M-phase, fixed q θ , update p by maximizing the likelihood function φ , as shown below:
[0045]
[0046] In the E-stage, q is updated by optimizing the objective function θ , as shown below:
[0047]
[0048] in, NB(i) is the neighbor of node i;
[0049] S214, alternating iterative training p φ and q θ Until convergence, the node connectivity prediction vector y is obtained V ;
[0050] S215, based on the connectivity vector y V, node priority, target tasks, establish a dynamically updated proactive routing strategy, and lay the routing foundation for real-time communication between nodes and high-level nodes.
[0051] Preferably, the specific process of constructing the reactive intelligent routing decision method in step S2 is as follows:
[0052] S221. Based on the route discovery process, a target conditional Markov decision process is established as follows:
[0053]
[0054] in, represents the network state space; represents the action space for selecting the next-hop routing node; represents the transition probability of the network state; G represents the target routing space; represents the network profit function; γ represents the discount factor;
[0055] S222. Establish a Markov decision process set as follows:
[0056]
[0057] in, p k Represents the causal relationship between network status and routing actions;
[0058] S223, randomly select an environment from K network environments, and use the interaction strategy π I Generate a network state transition sequence τ = {(s1, a1), (s2, a2), ...}, and use the inductive model F to build a causal model
[0059] S224. Obtaining a causal model through learning and routing strategy π G , generating real-time strategies between communication nodes.
[0060] Preferably, the specific process of constructing the opportunistic routing decision method in step S2 is as follows:
[0061] S231. Based on the node dynamic adjustment process, a Markov decision imitation learning model is established as follows:
[0062]
[0063] in, is the network state space, Action space for routing selection; represents the transition probability of the network state; r represents the set of instant network benefit functions; T represents the number of transitions of the network state;
[0064] S232: Construct a demonstration data set to obtain opportunistic routing decisions between nodes As shown below:
[0065]
[0066] in, represents the trajectory of potential available routes for a node; c i Indicates routing requirements;
[0067] S233, using imitation learning to build a node opportunity routing decision model, that is, to find a potential routing strategy π(a|s) from a set of available routing strategies π in a class of networks, simulating a demonstration data set The optimal routing strategy
[0068] Define the loss function:
[0069] in, is the indicator function;
[0070] Expected loss function:
[0071] The expected loss function for T steps is:
[0072] Among them, ρ π Represents the distribution of network status;
[0073] S234, by minimizing the loss function The opportunistic routing decision is as follows:
[0074]
[0075] The adaptive gradient descent algorithm is used to solve the optimization problem and obtain the delay-tolerant opportunistic routing strategy between nodes.
[0076] Preferably, the network node security access authentication process in step S3 is as follows:
[0077] S311, construct the public identification RID and initial invisible identification IID of the access node, then each proposed access node has a triple<RID,IID,Skey> ; Among them, Skey is a secure private key used for data encryption and decryption;
[0078] S312, when node A intends to access the network, it sends an access request carrying A's invisible identifier to the authentication server; the authentication server verifies A based on known legitimate access device information; if the verification is successful, it generates a new invisible identifier for A using dual identifier resolution mapping, and sends a successful access confirmation carrying A's new invisible identifier to A;
[0079] Construct the dual identity resolution mapping as follows:
[0080] [z a (X) RID ]→ψ[z a (X) IID ]
[0081] Where X represents the access node, a represents the access request, and z a (X) RID Indicates the access identifier, z a (X) IID Indicates the connection identifier;
[0082] S313: After receiving the access success confirmation, node A updates the invisible identifier.
[0083] Preferably, the network transmission connection security authentication process in step S3 is as follows:
[0084] S321, when node A needs to send information to node B, first send a message with node A's identity RID to the authentication server A |IID A and Node B public identifier RID B Transfer application;
[0085] S322, the authentication server verifies the identity information of nodes A and B. If the identity verification is successful, a set of encryption and decryption keys is generated using the dual identifications of both parties, as shown below:
[0086] (K A , K B )←Extract(pp, Skey, RID A |IID A , RID B |IID B )
[0087] Among them, pp represents the public parameter, which is sent to node A and node B in ciphertext form;
[0088] S323, node A uses the key obtained from the authentication server and its own invisible identifier to encrypt and encapsulate the data, and sends it to node B; when node B receives the data packet, it uses the secret key obtained from the authentication server to verify and decrypt the data packet; if the verification is successful, it sends a successful reception confirmation to node A;
[0089] S324, after receiving the confirmation information from node B, node A sends a transmission end confirmation to the authentication server; the authentication server regenerates new invisible identifiers for node A and node B and sends them to A and B respectively, and node A and node B update their new invisible identifiers for the next information transmission.
