High-dynamic mobile ad hoc network clustering multipath routing method, device, equipment and storage medium

Abstract

The invention discloses a high-dynamic mobile ad hoc network clustering multipath routing method, device and equipment and a storage medium, and the method comprises the steps: dynamically electing a cluster head through comprehensively considering indexes such as spatial proximity, a speed following coefficient, an average link retention rate, a correction node degree difference and an energy residual rate, and constructing an efficient cluster structure; a dynamic weight distribution mode based on an entropy method is adopted, index weights are adjusted in a self-adaptive mode according to the real-time state of the network, and it is ensured that cluster head election and routing decision are more scientific and flexible. In addition, a cluster maintenance method is also designed, and through an HELLO-ACK interaction mechanism and a backup cluster head notification mechanism, the stability and continuity of the network under the conditions of node dynamic change and fault are ensured. A multi-path routing scheme based on a clustering architecture is further established, and the fault-tolerant capability of the network and the reliability of data transmission are improved through an inter-cluster multi-path routing mechanism; by expanding a field in an RREQ message, efficient establishment of inter-cluster routing and path information aggregation are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of wireless communication and network technology, and specifically relates to a high dynamic mobile ad-hoc network clustering multi-path routing method, device, equipment and storage medium. BACKGROUND

[0002] Under the background of today's complex and changeable communication needs, high dynamic mobile ad-hoc networks (HD-MANET) show great application potential and value. Flying ad-hoc networks (FANET) as a typical scenario of high dynamic mobile ad-hoc networks, with its unique three-dimensional space deployment, high-speed mobile characteristics and diversified task requirements, has become a key entry point for researching high dynamic mobile ad-hoc network technology. Unmanned aerial vehicles (UFP) with their wide air view and high mobility play an important role in many fields such as military reconnaissance, environmental monitoring, traffic management, smart agriculture and disaster management. With the continuous progress of energy storage, computing power and wireless communication technology, unmanned aerial vehicles have become an ideal tool for dealing with various complex tasks, which can access specific areas on demand and capture real-time data to meet key requirements such as monitoring, environmental perception and connectivity. Through effective cooperation, FANET can provide efficient communication services in high dynamic environments, enhancing the robustness and reliability of the system. However, with the continuous expansion of application scenarios and the increasing coverage area, the entire high dynamic mobile ad-hoc network field is facing many challenges, such as the rapid change of network topology caused by the high mobility of nodes, network coverage and capacity limitations, dynamic environmental conditions and high delay caused by multi-hop transmission, especially when facing large-scale unmanned aerial vehicle clusters or other mobile carriers cooperating to complete complex tasks, which puts higher requirements on the stability, efficiency and scalability of the network.

[0003] Routing protocols play a key role in FANET, responsible for establishing and maintaining efficient data transmission paths in a dynamically changing network environment. Existing FANET routing protocols mainly fall into two technical routes: one is to improve traditional MANET routing protocols to adapt to the high dynamic characteristics of FANET; the other is to use intelligent optimization algorithms to model FANET routing problems and find the optimal routing strategy through iterative exploration. Under the drive of the two technical routes, FANET routing protocols can be divided into four categories: topology-based routing, location-based routing, swarm intelligence routing, and reinforcement learning routing. Topology-based routing requires selecting the transmission path before transmitting data to ensure that data packets can reach the destination node smoothly. This type of protocol can be divided into proactive and reactive. Proactive routing protocol, also known as table-driven routing, maintains a complete routing table at each node in the network and updates this information regularly to ensure that the information in the routing table is up-to-date. This protocol does not rely on data packet transmission to discover or maintain routing information. The main advantage of proactive routing protocol is that it can quickly respond to network changes because the routing information is pre-computed. However, in high dynamic scenarios, the links between nodes fluctuate frequently, and the stability of the route is low. Proactive routing needs to update routing information constantly, resulting in high overhead. Reactive routing initiates the routing discovery process on demand, sacrificing initial route setup delay for higher packet delivery rate, resulting in larger transmission delay. The above two routing methods can only compromise between real-time performance and overhead, and cannot minimize routing overhead while ensuring real-time performance. Traditional routing mainly borrows from MANET routing protocols, i.e., nodes calculate the shortest path based on passive acquisition of network topology and link state, etc. However, this has serious limitations in three-dimensional high-speed scenarios: network state information quickly becomes outdated, lacks efficient perception and prediction of future state, and link quality has not been comprehensively evaluated, thus weakening routing performance and increasing delay and packet loss. Thanks to the development of Global Navigation Satellite System (GNSS), geographic information-based routing protocols are widely used. Nodes do not need to explore the entire network state and can directly perform greedy forwarding based on local information, unifying data transmission and routing discovery. However, this protocol requires UFP nodes to have high-precision positioning capabilities to locate themselves and destination nodes. In high dynamic FANET scenarios, not only do they need to frequently obtain their own positioning information, but they also need to frequently discover neighbors, consuming a large amount of energy and spectrum resources. In addition, relying on distance for path selection can easily fall into local optimization and routing hole problems. In recent years, inspired by some emerging intelligent optimization algorithms, swarm intelligence routing and reinforcement learning routing have entered the field. These technologies are very suitable for solving FANET routing decision problems due to their distributed, adaptive, and global optimal solution characteristics.In particular, the application of deep reinforcement learning enables UFPs to autonomously learn optimal routing strategies by modeling the routing problem as a Markov decision process. However, in high dynamic scenarios, these protocols often suffer from blind search, local optimality, slow convergence, and other issues. Breaking through these technical bottlenecks and designing specialized routing protocols for emerging new UFP application scenarios remains challenging.

[0004] Existing FANET routing solutions are mainly based on the optimization and improvement of traditional ground mobile ad hoc network routing solutions. Although these improved solutions provide an important foundation for the initial study of FANET, due to the unique characteristics of FANET nodes such as high-speed movement and drastic changes in topology, directly applying these solutions can lead to frequent routing interruptions, high packet loss rates, and excessive routing maintenance overhead. To address the challenges posed by high dynamic environments, researchers have developed a variety of enhanced routing solutions specifically for FANET. The main goal of these solutions is to address the link instability caused by high-speed node movement and to improve routing stability, reduce latency, and reduce routing overhead while maintaining node connectivity as much as possible. Improved solutions often employ more complex routing strategies, such as predictive routing, environment-aware mechanisms, and multi-path routing, to adapt to the high dynamics of nodes and the rapid changes in network topology in FANET. In addition, these solutions take into account the unique communication environment of FANET, such as three-dimensional spatial deployment constraints, special movement patterns of aerial nodes (such as formation flight, task-oriented movement, etc.), and air-ground collaborative communication requirements, thus providing more reliable and efficient routing solutions for FANET. However, these improved solutions still have deficiencies in addressing the high-speed movement challenges of FANET.

[0005] Dynamic clustering is one of the solutions for FANET to cope with large-scale and highly dynamic topology challenges, aiming to optimize network topology to improve network performance and management level. The core goal is to reduce network management complexity, reduce redundant communication and energy consumption, improve routing efficiency and network stability, while enhancing network scalability and adaptability to high dynamic environment. The technology uses a multi-dimensional dynamic weight mechanism for cluster head election, considering node energy level, mobility, link stability and other key indicators, and designs a dynamic reconstruction mechanism to maintain networking. Typical applications include cooperative search, emergency networking and other scenarios, where adaptive clustering mechanism can effectively reduce communication overhead and improve network survival time. However, existing dynamic clustering technology still faces many challenges, such as insufficient dynamic adjustment of index weight, optimization of cluster head position and cluster size, and high dynamic topology adaptability. In recent years, combining reinforcement learning technology to achieve more accurate cluster head election and topology prediction has become an important development direction for dynamic clustering technology, but it also brings new challenges such as high-dimensional state space modeling and on-board real-time computing.

[0006] In the prior art, the invention patent with application number CN202311219314.6, "A clustering routing method and device for unmanned aerial platform ad hoc network", proposes a clustering routing method and device for unmanned aerial platform ad hoc network. The method is: combining the geographical position and flight speed of unmanned aerial platform nodes, all unmanned aerial platform nodes are divided into sub-clusters; in the divided sub-cluster, the largest load capacity unmanned aerial platform node is selected as the cluster head by comprehensively considering the remaining energy, storage capacity and load bearing of the unmanned aerial platform node; based on the formed sub-cluster and cluster head, the routing communication between sub-clusters and within clusters is carried out according to the existing routing protocol. The device includes a sub-cluster division module, a cluster head determination module and a routing communication module.

[0007] This method effectively improves the clustering effect by using a clustering method based on hierarchical k-means clustering, considering geographical position and flight speed, and reduces message flooding and resource overhead in the network. In the cluster head election process, the largest load capacity node is selected as the cluster head by comprehensively considering the remaining energy, storage capacity and load bearing of the unmanned aerial vehicle, significantly improving the survival time of the cluster head, reducing the network computing overhead, and improving the routing efficiency. In addition, routing communication is carried out based on the formed sub-cluster and cluster head, further optimizing the communication path and improving the overall throughput of the network.

[0008] In the invention patent with the application number CN202210260132.2, a self-organizing network hierarchical routing method for unmanned aerial vehicles based on fuzzy logic is proposed. This method includes two stages: clustering and routing. In the clustering stage, a clustering algorithm is used to divide the nodes in the network into different clusters. Based on the clustering results, a virtual backbone network is constructed. In the routing stage, an on-demand routing mechanism is used to start the routing discovery process to find the path to the destination node. This invention can be applied to large-scale unmanned aerial vehicle self-organizing networks. It can construct a hierarchical network structure through a distributed clustering method, reduce message flooding in large-scale networks, reduce the probability of message collision, and improve communication performance. At the same time, the complexity of this invention is relatively low.

[0009] In a specific implementation, all unmanned aerial vehicle nodes periodically broadcast HELLO messages containing their own identity, location, and mobility, and maintain a neighbor table recording the identity, mobility, and fitness of neighboring nodes. Nodes calculate their own mobility through a remoteness function and determine whether to become a candidate cluster head by judging whether their mobility is lower than that of more than half of the nodes in the neighbor table. The candidate cluster head further calculates the energy factor, bandwidth factor, and location factor, and converts these factors into fuzzy values using fuzzy logic. The fitness is calculated through IF / THEN rules and output membership functions. The candidate cluster head broadcasts the fitness to neighboring nodes, and if its fitness is higher than that of all neighboring nodes, it declares itself as a cluster head, completing clustering. After clustering is completed, if a node belongs to multiple clusters, it becomes a gateway node. Nodes perform routing discovery and data transmission through the backbone network formed by clustering. Source nodes directly send data or establish a route by broadcasting a routing request message based on whether the destination node is a neighbor node.

[0010] This method exhibits significant performance advantages in large-scale unmanned aerial vehicle self-organizing networks. Through a dynamic clustering strategy, it effectively reduces message flooding and resource overhead in the network, reduces the probability of message collision, and significantly improves communication performance. The cluster head selection mechanism based on node mobility avoids high-mobility nodes participating in the election, prolonging the survival time of the cluster. The introduction of fuzzy logic considers key indicators such as node energy, bandwidth, and distance, making the cluster head election more scientific and further improving clustering efficiency. In addition, routing discovery and data transmission are performed using a backbone network, optimizing the communication path and improving network throughput.

[0011] The main shortcomings of the above-mentioned prior art solutions are as follows:

[0012] 1. Traditional routing protocols are difficult to adapt to high dynamics: Most existing FANET routing protocols are based on improvements of traditional MANET protocols such as AODV and OLSR, which are designed for low mobility ground networks and are difficult to adapt to the high mobility and rapidly changing network topology of FANET. In addition, geographic location-based routing protocols such as GPSR cannot adaptively discover reliable communication links, and the link quality fluctuates greatly, easily leading to data transmission interruption. In high dynamic scenarios, traditional routing protocols need to update routing information frequently, resulting in large control overhead and network delay. Multi-hop routing, although it improves the coverage and reliability of the network, also brings problems such as increased network delay, decreased throughput, and uneven energy consumption.

[0013] 2. There are limitations in the application of intelligent optimization: Some routing protocols based on reinforcement learning such as Q-Learning can adapt to network dynamics, but in high dynamic scenarios they are prone to local optimization and slow convergence. The parameters of Q-Learning such as learning rate and discount factor are difficult to adaptively adjust according to network conditions, affecting routing performance. Routing algorithms based on swarm intelligence such as ant colony algorithm can find global optimal solutions, but in large-scale high dynamic networks the computational complexity is high and it is difficult to respond to network changes in real time.

[0014] 3. Network management is inefficient: In traditional flat architecture, all nodes are equal in status and lack hierarchical structure and management mechanisms, resulting in low efficiency in network coordination and task allocation. When the node density is high or the task is complex, the flat routing scheme exposes serious network congestion problems due to the surge of routing forwarding traffic. Frequent node movement and rapid topology changes force routing protocols to update frequently, causing slow convergence and high delay, making it difficult to meet the real-time communication needs of high dynamic environments.

[0015] 4. Clustering routing algorithms lack adaptability: Existing clustering algorithms have obvious defects in high dynamic FANET. First, the cluster head election mechanism is imperfect, mostly based on a single attribute such as residual energy to elect cluster heads, lacking a comprehensive consideration of nodes, making it difficult to maintain efficient routing. Second, the weight adjustment lacks dynamism, and fixed weights are difficult to adapt to node dynamics, affecting clustering effectiveness. In addition, the clustering scale and cluster head location are unreasonable, resulting in low clustering coverage and affecting network performance. At the same time, the dynamic topology adaptability is poor, making it difficult to cope with rapidly changing network topology. Finally, there is a lack of multi-path transmission mechanism, and single-path routing is difficult to cope with link failure, reducing network fault tolerance. SUMMARY

[0016] In view of the deficiencies of the prior art, the present application provides a high-dynamic mobile ad hoc network clustering multipath routing method, device, equipment and storage medium, focusing on the high-dynamic flight ad hoc network scene, and designing a weighted dynamic clustering-based multipath invulnerability routing protocol, which aims to solve the defects of the existing clustering algorithm in dynamic adaptability, cluster head election mechanism and routing stability.

[0017] The high-dynamic mobile ad hoc network clustering multipath routing method of the present application has the following specific steps:

[0018] Step 1: initialization of state information of each node deployed in the target area.

[0019] Step 2: each node periodically broadcasts a HELLO probe packet carrying node ID, position, speed and energy information; the node obtains neighbor state information by listening to the HELLO message, calculates five indexes of spatial proximity, speed following coefficient, average link retention rate, modified node degree difference and energy remaining rate, and broadcasts the index values to the outside through a METRIC message.