[0090] Therefore, the present invention adopts the above-mentioned intelligent networking method for highly dynamic large-scale mobile self-organizing networks, realizes the organization and management of large-scale network nodes, solves the routing decision problems in various situations, ensures the security of network node access and transmission, and is suitable for highly dynamic large-scale new network application scenarios.
[0091] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0092] Figure 1 It is a diagram of the intelligent identification self-organizing network architecture of an intelligent networking method for highly dynamic large-scale mobile self-organizing networks of the present invention;
[0093] Figure 2 It is a node organization diagram of distributed virtual mapping of an intelligent networking method for highly dynamic large-scale mobile ad hoc networks of the present invention;
[0094] Figure 3 The invention discloses a "send-authenticate-receive" three-party collaborative trusted transmission control process for an intelligent networking method for highly dynamic large-scale mobile ad hoc networks. DETAILED DESCRIPTION
[0095] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention more clearly understood, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0096] like Figure 1 As shown, an intelligent networking method for highly dynamic large-scale mobile ad hoc networks mainly includes three parts: an intelligent networking model, a routing decision method and a security authentication method.
[0097] S1. Based on the actual application of highly dynamic large-scale mobile ad hoc networks and network characteristics, an intelligent networking architecture model is established. From a logical perspective, a security node and functional node organization model is established. From a physical perspective, an intelligent network domain division model is established.
[0098] S2. In order to ensure the communication needs of nodes, random communication needs, and communication needs without available links, a proactive intelligent routing decision method based on graph Markov neural network, a reactive routing decision method based on causal reasoning, and a delay-tolerant opportunistic routing decision method based on imitation learning are established respectively.
[0099] S3. To address the security issues of node access and transmission connections, a node security authentication method based on “public-invisible” dual identification and a “send-authentication-receive” tripartite collaborative transmission security authentication method are designed.
[0100] Example
[0101] S1. Establish an intelligent networking architecture model;
[0102] The intelligent networking model mainly includes three parts: self-organizing network architecture, node organization and network domain division.
[0103] S11, ad hoc network architecture model;
[0104] In an environment with high dynamic and large-scale networking application requirements, the self-organizing network formed will cover multiple different network domains.
[0105] Network nodes can be divided into two categories according to their functions: one is the support nodes, which are network nodes specifically used for routing relay, such as communication satellites, operator fixed base stations, drone / unmanned vehicle mobile base stations, etc.; the other is the functional nodes, which are network nodes that complete the established target tasks and also have certain relay functions, such as reconnaissance drones and unmanned minesweepers.
[0106] According to the functions of realizing networking, it is divided into four modules: node organization engine, domain division engine, routing decision engine and security authentication engine.
[0107] The network assurance node deploys all engine modules and completes the functions of each engine through distributed collaboration; the network function node mainly deploys the routing decision engine and security authentication engine.
[0108] The node organization engine rationally organizes all nodes connected to the network, which are divided into security node organization and functional node organization. The two are mapped and associated with each other to achieve efficient management of network nodes without topology.
[0109] The network domain division engine combines node organization and scenario cognition to appropriately divide the network into domains and optimize the deployment of security nodes.
[0110] The routing decision engine maintains intelligent routing in various situations based on network status, node level, task level, etc.
[0111] The security authentication engine is based on existing encryption and decryption technologies and uses "public-invisible" dual identification to achieve secure management of node access and transmission connections.
[0112] S111. Use mathematical optimization to model the highly dynamic large-scale ad hoc network architecture as a generalized network utility maximization problem.
[0113] First, define the security node set and the functional node set as NID c and NID nc The set of available paths from the source node q to the destination node via the security node or the functional node is the adjacency matrix H q , then the available path matrix can be expressed as H = [H1, ..., H S ]; where S represents the number of source nodes.