[0020] Step 3: the node constructs a local index matrix according to the index transmission message listened to, determines the weight of each index, and obtains the cluster head election score through weighted calculation.

[0021] Step 4: each node broadcasts a CHELEC message carrying its election score; the node receives the election score message of the neighbor node, arranges the candidate cluster head list in descending order according to the score; the node checks whether its score is higher than the highest score in the candidate cluster head list, if yes, the node is self-boosted as a cluster head, updates its state to "network access", and broadcasts a CHDECL message carrying the cluster head declaration; if not, step 5 is executed for cluster head coexistence judgment.

[0022] Step 5: each node dynamically calculates the number of cluster heads N CH required by the current neighborhood according to the size of its neighbor nodes; when the node score is lower than the highest score in the candidate cluster head list, it is judged whether it can be a coexisting cluster head according to the ranking. If it is ranked ahead of N CH , the node can be a coexisting cluster head, and broadcasts a CHDECL message carrying the cluster head declaration; otherwise, it continues to wait for the cluster access opportunity.

[0023] Step 6: The node receives all neighbor broadcasts of CHDECL messages carrying cluster head declaration, checks whether the message source contains the highest scoring candidate in the candidate cluster head list; if yes, the network access handshake process: sends a JOINREQ message to the cluster head, waits for the cluster head to return a JOINACPT message; after receiving the JOINACPT message sent by the cluster head, broadcasts a CMNOTIF message to confirm joining, and updates its own state to "network access"; if no, after checking all received neighbor CMNOTIF messages, the following judgment is performed: check whether the message source exists in the candidate cluster head list, if yes, remove the node sending the message from the list to prevent the cluster head node from suppressing the pending node from entering the cluster; if no, determine whether the node has completed cluster entry, if not, requery the candidate cluster head list and select to join the remaining highest scoring cluster head. If the node cannot meet the cluster entry condition all the time, go to Step 1 to recompute and broadcast the weighted score of the node.

[0024] The above process will be executed in a loop until all nodes in the network complete cluster entry.

[0025] Step 7: Establish a cluster maintenance mechanism to ensure the stability and reliability of the network, including a HELLO-ACK interaction mechanism to periodically check the connection status with the member nodes in the cluster, and a backup cluster head announcement mechanism to realize emergency switching in case of failure.

[0026] Step 8: The source node initiates communication demand to establish multi-path routing, dividing the path discovery process into intra-cluster routing discovery and inter-cluster routing discovery; when the upper layer application of a node generates data transmission demand, the node becomes a source node; when the source node generates communication demand, it first queries its local neighbor table: if it contains the destination node, it directly performs one-hop transmission; if it does not find the destination node, it sends a routing request to the cluster head to which the source node belongs; then, the source cluster head judges whether the destination node is located in the same cluster according to the cluster topology: if yes, the source cluster head starts intra-cluster routing and completes data forwarding according to the intra-cluster routing table; otherwise, the source cluster head starts inter-cluster routing.

[0027] The advantages of the present application are:

[0028] 1. Higher data transmission reliability.

[0029] Through the multi-path transmission mechanism and dynamic cluster maintenance strategy, the packet loss rate of data transmission is significantly reduced. The multi-path mechanism provides redundant transmission paths, so that even if part of the link fails, data can still be transmitted through other paths, ensuring the stability and reliability of data transmission. Especially in the scenario of high-speed node movement and frequent changes in network topology, the multi-path transmission strategy of the present scheme can effectively cope with link interruption and channel quality fluctuations, significantly outperforming traditional single-path routing protocols.

[0030] 2. Adapt to the dynamic changes of link and topology.

[0031] The introduction of the dynamic weight distribution mechanism and the cluster head coexistence mechanism can realize the real-time sensing of the changes of network topology, such as node position, speed, link duration and node density. Through the dynamic weight distribution based on the entropy value method, the cluster head election and clustering strategy can be adaptively adjusted, the network dynamic change can be quickly responded, and the resource utilization rate and communication performance of the network in a high dynamic scene are significantly improved.

[0032] Advantage three: improve network throughput and resource utilization.

[0033] 3. Through the optimization of cluster head election, multi-path transmission and lightweight cluster maintenance mechanism, the throughput of the network is significantly improved, especially when the network state fluctuates greatly. The multi-path transmission mechanism can transmit data on different paths at the same time, effectively utilize network resources, reduce time slot conflicts and communication delay.

[0034] Advantage four: reduce control overhead and clustering time.

[0035] 4. The weighted clustering algorithm is adopted, the cluster head election process is optimized, the rounds required for clustering are reduced, the control overhead and clustering time are significantly reduced. Through the introduction of the cluster head coexistence mechanism, the problem of mutual coverage of competing nodes is also effectively solved, and the clustering process is further accelerated. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 It is a disaster communication FANET scene diagram composed of multiple unmanned flight platforms and a ground control center;

[0037] Figure 2 It is a whole flow chart of the clustering multi-path routing method for high dynamic mobile ad hoc network of the application;

[0038] Figure 3 It is a whole flow chart of the weighted dynamic clustering in the method of the application;

[0039] Figure 4 It is a weight gradient dependence schematic diagram in the cluster head election process in the method of the application;

[0040] Figure 5 It is a backup cluster head mechanism flow chart in the method of the application;

[0041] Figure 6 It is a multi-path routing flow chart in the method of the application.

[0042] Figure 7 It is an inter-cluster multi-path routing schematic diagram;

[0043] Figure 8 It is a whole structure block diagram of the clustering multi-path routing device for high dynamic mobile ad hoc network of the application. DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below with reference to the accompanying drawings.

[0045] The present invention proposes a clustered multipath routing method for highly dynamic mobile ad hoc networks. The physical entity comprises multiple unmanned aerial platforms (which can be implemented in the form of various computing devices) with the same data forwarding capabilities, eliminating the need for ground infrastructure and a centralized control center. Each node is composed of a device module, including a memory, a processor, and a computer program stored on the memory and executable on the processor. The computer device, which can be a server, includes a processor, memory, a network interface, and a database connected via a system bus. The device's processor provides computing and control capabilities. The device's memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and computer program in the non-volatile storage medium. The device's database stores data related to the highly dynamic mobile ad hoc network, such as node location information and link quality data. The device's network interface communicates with other external nodes via a network connection. When executed by the processor, the computer program implements the clustered multipath routing method for highly dynamic mobile ad hoc networks of the present invention.

[0046] In this implementation, a disaster recovery communication FANET scenario consisting of multiple unmanned aerial platforms and a ground control center (GCC) is considered. Figure 1 As shown in the figure, in this network, unmanned aerial platforms are organized into single-hop clusters. Each cluster consists of a cluster head (CH) and several cluster member nodes (CM). CM nodes are not only capable of performing specific tasks but are also equipped with high-resolution cameras and various environmental sensors, acting as flight monitoring units. In addition to managing and maintaining the cluster structure, the CH is responsible for aggregating information within the cluster and transmitting this data to the GCC. It also receives commands from the GCC and executes mission decision-making and dispatch functions. The cluster head and member nodes always dynamically maintain a complete neighbor list.

[0047] Initially, the neighbor list is empty. Subsequently, through regular information exchange between nodes, the location, speed, and metrics of neighbor nodes are dynamically updated. The specific format of the neighbor list is shown in Table 1.

[0048] Table 1 Neighbor list format

[0049]

[0050] Wherein, the meaning of each field is as follows:

[0051] • NodeID represents the unique identification of the neighbor node (such as IP address, MAC address or custom ID).

[0052] • TimeStamp represents the neighbor discovery timestamp, recording the time when the neighbor is first discovered, used for aging mechanism (eliminating outdated entries).

[0053] • NodePosition represents the real-time coordinates (latitude, longitude, height) of the neighbor node, obtained through GPS or inertial navigation.

[0054] • NodeVelocity represents the velocity vector (including size and direction) of the neighbor node.

[0055] • NodeEnergy represents the remaining energy of the neighbor node.

[0056] • NodeScore represents the weighted score calculated by the neighbor node in the cluster head election, the higher the score, the greater the possibility of winning the election.

[0057] • ClusterID represents the cluster head ID of the cluster to which the neighbor node belongs, initially this field is 0.

[0058] • NodeRole represents the role of the neighbor node in the cluster, 0 represents cluster member, 1 represents cluster head, 2 represents backup cluster head, and 3 represents inter-cluster relay.

[0059] • NodeStatus represents the node network state identification, 0 represents pending, 1 represents network access, and 2 represents network disconnection.

[0060] Since the node selects the optimal neighbor as the CH based on the weighted score when entering the cluster, the neighbor list of the CH is different from its member list. In addition, when the ordinary cluster member (CM) finds non-cluster nodes in its neighbor list, it will record the cluster head IDs of all connected cluster groups and timely report to the CH through the HELLO message. After receiving the report, the CH will mark this node as an inter-cluster relay (ICR) and additionally record the cluster head IDs of all clusters connected by the node in the neighbor list, so as to realize more efficient inter-cluster communication and network management.

[0061] In inter-cluster communication, there is usually no direct communication link between CH nodes, and data exchange with other CHs needs to be achieved through inter-cluster relay nodes. GCC, as the ground control center, is mainly responsible for receiving and integrating the data collected by the unmanned flight platform; at the same time, it can command, initiate tasks, and coordinate rescue work. There are two types of communication in the network: between UFP and GCC (U2G) and between UFPs (U2U). In the routing process, any node can be regarded as a source node, an intermediate node and a target node.

[0062] It is assumed that all UFP nodes are equipped with an omnidirectional antenna above their rotors, and can use the GPS positioning system to perceive their position and velocity information within a tolerable error range, i.e. the i unmanned flight platform n i has a three-dimensional position vector p i = [x i , y i , z i ] and a velocity vector v For any two unmanned flight platforms, if the distance between them is within the communication radius (R), they can communicate with each other and are neighbor nodes. In the task area, the motion of each node exhibits Markov characteristics, i.e. the motion state at the current time only depends on the motion state at the previous time, while allowing moderate randomness. Therefore, the present application proposes a three-dimensional Gauss-Markov group (GMG) mobility model, which combines the advantages of the reference point group (RPG) mobility model and the Gauss-Markov (GM) mobility model. The GM model can effectively depict the time correlation and randomness of node motion, while the RPG model is more suitable for describing the spatial correlation between nodes.

[0063] Initially, the unmanned flight platforms are deployed in a clustered form, a node is randomly selected as a reference node in each cluster, and a GM mobility model is used to generate a moving trajectory, and the remaining ordinary nodes move around the node. The motion of the reference node is time-dependent, and the changes in its speed and direction are continuous. The reference node moves with a group movement vector , and each group of ordinary nodes updates its position by superimposing a random motion vector on the reference node. The length d of the random vector is uniformly distributed within a sphere centered at the reference node with a radius R, and the azimuth angle θ and the pitch angle φ are uniformly distributed within the intervals [0°, 360°] and [-90°, 90°], respectively. Each random motion vector is independent of the historical position information of the corresponding ordinary node. Through the GMG mobility model, the unmanned flight platform nodes can move according to the roughly predetermined trajectory, while retaining a certain dynamic and flexibility, providing a basis for the design of subsequent clustering routing schemes.

[0064] In the actual operation of FANET, the position, speed, energy and other state information of unmanned aerial platform changes all the time. Although the network is initially deployed in a clustered form and the reference nodes are selected based on the GMG mobile model to generate trajectories, the reference nodes are randomly designated and cannot directly assume the role of cluster head. As the core role in the subset cluster network structure, the cluster head undertakes key tasks such as routing forwarding, task allocation, relay communication and ground communication. Once the cluster head is improperly selected or fails, it will lead to serious consequences such as communication interruption and data packet loss.

[0065] Therefore, in order to ensure the continuous and efficient operation of the network in a dynamic environment, a clustered multipath routing method for high dynamic mobile ad hoc network is designed based on the above-mentioned physical entity composed of multiple node devices and the mobility model involved in the present application, a reasonable evaluation index system is used for cluster head election, and an effective cluster maintenance mechanism is established, such as Figure 2 As shown in the figure, the method comprises the following steps:

[0066] Step 1: network initialization.

[0067] A large number of unmanned aerial platform nodes are deployed in the target area, all nodes are in the "pending" (Pending) state, the node state information is initialized, including node ID, position, speed, energy and the like.

[0068] Step 2: neighbor node discovery and index calculation.

[0069] Each node periodically broadcasts a HELLO probe packet (for node discovery) carrying node ID, position, speed and energy information. The node obtains neighbor state information by listening to the HELLO message, calculates five indexes including spatial proximity, speed following coefficient, average link retention rate, modified node degree difference and energy remaining rate, and broadcasts these index values through METRIC message (for transmitting index information) to the outside.

[0070] The above-mentioned spatial proximity can reflect the node distribution of the entire network; the speed following coefficient is used to measure the stability of the node in the network topology; the average link retention rate reflects the quality of communication between the node and its neighbors through the survival time of the link; the modified node degree difference can reflect whether the network load is balanced; considering the limited energy reserves of the node, the energy remaining rate also needs to participate in the cluster head evaluation. The node obtains neighbor state information by periodically exchanging HELLO messages, and calculates each index in the following manner.

[0071] (1) Spatial proximity

[0072] This index reflects the spatial distribution characteristics of the node in the network by evaluating the relative distance relationship between the node and all its neighbor nodes. The normalized spatial proximity P i of node ni is defined as follows

[0073]

[0074] where N i represents the set of symmetric one-hop neighbors of node n i , p i (t) = [x i (t), y i (t), z i (t)] represents the position of node n i in the x, y, z axes at time t, and R represents its communication radius. p j is the position vector of neighbor node n j at time t.

[0075] The cluster head needs to have strong regional coverage capability, and the nodes with high spatial proximity indicate that they are close to the physical distance of the neighbor nodes, which can more effectively realize the communication within the region, reduce signal attenuation and energy consumption. In addition, good spatial distribution characteristics help to optimize the network topology and improve data transmission efficiency.

[0076] (2) Speed following coefficient

[0077] This index is used to quantify the alignment degree between the speed of a node and the speed of its adjacent nodes, reflecting the movement similarity between nodes. The speed following coefficient F i of node n i is defined as follows

[0078]

[0079] where v represents the speed of node n i in the x, y, z axes at time t.