[0114] Then, let w q,j represents the flow of the source node q on the jth path, and satisfies w q,j ≥0 and in, m represents the number of available paths from the source node q to the destination node; the routing matrix of the highly dynamic large-scale ad hoc network is:
[0115] R=HW
[0116] Where W is the vector w q A diagonal matrix composed of q=1,...,S.
[0117] Finally, let represents the sending rate of the source node q in the network domain o, c l Representing the capacity of the lth path (such as link bandwidth, node processing capacity, node energy, etc.), the generalized network utility maximization problem is as follows:
[0118]
[0119] subject to Rx≤c
[0120] in, Represents the assurance node vector, which consists of the location, resources, and authentication capabilities of the assurance node; Represents the function node vector, which is composed of the function node's location, task, and relationship with other nodes.
[0121] S112. The high dynamic and large-scale characteristics of the network lead to the network utility function The forms of are various, and the routing matrix and path capacity change rapidly. Therefore, mathematical optimization, causal reasoning and reinforcement learning are used to solve the optimization problem. The specific process is as follows:
[0122] Introduce the causal model to learn the adjacency matrix H q , and then get the routing matrix R, and model the causal model as a Markov decision process:
[0123]
[0124] in, Represents the state space, which consists of network scenario, path capacity c, Optimization Problem and its dual form; represents the action space, which consists of optimization algorithms; represents the state transition probability; represents the profit function.
[0125] S113, find the best optimization algorithm through reinforcement learning and solve the best routing strategy R * and transport policy x * .
[0126] Therefore, by solving the data model of the highly dynamic large-scale MANET architecture, efficient management of the entire network can be achieved.
[0127] S12, node organization;
[0128] S121, node virtual organization;
[0129] Under the requirements of highly dynamic large-scale networking applications, traditional node management methods that rely on network topology are expensive or ineffective. Distributed virtual mapping is an effective way to achieve node topology-free management.
[0130] First, the security nodes are virtually organized according to logical connectivity to form a virtual mapping of security nodes in a highly dynamic large-scale ad hoc network.
[0131] Then, according to different types of functional nodes, such as clustering, peering, grouping, etc., virtual mapping of multiple functional nodes is established;
[0132] Finally, according to the physical or logical relationship, the virtual mapping of the functional node is connected with the virtual mapping of the guarantee node to form a distributed virtual mapping organization of the network nodes, such as Figure 2 shown.
[0133] like Figure 2 As shown, based on the routing strategy R * and transport policy x *, the mathematical model of highly dynamic large-scale ad hoc networks is decomposed into two interrelated sub-optimization problems, as follows:
[0134] First, the Lagrange multiplier method is used to transform the generalized network utility maximization problem into the following optimization problem:
[0135]
[0136] Among them, r q represents the qth column vector of matrix R, R lq represents the element in the lth row and qth column of the matrix R, Indicates H q is the set of all column vectors of , λ represents the Lagrange multiplier vector.
[0137] Through the known routing strategy R * and transport policy x * , we can get the optimal λ * ; Use nonlinear function approximation algorithm to convert the function Breaking it down, it looks like this:
[0138]
[0139] Then solve the sub-optimization problems separately:
[0140]
[0141]
[0142] Get the guaranteed node vector and function node vector
[0143] Then, based on the obtained and And optimize the generalized network utility maximization problem, get the routing strategy and transmission strategy Then we can get the equilibrium solution and the model basis for managing assurance nodes and functional nodes.
[0144] Finally, the difference between the overall optimization problem solution and the sub-optimization solution is evaluated. The difference is defined as follows:
[0145]
[0146] By iteratively optimizing the difference, the optimal decomposition of the overall problem is obtained, thereby improving the overall performance of the network.
[0147] S122, ensure node optimization deployment;
[0148] Combining the self-organizing network architecture, security node organization and functional node organization, causal reinforcement learning is introduced to establish an optimized deployment method for security nodes to achieve efficient coverage of the target network area, as follows:
[0149] First, the process of dynamic adjustment of security node i according to changes in the network environment is modeled as a distributed structural causal model, as shown below:
[0150]
[0151] Among them, U i Represents exogenous variables, such as weather changes, geographical environment, human interference, etc.; V i represents endogenous variables, such as network status, node deployment strategy, network utility function, etc.; F i Represents structural functions, such as state transition, node deployment strategy, network utility function, etc.; P i Represents the distribution of all exogenous variables, such as Gaussian distribution, etc.; n is the number of guarantee nodes.