[0080] In a network with strong mobility, the cluster head node needs to maintain a relatively consistent motion pattern with the neighbor nodes to reduce the frequency of dynamic changes in the network topology. A higher speed following coefficient indicates that the movement trajectory of the node is similar to that of the neighbor nodes, which can reduce the frequency of link breakage and cluster head replacement, thereby improving the stability of the network.

[0081] (3) Average link retention rate

[0082] This index evaluates the stability of the link based on the link survival time. The normalized predicted link survival time of link L ij can be represented as Here, the stability of the communication link between node n i and all its neighbors is averaged to obtain the average link retention rate R i .

[0083]

[0084] where δ ij is the link lifetime, indicating the time period that the communication link between node u j and u i remains active during the movement of u ij ; s i is the relative speed between two nodes; a node with a higher average link lifetime indicates that its communication link with neighbor nodes is more persistent, which can effectively reduce the risk of data transmission interruption and ensure the continuity and reliability of the network.

[0085] (4) Corrected node degree difference

[0086] This index is used to evaluate the deviation between the actual node degree and the ideal node degree, reflecting the balance of network load. In an ideal state, the optimal clustering scale in a distributed network of N nodes is Therefore, the ideal node degree is defined as The node degree difference is the absolute deviation between the actual node degree and the ideal node degree, i.e. i =||N i |-D ideal |.

[0087] However, in the actual network, the degree of some nodes may be significantly higher than the ideal node degree D ideal , resulting in a disproportionate reduction in the calculated deviation, thereby introducing gradient dependence in network evaluation. In order to eliminate this effect, the node degree difference needs to be corrected, and the corrected node degree difference D i is defined as follows:

[0088]

[0089] The corrected node degree difference can help the system select nodes with more balanced connectivity as cluster heads, thereby avoiding the cluster head being too concentrated or too sparse in the topology. By controlling the node degree difference, the size and distribution of the cluster can be made more uniform, thereby improving the clustering efficiency of the network.

[0090] (5) Energy remaining rate

[0091] The index is defined as the ratio of residual energy to nominal energy, reflecting the current energy state of the node. The energy consumption of FANET nodes mainly comes from flight, communication and calculation. Among them, the motion energy consumption is much higher than the latter two, and the communication energy consumption is significantly higher than the calculation energy consumption, so the calculation energy consumption can usually be ignored. Although the motion energy consumption accounts for a large proportion, it is mainly determined by the flight control system such as flight mission and path planning, and is limited by the physical characteristics of the aircraft and the task requirements, and cannot be optimized by routing protocol; while the communication energy consumption is directly related to the routing protocol, and can be directly reduced by optimizing the data transmission path. Therefore, this paper will focus on the communication energy consumption.

[0092] To quantify the communication energy consumption, a model of energy consumption for communication between two nodes is established. It is assumed that node n i has a data packet of size l bits to send to node n j . The total communication energy consumption of this process includes the transmission energy consumption of the sending node n i and the receiving energy consumption of the receiving node n j . The mathematical expression is as follows:

[0093]

[0094] In the formula, ε elec and ε fs are the coefficients of the transmitter (receiver) and the transmission power amplifier respectively. d is the distance between nodes.

[0095] The energy residual rate of node n i is defined as follows:

[0096]

[0097] In the formula, E rem,i = E o -E C,i represents the residual energy of the node, E o is the nominal energy, and E C,i is the energy consumed by node n i .

[0098] In the unmanned aerial platform network, energy management is the key to ensure the sustainable operation of the system, especially in emergency applications such as disaster relief, which requires high timeliness. The energy state directly affects the operational efficiency of the unmanned aerial platform. The cluster head is responsible for core tasks such as data aggregation and communication coordination, and its energy reserve directly affects the network efficiency. High-energy nodes are more capable of these tasks, which helps to optimize energy distribution and prolong network life.

[0099] Step 3: Calculate the cluster head election score.

[0100] ​The node constructs a local index matrix according to the monitored indicators, determines the weight of each indicator, and obtains a cluster head election score through weighted calculation:

[0101]

[0102] In the formula, w i is a weight coefficient, i=1, 2,..., 5, and w1+w2+w3+w4+w5=1.

[0103] The traditional clustering algorithm assigns a fixed weight coefficient to the index, which cannot adapt to the dynamic topology and energy consumption changes of the FANET. Therefore, the present application introduces a dynamic weight distribution method based on the entropy value method, and the node adjusts the index weight according to the local information to adapt to the dynamic changes of the network environment. Compared with other methods, the entropy value method has advantages in objectivity, adaptability and computational complexity, and can better meet the needs of dynamic weight distribution. First, the entropy value method completely depends on the data itself, avoiding subjective bias of human intervention, and ensuring the fairness of weight distribution; second, its dynamic adjustment ability is outstanding, and it can respond to the changes of index data in real time, especially suitable for high dynamic FANET scene; in addition, the entropy value method has simple and lightweight calculation process, and is suitable for large-scale data processing and real-time calculation demand. Therefore, the present application calculates the dynamic weight of the index based on the entropy value method. The specific process is as follows:

[0104] (1) Construct a local index matrix: in the FANET, each node collects the position, speed, energy and node degree of the surrounding neighbor nodes through the neighbor discovery (Neighbor Discovery) process, and calculates its own index. The foregoing index definition formula has completed the normalization processing, and the value is mapped to the [0, 1] interval, eliminating the order of magnitude difference. Then, the node broadcasts the index value to the neighbor nodes, and constructs a local index matrix X with a dimension of (|N b |+1)×m. Where, (|N b |+1) represents the total number of the current node and its neighbor nodes, and m is the number of indexes; the matrix element x ij represents the jth index value of the ith neighbor node.

[0105] (2) Calculate the information entropy and redundancy: the information entropy E j is used to measure the uncertainty of the jth index. The higher the information entropy, the less information the index provides in the current network environment, and the relatively lower the importance. Then, the redundancy D j is calculated according to the information entropy, which is used to reflect the correlation between the indexes and their repeated contribution to the overall information.

[0106] (3) Dynamic weight of the index: the weight distribution follows the principle of entropy method, that is, the index with lower information entropy is given higher weight, ensuring that the index with high information quantity occupies a larger proportion in decision-making.

[0107] The pseudo code for calculating the dynamic weight based on the entropy method is as follows:

[0108]

[0109] Step 4: Cluster head election.

[0110] As shown in Figure 3 , each node broadcasts CHELEC message (for cluster head election) carrying its election score. The node receives the election score message of the neighbor node, and arranges the candidate cluster head list in descending order according to the score. The node checks whether its own score is higher than the highest score in the candidate cluster head list. If yes, the node will be self-boosted as a cluster head, updates its own state to "network access", and broadcasts CHDECL message carrying the cluster head declaration; if not, step 5 is executed for cluster head coexistence judgment.

[0111] Step 5: Cluster head coexistence judgment.

[0112] As shown in Figure 4 , the cluster head election process may cause the competition nodes to cover each other due to the gradient difference distribution of the network center node weight, thereby inhibiting the generation of other cluster heads, and forming an isolated communication area. In order to optimize this problem, the cluster head coexistence mechanism is proposed in the present application, which allows multiple nodes with similar or partially overlapping coverage areas to jointly serve as cluster heads. Each node dynamically calculates the number of cluster heads N CH required by the current neighborhood according to the size of its neighbor nodes.

[0113]

[0114] In the formula, , represents the ideal cluster size.

[0115] When the node score is lower than the highest score in the candidate cluster head list, whether it can be a coexisting cluster head is judged according to the ranking. If the ranking is ahead of N CH , the node can be a coexisting cluster head, and broadcasts CHDECL message carrying the cluster head declaration; otherwise, it continues to wait for the cluster access opportunity.

[0116] Step 6: Cluster group formation.

[0117] The node receives all neighbor broadcast CHDECL messages carrying the cluster head declaration, and checks whether the message source contains the candidate with the highest score in the candidate cluster head list.

[0118] If included, the network access handshake process: send JOINREQ message to the cluster head (for cluster access application), wait for the cluster head to return JOINACPT message (for agreeing to cluster access); after receiving the JOINACPT message sent by the cluster head, broadcast CMNOTIF message (for member notification) to confirm joining, and update the state of the node to "network access".

[0119] If not included, after checking all received CMNOTIF messages of the neighbors, the following judgment is performed: check whether the message source exists in the candidate cluster head list, if it exists, remove the node sending the message from the list to prevent the cluster head node from inhibiting the pending node from accessing the cluster; if it does not exist, determine whether the node has completed cluster access, if not, requery the candidate cluster head list and select to join the cluster head with the highest remaining score. If the node cannot meet the cluster access condition all the time, the neighbor discovery process (step 2) is entered to recalculate and broadcast the weighted score of the node.

[0120] The above process (steps 2-6) will be executed in a loop until all nodes in the network complete cluster access.

[0121] Step 7: Maintain dynamic cluster.

[0122] After the cluster is formed, the topology of the FANET is determined. However, due to the high dynamic nature of the unmanned flight platform nodes, the node state will change constantly, including the joining of new nodes and the departure of existing nodes. Therefore, an efficient cluster maintenance mechanism needs to be established to ensure the stability and reliability of the network. The present application describes the cluster maintenance mechanism from two aspects:

[0123] (1) HELLO-ACK interaction mechanism

[0124] Each unmanned flight platform node in the network must participate in the maintenance of the cluster. The cluster head node periodically checks the connection state with the member nodes in the cluster through the HELLO-ACK interaction mechanism. Specifically:

[0125] The cluster head periodically broadcasts a HELLO message to the member nodes, and the member nodes respond with an ACK message after receiving it. If the CH does not receive an ACK response from a certain CM within the preset bT time, it is determined that the CM has left its control range, and it is removed from the neighbor list (including the member list) and the routing information. Wherein, T is the HELLO message period, and b is an adjustable parameter, which is dynamically adjusted according to the network environment, task requirements and node role. For example, backup cluster head b = 1, insensitive node b = 2, and BCH takeover node b = 3.

[0126] (2) Backup cluster head announcement mechanism

[0127] To reduce the control overhead caused by frequent re-clustering, a backup cluster head (BCH) mechanism is adopted to realize emergency switching in case of failure. Considering that it is difficult for the backup cluster head to effectively replace the cluster head function in a high dynamic scenario, after clustering is completed, only a single BCH is set for each cluster in order to quickly take over when the CH fails. Once the CH and BCH fail in succession, the nodes will trigger the re-execution of the clustering process. The announcement of the BCH is realized through the HELLO message periodically broadcasted by the CH, as shown in Figure 5 The specific process is as follows:

[0128] ① BCH selection and announcement: the CH selects the node with the highest weighted score from the member list as the BCH, and broadcasts the identity information of the BCH in the BCH field of the HELLO message;

[0129] ② BCH role activation: after receiving the message, the selected node marks itself as a "BCH", activates the role immediately and synchronizes the state information, so as to ensure that the CH can be seamlessly taken over when the CH fails, and the continuity of communication is ensured;

[0130] ③ Member node processing: after receiving the message, other nodes extract the BCH field and query the local neighbor list: if the node exists in the neighbor list, the node is marked as a BCH takeoverable node; if the node is not found and the relay node of the BCH only contains the CH, the node is marked as a non-sensitive node, and the cluster head maintenance period is set to two HELLO periods, so that the node can quickly exit the cluster when the CH fails.

[0131] When the backup cluster head does not receive the cluster head confirmation message sent by the CH in step ① within a preset time, it is determined that the cluster head fails, and the backup cluster head upgrades itself to a "CH" and broadcasts a new HELLO message. The backup cluster head takeoverable node updates the cluster head information and joins the new cluster; the non-sensitive node waits for the cluster head to expire and then executes the cluster exit and cluster entry process.

[0132] Step 8: The source node initiates a communication demand to perform multi-path routing.

[0133] The path discovery process is divided into two parts: intra-cluster route discovery and inter-cluster route discovery. As shown in Figure 6 When the upper layer application of a node generates a data transmission demand, the node becomes a source node. When the source node generates a communication demand, it first queries its local neighbor table: if the destination node is included, one-hop transmission is directly performed; if the destination node is not found, a route request is sent to the cluster head to which the source node belongs (the source cluster head). Subsequently, the source cluster head determines whether the destination node is located in the same cluster according to the cluster topology: if yes, the source cluster head starts intra-cluster routing and completes data forwarding according to the intra-cluster routing table; otherwise, the source cluster head starts inter-cluster routing, which is specifically:

[0134] First, to support inter-cluster multipath routing, the present application extends the following fields of RREQ packet: source cluster head ID, path stability and path load. Among them, the source cluster head ID identifies the initiating cluster of the routing request, which is used for loop detection and path tracing; the link stability η ij Refers to the quantized standard link survival time; the link load ι ij Depending on the node with higher queue occupancy rate among the two ends of the link, that is, for the link L ij , the path weight index is defined as: Among them And Respectively represent the queue occupancy rate of node u i And u j . The RREQ packet can be uniquely identified by the triple <source cluster head ID, source node ID, sequence number>.

[0135] Subsequently, the source cluster head starts the inter-cluster routing process: the source cluster head broadcasts RREQ to all nodes in the cluster; since ordinary cluster members do not participate in inter-cluster routing, ordinary cluster members directly discard RREQ after receiving it. After receiving RREQ, the inter-cluster relay (ICR) checks the source cluster head ID of RREQ: if RREQ comes from the cluster head of the cluster, the ICR forwards RREQ to the cluster heads of other clusters connected by the ICR, realizing cross-cluster propagation. If RREQ comes from the cluster head of other cluster, the ICR forwards RREQ to the cluster head of the cluster, so that the CH collects path information for path information aggregation.

[0136] When RREQ reaches the CH that can address the destination node, the destination cluster head generates a routing reply RREP and sends RREP back to the source cluster head along the reverse path of RREQ, completing routing establishment.

[0137] Since the throughput of single path is low and the anti-destroying ability is poor, the traditional single path routing scheme may not be suitable for transmission in high dynamic network. Compared with single path routing, multipath routing is a better choice to realize inter-cluster multipath propagation. Inter-cluster multipath refers to the existence of multiple transmission paths to the destination cluster head, which pass through different relay cluster heads respectively. After receiving multiple identical RREQs, the destination cluster head selects the optimal path as the main path and at most two suboptimal paths as backup paths according to the path weight index. Among them, the path weight index is calculated according to the path stability, path load and hop count of the current path, and is sorted by priority. Selecting three paths (one main path and two backup paths) is a compromise scheme that takes into account performance, reliability and resource overhead, which can guarantee high fault tolerance and fast routing switching while avoiding excessive routing maintenance overhead, adapting to network dynamics and resource-constrained environments, such as Figure 7 As shown.