[0152] Secondly, according to the conversion relationship between causal model and Markov decision process, the structural causal model is characterized as a multi-agent interference Markov decision process.
[0153] Then, in order to achieve efficient coverage of the overall network by guaranteeing nodes, the optimal deployment strategy π is found by maximizing the long-term cumulative utility function of the network. * .
[0154] Finally, the optimal security node deployment strategy is obtained to achieve efficient coverage of the target area by security nodes.
[0155] S13, network domain division method;
[0156] S131, scene recognition;
[0157] By utilizing existing networks, environmental perception technologies, and prior knowledge, a scene recognition method based on multimodal meta-transfer learning is established, as follows:
[0158] First, the location, resources, status, environment, weather and other perception information obtained by node i are represented as a set Among them, the vector s i,j represents the j-th type of perception information of node i, and m represents the number of types of perception information.
[0159] Secondly, we use multimodal feature learning to perceive the information set S from the network scene. i Get the network scene dataset Initialize the feature extractor Θ of the training network scene using the deep neural network model i and classifier θ i, according to the stochastic gradient descent learning algorithm, we get Θ i , as shown below:
[0160]
[0161] Among them, α represents the learning rate, represents the empirical loss function.
[0162] Then, for a specific network scenario perception task Train the network scene classifier θ′ i , as shown below:
[0163]
[0164] Among them, β represents the learning rate, represents the data training set, and Represents different meta-operations, such as zooming and panning in network scenarios. Each meta-operation The learning rules are as follows:
[0165]
[0166] Where η represents the learning rate, represents the test data set. At the same time, the basic classifier θ of the network scene i , updated as follows:
[0167]
[0168] In this way, the basic perception model θ of the network scene can be obtained through training i , including the spatial location of the network, network resources, environmental weather and other information. Using the basic perception model θ i , through the classifier θ′ i Perceive specific network scenarios and update the model θ i , the network scenario where node i is located can be quickly identified through loop iteration.
[0169] Finally, the network scenario in the entire domain is perceived through interaction between nodes.
[0170] S132, network domain division;
[0171] Under the demand for highly dynamic and large-scale network applications, based on the network scenario recognition results of step S131, a mathematical optimization model for network domain division is established, which can realize efficient network management in combination with actual scenarios. The specific process is as follows:
[0172] First, the Lagrange multiplier method is used to transform the generalized network utility maximization problem into the following optimization problem:
[0173]
[0174] Next, the above optimization problem is decomposed into n sub-problems, and the Oth sub-optimization problem is:
[0175]
[0176] According to the optimal guarantee node vector and function node vector The Oth sub-optimization problem is:
[0177]
[0178] In order to solve each sub-optimization problem, an idea of automatically solving sub-optimization problems is established based on distributed causal reinforcement learning:
[0179] First, for each domain o∈{1,...,n}, a multi-domain causal model is constructed as follows:
[0180]
[0181] in, It consists of the optimal protection node vector, the optimal function node vector, the network domain O, the network scenario, etc. Indicates the association relationship between assurance nodes, the association relationship between function nodes, and the association relationship between assurance nodes and function nodes; express The probability distribution of Indicates the shared information between domain o and other domains.
[0182] Then, we learn through distributed causal discovery Will It is characterized as a generalized Markov decision process as follows:
[0183]
[0184] Among them, the state space Includes spatial deployment of security nodes and functional nodes, network scenarios, network domain information, etc.; represents the action space of domain o, which consists of a set of distributed optimization algorithms; γ∈[0, 1] represents the discount rate.
[0185] Therefore, the solution task of each sub-problem o can be defined as:
[0186]
[0187] in, represents the Markov decision process, r orepresents the profit function.
[0188] Next, distributed causal reinforcement learning is used to select the best optimization algorithm to solve each sub-optimization problem and obtain the optimal routing strategy. and transmission strategy
[0189] Finally, based on and Optimize network management model.
[0190] S2. Construct an intelligent routing decision method for highly dynamic large-scale mobile ad hoc networks.
[0191] Routing decision methods mainly include proactive intelligent routing decision, reactive intelligent routing decision and opportunistic routing decision.
[0192] S21, proactive intelligent routing decision method;
[0193] Based on the node organization model, network domain division model, node priority, etc., a node connectivity prediction method based on graph Markov neural network is designed to dynamically maintain the proactive routing strategy between nodes and high-priority nodes with potential transmission tasks, as follows:
[0194] S211. Construct a graph network model based on the node organization model, domain division model, etc., such as hour:
[0195] G=(V,E,x V )
[0196] Where V represents a node set; represents the set of links between nodes; X V Represents the characteristic information of the node, such as historical trajectory, established tasks, priority, etc.
[0197] If (i, j) ∈ E, it means that there are one or more links between nodes i∈V and i∈V, then nodes i∈V and i∈V are said to be connected; let the matrix C = [c ij ] |V|×|V| represents the connectivity matrix, where |V| represents the number of nodes.
[0198] When (i, j)∈E, then c ij =1, otherwise c ij = 0; the connectivity matrix C is vectorized as
[0199] y V :y V =vec(C)
[0200] S212. Use the conditional random field model to establish a joint distribution to achieve connectivity prediction between nodes, as shown below:
[0201] p φ (y V |x V , E)
[0202] Among them, φ represents the model parameters.
[0203] Since log p is directly maximized φ (y L |x V , E) is more difficult, y L Represents some nodes The connectivity of , so its lower bound is optimized:
[0204]
[0205] Among them, q θ (y U |x V ) represents the connectivity distribution of the node set U.
[0206] S213, using the variational expectation maximization framework (Expectation Maximization, EM) to optimize the above lower bound to obtain the model q θ .
[0207] In the M-phase, fixed q θ , update p by maximizing the following likelihood function φ :
[0208]
[0209] In the E-stage, q is updated by optimizing the following objective function θ :
[0210]
[0211] in, NB(i) is the neighbor of node i.
[0212] S214, alternating iterative training p φ and q θ Until convergence, the node connectivity prediction vector y is obtained V .
[0213] S215, based on the connectivity vector y V , node priority, target tasks, etc., establish a dynamically updated proactive routing strategy to lay the routing foundation for real-time communication between nodes and high-level nodes.
[0214] S22, reactive intelligent routing decision method;
[0215] Based on the node organization model, network domain division model, node established tasks, proactive routing strategy, etc., a reactive intelligent routing decision method based on causal reasoning is designed to maintain the routing strategy between random communication task nodes in real time. The details are as follows:
[0216] S221. Model a target conditional Markov decision process based on the route discovery process, as shown below:
[0217]
[0218] The goal is to maximize network revenue Finding routing strategy π G .
[0219] in, Represents the network state space, such as network scenarios, security node deployment, network resources, etc. represents the action space for selecting the next-hop routing node; represents the transition probability of the network state; G represents the target routing space; represents the network profit function, that is, the one-step instant profit based on the target g∈G is r(s, a, g); γ represents the discount factor.
[0220] S222. In order to improve the generalization ability of the routing strategy, a Markov decision process set is established as follows:
[0221]
[0222] in, p k Represents the causal relationship between network status and routing actions.
[0223] S223, randomly select an environment from K network environments, and use the interaction strategy π I Generate a network state transition sequence τ = {(s1, a1), (s2, a2), ...}, and use the inductive model F to construct a causal model
[0224] S224. Using causal models Set the routing policy π G Placed in context, it ensures that routing tasks can be performed in the new network environment.
[0225] To learn the inductive model F, the set of all Markov decision processes is divided into two disjoint sets and
[0226] In the training phase, supervised learning methods are used to Learn F and use the DAgger algorithm to train the learning routing strategy π G .
[0227] In the testing phase, we evaluate whether we can build a causal model by learning F in a new network environment. Can causal models be used? Finding routing strategy π G .
[0228] S225. Obtaining a causal model through learning and routing strategy π G . Aiming at specific random communication needs, based on routing strategy π G , generating real-time strategies between communication nodes.