[0138] After the inter-cluster multipath routing is established, the source node has the ability to transmit data through multiple independent paths in parallel. The present application proposes an adaptive multipath transmission strategy based on network dynamic characteristics, which realizes efficient and reliable data transmission through path management. In terms of data transmission, a differentiated strategy is adopted according to the dynamic change characteristics of the network topology: for high dynamic network environment, a reliable transmission scheme based on redundancy coding is adopted, the source node sends duplicate packets in parallel through two or more independent paths, and the data redundancy mechanism is used to replace the traditional retransmission mode, which significantly improves the transmission reliability in the scene of fast changing topology; in the scene of relatively stable network topology, a multipath concurrent transmission mechanism based on load balancing is enabled, the data flow is dynamically scheduled to different paths through flow balancing distribution, which not only realizes the optimal utilization of network bandwidth resources, but also effectively improves the overall throughput of the system. This dual-mode adaptive transmission mechanism realizes dynamic balance between reliability and transmission efficiency by real-time sensing of network state, and can meet the data transmission needs in different application scenarios.

[0139] As shown in Figure 8 , based on the above method, the present application proposes a clustering multipath routing device for high dynamic mobile ad hoc networks, which adopts a hierarchical modular architecture design, integrates various functional modules, and works cooperatively to cope with high dynamic network environment, significantly improves network throughput, resource utilization and stability, and is suitable for high dynamic scenarios such as unmanned aerial vehicle formation, vehicle ad hoc network, emergency communication network, etc., providing reliable support for efficient communication in complex network environment. As shown in Figure 5 , the figure shows the interaction between modules and the exchange of information between node devices. This architecture supports routing decisions and efficient communication in distributed systems, especially in mobile devices and dynamic network environments.

[0140] In this architecture, the upper layer is the application module responsible for generating data transmission requests according to user or system task requirements, and scheduling the middle layer routing module to determine the best transmission path, and then starting data transmission.

[0141] The middle layer includes a weighted dynamic clustering module and a multipath routing module. The weighted dynamic clustering module is deployed in all network nodes, which contains an index calculation module for calculating various indexes of node participating in cluster head election according to network topology information, and a dynamic weight allocation module based on entropy method for adaptively adjusting index weight according to real-time state of the network, ensuring that cluster head election and routing decision are more scientific and flexible, effectively solving the defects of existing clustering algorithms in dynamic adaptability, cluster head election mechanism and routing stability.

[0142] The multi-path routing module improves the fault tolerance and reliability of data transmission of the network through an inter-cluster multi-path routing mechanism. By expanding fields in the RREQ message, such as source cluster head ID, path stability, and path load, efficient establishment of inter-cluster routing and aggregation of path information are achieved. At the same time, the destination cluster head selects the optimal path as the main path according to the path weight index, and at most selects two suboptimal paths as backup paths, taking into account performance, reliability, and resource overhead.

[0143] The middle layer also includes a cluster maintenance module, which ensures the stability and continuity of the network under dynamic changes and failure conditions of nodes through a HELLO-ACK interaction mechanism and a backup cluster head announcement mechanism. The HELLO-ACK interaction mechanism uses a differentiated strategy to achieve efficient and reliable data transmission according to the dynamic change characteristics of the network topology. In a highly dynamic network environment, a reliable transmission scheme based on redundancy coding is used; in a relatively stable network topology scenario, a multi-path concurrent transmission mechanism based on load balancing is enabled, and through real-time sensing of network state, a dynamic balance between reliability and transmission efficiency is achieved.

[0144] The bottom layer is a communication module responsible for basic data transmission and interacting with neighbor nodes.

[0145] The application also provides a computer-readable storage medium having a computer program stored thereon. The program, when executed by a processor, can implement the steps of any of the aforementioned weighted clustering multi-path routing methods for flight ad hoc networks. The routing method can be implemented in the form of a software functional unit and sold or used as an independent product. These software functional units can be stored in a computer-readable storage medium, and users can realize the clustering routing and forwarding function by installing and executing the software.

[0146] The computer-readable storage medium includes but is not limited to the following forms:

[0147] (1) Physical storage medium

[0148] · U disk: portable storage device, suitable for program distribution in small networks.

[0149] · Mobile hard disk: large-capacity storage device, suitable for storing programs and data of large-scale networks.

[0150] · Magnetic disk: traditional storage medium, suitable for offline storage.

[0151] · Optical disc: including CD, DVD, Blu-ray disc, etc., suitable for program distribution and long-term storage.

[0152] · Computer memory: including RAM and ROM, used for temporary storage and fast access of program code.

[0153] (2) virtual storage media

[0154] • an electrical carrier signal carrying the program code.

[0155] • a telecommunications signal carrying the program code.

[0156] • a software distribution medium carrying the program code.

Claims

1. A high dynamic mobile ad hoc network clustering multipath routing method, device, equipment and storage medium, characterized in that: The steps are: Step 1: initialization of state information of each node deployed in the target area; Step 2: each node periodically broadcasts a HELLO probe packet carrying node ID, position, speed and energy information; The node obtains neighbor state information by listening to the HELLO message, calculates five indexes of spatial proximity, speed following coefficient, average link retention rate, modified node degree difference and energy remaining rate, and broadcasts the index values outside through the METRIC message; Step 3: the node constructs a local index matrix according to the index transfer message listened to, determines the weight of each index, and obtains the cluster head election score through weighted calculation; Step 4: each node broadcasts a CHELEC message carrying its election score; The node receives the election score message of the neighbor node, arranges the candidate cluster head list in descending order according to the score, checks whether the score of the node is higher than the highest score in the candidate cluster head list, if yes, the node is self-boosted as a cluster head, updates the state of the node to "network access", and broadcasts a CHDECL message carrying the cluster head declaration; if not, step 5 is executed for cluster head coexistence judgment; Step 5: Each node dynamically calculates the number of cluster heads N required by the current neighborhood according to the size of its neighbor nodes CH ; when the node score is lower than the highest score in the candidate cluster head list, it is determined whether it can be a coexisting cluster head according to the ranking; if it ranks ahead of N CH , the node can be a coexisting cluster head, and broadcasts a CHDECL message carrying a cluster head declaration. Otherwise, continue to wait for the cluster access opportunity; Step 6: the node receives all neighbor broadcasted CHDECL messages carrying the cluster head declaration, checks whether the message source contains the candidate with the highest score in the candidate cluster head list; if yes, the network access handshake process is performed: sending a JOINREQ message to the cluster head, waiting for the cluster head to return a JOINACPT message; after receiving the JOINACPT message sent by the cluster head, broadcasting a CMNOTIF message to notify the confirmation of joining, and updating the state of the node to "network access"; if not, after checking all received neighbor CMNOTIF messages, the following judgment is performed: checking whether the message source exists in the candidate cluster head list, if yes, removing the node sending the message from the list to prevent the cluster head node from inhibiting the pending node from joining the cluster; if not, judging whether the node has completed the cluster access, if not, requerying the candidate cluster head list and selecting to join the cluster head with the highest remaining score; if the node cannot meet the cluster access condition all the time, step is executed to re-calculate and broadcast the weighted score of the node; The above process will be executed in a loop until all nodes in the network complete the cluster access; Step 7: a cluster maintenance mechanism is established to ensure the stability and reliability of the network, including a HELLO-ACK interaction mechanism to periodically check the connection state with the cluster member nodes, and a backup cluster head announcement mechanism to realize emergency switching in case of failure; Step 8: a source node initiates a communication demand to establish a multi-path routing, and the path discovery process is divided into intra-cluster routing discovery and inter-cluster routing discovery; when an upper-layer application of a node generates a data transmission demand, the node becomes a source node; When the source node generates a communication demand, it first queries the local neighbor table: if the destination node is included, one-hop transmission is directly performed; if the destination node is not found, a routing request is sent to the cluster head to which the source node belongs; then, the source cluster head judges whether the destination node is located in the same cluster according to the topology of the cluster: if yes, the source cluster head starts intra-cluster routing and completes data forwarding according to the intra-cluster routing table; otherwise, the source cluster head starts inter-cluster routing.