[0229] S23, Opportunistic Routing Decision Method;
[0230] When no link is available, an opportunistic routing decision method based on imitation learning is established based on the node organization model, node established tasks, network domain division model, communication task QoS, established routing strategy, etc., to generate a routing strategy under delay tolerance conditions, as follows:
[0231] S231. Based on the node organization model, node established tasks, network domain division model, etc., the node dynamic adjustment process is modeled as a Markov decision imitation learning model, as shown below:
[0232]
[0233] in, The network state space, such as network scene images, node location information, network resources, etc. Action space for routing selection; represents the transition probability of the network state; r represents the set of instant network benefit functions; T represents the number of transitions of the network state.
[0234] S232. According to the node organization and network domain division, simulate the actual network environment to obtain the trajectory set of the node's potential available routing strategy, and build a demonstration data set to obtain the opportunistic routing decision between nodes. As shown below:
[0235]
[0236] in, represents the trajectory of potential available routes for a node; c i Indicates routing requirements.
[0237] S233, using imitation learning to build a node opportunity routing decision model, that is, to find a potential routing strategy π(a|s) from a set of available routing strategies π in a class of networks, simulating a demonstration data set The optimal routing strategy
[0238] Define the loss function:
[0239] in, is the indicator function.
[0240] Expected loss function:
[0241] The expected loss function for T steps is:
[0242] Among them, ρ π Represents the distribution of network status.
[0243] S234, by minimizing the loss function The potential opportunistic routing decisions can be obtained as follows:
[0244]
[0245] The adaptive gradient descent algorithm is used to solve the optimization problem and obtain the delay-tolerant opportunistic routing strategy between nodes.
[0246] S3. Design security authentication strategies for node organization and routing transmission.
[0247] The security authentication method mainly includes two parts: node security access authentication and transmission connection security authentication.
[0248] S31, network node security access authentication process, as shown below:
[0249] S311, construct the public identification RID (RouterID) and the initial invisible identification RID (RouterID) of the access node; select the organic combination of the ID, product serial number and other physical information of the access node as the public identification RID; extract the organic combination of the parameter information, buffer information, backup file information of the access node as the invisible identification IID; then each proposed access node has a triple<RID,IID,Skey> ; Among them, Skey is the secure private key used for data encryption and decryption.
[0250] S312, when node A intends to access the network, it sends an access request carrying A's invisible identifier to the authentication server; the authentication server verifies A based on known legitimate access device information; if the verification is successful, it generates a new invisible identifier for A using dual identifier resolution mapping, and sends a successful access confirmation carrying A's new invisible identifier to A;
[0251] In order to achieve rapid conversion of "public-invisible" identification, a dual identification resolution mapping is constructed as follows:
[0252] [z a (X) RID ]→ψ[z a (X) IID ]
[0253] Where X represents the access node, a represents the access request, and z a (X) RID Indicates the access identifier, z a (X) IID Indicates the connection identifier.
[0254] S313: After receiving the access success confirmation, node A updates the invisible identifier.
[0255] S32, network transmission connection security authentication process, such as Figure 3 As shown:
[0256] S321, when node A needs to send information to node B, it first sends a message carrying node A's identity RID to the authentication server. A |IID A (RouterIDA, InvisibleIDA) and the public identifier RID of node B B (RouterIDB) transmission request;
[0257] S322, the authentication server verifies the identity information of nodes A and B. If the identity verification is successful, a set of encryption and decryption keys is generated using the dual identifications of both parties:
[0258] (K A , K B )←Extract(pp, Skey, RID A |IID A , RID B |IID B )
[0259] Among them, pp represents the public parameter and is sent to node A and node B in ciphertext form.
[0260] S323, node A uses the key obtained from the authentication server and its own invisible identifier to encrypt and encapsulate the data, and sends it to node B; when node B receives the data packet, it uses the secret key obtained from the authentication server to verify and decrypt the data packet; if the verification is successful, it sends a successful reception confirmation to node A.
[0261] S324, after receiving the confirmation information from node B, node A sends a transmission end confirmation to the authentication server; the authentication server regenerates new invisible identifiers for node A and node B and sends them to A and B respectively, and node A and node B update their new invisible identifiers for the next information transmission.
[0262] Therefore, the present invention adopts the above-mentioned intelligent networking method for highly dynamic large-scale mobile self-organizing networks, realizes the organization and management of large-scale network nodes, solves the routing decision problems in various situations, ensures the security of network node access and transmission, and is suitable for highly dynamic large-scale new network application scenarios.
[0263] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. An intelligent networking method for highly dynamic large-scale mobile ad hoc networks, characterized in that: The following steps are involved: S1. Establish a networking model for highly dynamic large-scale mobile ad hoc networks: establish a security node and functional node organization model from a logical perspective, and establish an intelligent network domain division model from a physical perspective; S2. According to the communication requirements, an intelligent routing decision method for highly dynamic large-scale mobile ad hoc networks is constructed, which is specifically divided into three cases: proactive intelligent routing decision, reactive intelligent routing decision and opportunistic routing decision; S3. Design security authentication strategies for node organization and routing transmission, including node security access authentication and transmission connection security authentication.
2. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: In step S1, in establishing a networking model, network nodes are divided into security nodes and functional nodes according to their functions; and divided into four functional modules: node organization engine, network domain division engine, routing decision engine and security authentication engine according to the networking functions to be realized; The network assurance node deploys all engine modules and completes the functions of each engine through distributed collaboration; The network function node deploys the routing decision engine module and the security authentication engine module.
3. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The specific steps of step S1 to establish the self-organizing network architecture model are as follows: S111. Using mathematical optimization, we construct the generalized network utility maximization problem of the ad hoc network architecture as follows: subject to Rx≤c in, represents the guarantee node vector; represents a function node vector; represents the sending rate of the source node q in the network domain o; Identify the network utility function; S112. Using mathematical optimization, causal reasoning and reinforcement learning, we model the Markov decision process and solve the generalized network utility maximization problem as follows: in, represents the state space; represents the action space; represents the state transition probability; represents the profit function; S113, obtain the best optimization algorithm through reinforcement learning, and solve the best routing strategy R * and transport policy x * .
4. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The specific steps of step S1 node organization are as follows: S121, node virtual organization; The security nodes are virtually organized according to logical connectivity to form a virtual mapping of security nodes; According to different types of functional nodes, virtual mapping of multiple functional nodes is established; Based on physical or logical relationships, the virtual mapping of the functional nodes is connected with the virtual mapping of the guarantee nodes to form a distributed virtual mapping organization of network nodes; S122, ensure node optimization deployment; First, the distributed structural causal model is established to ensure that node i is dynamically adjusted according to changes in the network environment; Then, according to the conversion relationship between the causal model and the Markov decision process, the structural causal model is characterized as a multi-agent interference Markov decision process; the optimal deployment strategy π is found by maximizing the long-term cumulative utility function of the network. * ; Obtain the best security node deployment strategy to achieve efficient coverage of the target area by security nodes.
5. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: In step S1, the specific process of the network domain division method is as follows: S131, scene recognition; Utilize existing network, environment perception technology and prior knowledge to establish a scene recognition model based on multimodal meta-transfer learning; perceive the network scene in the entire network domain through interaction between nodes; S132, network domain division; Based on the network scenario recognition results, a mathematical optimization model for network domain division is established; based on distributed causal reinforcement learning, the optimization problem is automatically solved to obtain the optimal routing strategy. and transmission strategy Implement an optimized network management model.
6. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The specific process of constructing the proactive intelligent routing decision method in step S2 is as follows: S211. Based on the node organization model and the network domain division model, a graph network model is constructed as shown below: G=(V,E,x V ) Where V represents a node set; represents the set of links between nodes; x V Represents the characteristic information of the node; S212. Using the conditional random field model, a joint distribution is established to predict connectivity between nodes, as shown below: p φ (y V |x V ,E) Where φ represents the model parameter; for directly maximizing logp φ (y L |x V , the lower bound of E) is optimized as follows: Among them, y L Represents some nodes The connectivity, q θ (Y U |x V ) represents the connectivity distribution of the node set U; S213, using the variational expectation maximization framework EM to optimize the lower bound in step S212 to obtain the model q θ : In the M-phase, fixed q θ , update p by maximizing the likelihood function φ , as shown below: In the E-stage, q is updated by optimizing the objective function θ , as shown below: in, NB(i) is the neighbor of node i; S214, alternating iterative training p φ and q θ Until convergence, the node connectivity prediction vector y is obtained V ; S215, based on the connectivity vector y V , node priority, target tasks, establish a dynamically updated proactive routing strategy, and lay the routing foundation for real-time communication between nodes and high-level nodes.
7. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The specific process of constructing the reactive intelligent routing decision method in step S2 is as follows: S221. Based on the route discovery process, a target conditional Markov decision process is established as follows: in, represents the network state space; represents the action space for selecting the next-hop routing node; represents the transition probability of the network state; G represents the target routing space; represents the network profit function; γ represents the discount factor; S222. Establish a Markov decision process set as follows: in, p k Represents the causal relationship between network status and routing actions; S223, randomly select an environment from K network environments, and use the interaction strategy π I Generate a network state transition sequence τ = {(s1, a1), (s2, a2), ...}, and use the inductive model F to build a causal model S224. Obtaining a causal model through learning and routing strategy π G , generating real-time strategies between communication nodes.
8. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The specific process of constructing the opportunity routing decision method in step S2 is as follows: S231. Based on the node dynamic adjustment process, a Markov decision imitation learning model is established as follows: in, is the network state space, Action space for routing selection; represents the transition probability of the network state; r represents the set of instant network benefit functions; T represents the number of transitions of the network state; S232: Construct a demonstration data set to obtain opportunistic routing decisions between nodes As shown below: in, represents the trajectory of potential available routes for a node; c i Indicates routing requirements; S233, using imitation learning to build a node opportunity routing decision model, that is, to find a potential routing strategy π(a|s) from a set of available routing strategies ∏ in a class of networks, simulating a demonstration data set The optimal routing strategy Define the loss function: in, is the indicator function; Expected loss function: The expected loss function for T steps is: Among them, ρ π Represents the distribution of network status; S234, by minimizing the loss function The opportunistic routing decision is as follows: The adaptive gradient descent algorithm is used to solve the optimization problem and obtain the delay-tolerant opportunistic routing strategy between nodes.
9. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: The network node security access authentication process in step S3 is as follows: S311, construct the public identification RID and initial invisible identification IID of the access node, then each proposed access node has a triple<RID,IID,Skey> ; Among them, Skey is a secure private key used for data encryption and decryption; S312, when node A intends to access the network, it sends an access request carrying A's invisible identifier to the authentication server; the authentication server verifies A based on known legitimate access device information; if the verification is successful, it generates a new invisible identifier for A using dual identifier resolution mapping, and sends a successful access confirmation carrying A's new invisible identifier to A; Construct the dual identity resolution mapping as follows: [z a (X) RID ]→ψ[z a (X) IID ] Where X represents the access node, a represents the access request, and z a (X) RID Indicates the access identifier, z a (X) IID Indicates the connection identifier; S313: After receiving the access success confirmation, node A updates the invisible identifier.
10. The intelligent networking method for highly dynamic large-scale mobile ad hoc networks according to claim 1, characterized in that: In step S3, the network transmission connection security authentication process is as follows: S321, when node A needs to send information to node B, first send a message with node A's identity RID to the authentication server A |IID A and Node B public identifier RID B Transfer application; S322, the authentication server verifies the identity information of nodes A and B. If the identity verification is successful, a set of encryption and decryption keys is generated using the dual identifications of both parties, as shown below: (K A ,K B )←Extract(pp,Skey,RID A |IID A ,,RID B |IID B ) Among them, pp represents the public parameter, which is sent to node A and node B in ciphertext form; S323, node A uses the key obtained from the authentication server and its own invisible identifier to encrypt and encapsulate the data, and sends it to node B; when node B receives the data packet, it uses the secret key obtained from the authentication server to verify and decrypt the data packet; if the verification is successful, it sends a successful reception confirmation to node A; S324, after receiving the confirmation information from node B, node A sends a transmission end confirmation to the authentication server; the authentication server regenerates new invisible identifiers for node A and node B and sends them to A and B respectively, and node A and node B update their new invisible identifiers for the next information transmission.
Citation Information
Patent Citations
Social network-based vehicle-mounted self-organization network routing method
CN103702387A
Routing method for linear topology wireless ad hoc network
CN110831104A
Low-delay network slicing method based on deep reinforcement learning
CN113098714A
Intelligent driving behavior decision-making method and device fusing complex network theory and partially observable Markov decision-making process
CN116027788A
Off-line double-Q learning routing method based on geographic position
CN116456416A