2. The clustering multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: In step 2, each index is calculated by the following method: (1) Node n i The normalized spatial proximity P i is defined as: where N i represents the set of symmetric one-hop neighbors of node n i , p i (t) = [x i (t), y i (t), z i (t)] represents the position of node n i in the x, y, z axes at time t, and R represents its communication radius; p j is the position vector of neighbor node n j at time t. (2) Node n i Speed following coefficient F i is defined as: wherein representing node n i Velocity of the x, y, z axes at time t (3) Node n i The stability of the communication link between the node n and all its neighbors is averaged to obtain the average link retention rate R i is: In the formula, is the normalized predicted link survival time for the link L ij . delta ij is the link lifetime, denoting the time for which a node u j is in contact with u i is the time period during which a valid communication connection is maintained while moving; s ij is the relative velocity between two nodes; (4) Correcting the node degree difference In the ideal state, the optimal cluster size in the distributed network of N nodes is Therefore, the ideal node degree is defined as D ideal is the ideal node degree; (5) Node n i of the energy remaining ratio is defined as follows: where E rem,i = E o - E C,i represents the residual energy of a node, E o is the nominal energy; E C,i is the energy consumed by a node n i .

3. The clustering multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: In step 3, the dynamic weight of the index is calculated based on the entropy value method, and the method is: Firstly, in FANET, each node collects the information of position, velocity, energy and node degree of surrounding neighbor nodes through the neighbor discovery process, and calculates its own indicators; then, the node broadcasts the indicator values to the neighbor nodes, and constructs a local indicator matrix X with dimension of (|N b |+1) x m; where, (|N b |+1) represents the total number of the current node and its neighbor nodes, and m is the number of indicators; the matrix element x ij represents the jth indicator value of the ith neighbor node. Further, the uncertainty of the jth indicator is measured by information entropy E j ; the higher the information entropy, the less information the indicator provides in the current network environment, and the lower its importance. Redundancy D j is calculated according to the information entropy, to reflect the correlation between indicators and their repeated contribution to the overall information. Finally, the dynamic weight of the index is calculated, and the weight distribution follows the principle of entropy value method, that is, the index with lower information entropy is given higher weight, ensuring that the index with high information quantity occupies a larger proportion in decision-making.

4. The clustering multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: In step 7, the HELLO-ACK interaction mechanism is as follows: the cluster head periodically broadcasts the HELLO message to the member nodes, and the member nodes respond with the ACK message after receiving it; if the CH does not receive the ACK response of a certain CM within the preset bT time, it is determined that the CM has left its control range, and it is removed from the neighbor list and routing information; where T is the HELLO message period, and b is an adjustable parameter whose value is dynamically adjusted according to the network environment, task requirements and node role.

5. The cluster-based multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: In step 7, the backup cluster head announcement mechanism has the following specific process: ① BCH selection and announcement: the CH selects the node with the highest weighted score from the member list as the BCH, and broadcasts its identity information in the BCH field of the HELLO message; ② BCH role activation: the selected node marks itself as "BCH" after receiving the message, and immediately activates the role and synchronizes the state information to ensure seamless takeover when the CH fails, ensuring the continuity of communication; ③ Member node processing: other nodes extract the BCH field and query the local neighbor list after receiving the message: if the node exists in the neighbor list, mark it as a BCH manageable node; If the node is not found and the relay node of the BCH contains only the CH, mark the node as a non-sensitive node, and set the cluster head maintenance period to two HELLO periods to quickly exit the cluster when the CH fails; When the backup cluster head does not receive the cluster head confirmation message sent by the CH in ① within the preset time, it is determined that the cluster head has failed, and the backup cluster head upgrades itself to "CH" and broadcasts a new HELLO message; The backup cluster head manageable node updates the cluster head information and joins the new cluster; the non-sensitive node waits for the cluster head to expire and then executes the cluster exit and entry process.

6. The clustering multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: In step 8, the multi-path routing method is as follows: First, the RREQ message is expanded with the following fields: source cluster head ID, path stability, and path load; the source cluster head ID identifies the initiating cluster of the routing request and is used for loop detection and path tracing; link stability η ij Reference quantified standard link survival time; link load ι ij It depends on the node with higher queue occupancy rate at both ends of the link, that is, for link L ij In terms of in and Represents node u respectively i and u j The queue occupancy rate; RREQ message can be uniquely identified by the triple <source cluster head ID, source node ID, sequence number>; Subsequently, the source cluster head starts inter-cluster routing: the source cluster head broadcasts RREQ to all nodes in the cluster; ordinary cluster members receive RREQ and discard it directly; inter-cluster relays receive RREQ and check the source cluster head ID of RREQ: if RREQ comes from the cluster head of the cluster, the ICR forwards RREQ to the cluster heads of other clusters connected by it, realizing cross-cluster propagation; if RREQ comes from the cluster head of other cluster, the ICR forwards RREQ to the cluster head of the cluster, so that the CH collects path information for path information aggregation; When RREQ reaches the CH that can address the destination node, the destination cluster head generates a route reply RREP and sends it back to the source cluster head along the reverse path of RREQ, completing route establishment.

7. The clustering multi-path routing method for high-mobility ad hoc networks of claim 5, wherein: After receiving multiple identical RREQs, the destination cluster head selects the optimal path as the main path and at most two suboptimal paths as backup paths according to the path weight index; where the path weight index is calculated according to the path stability, path load and hop count of the current path, and is sorted by priority.

8. The clustering multi-path routing method for high-mobility ad hoc networks of claim 1, wherein: After the multi-path routing is established, an adaptive multi-path transmission strategy based on network dynamic characteristics is adopted to realize efficient and reliable data transmission through path management; in terms of data transmission, a differentiated strategy is adopted according to the dynamic change characteristics of the network topology: for a high dynamic network environment, a reliable transmission scheme based on redundant coding is adopted, and the source node sends duplicate data packets in parallel through two or more independent paths; in the case of a relatively stable network topology, a multi-path concurrent transmission mechanism based on load balancing is enabled, and the data flow is dynamically scheduled to different paths through flow balancing distribution.

9. The implementation device based on the cluster multi-path routing method for high dynamic mobile ad hoc network according to claim 1, characterized in that: A hierarchical modular architecture is adopted to integrate various functional modules and work cooperatively to cope with high dynamic network environment; the upper layer is an application module responsible for generating data transmission requests according to user or system task requirements, and scheduling the middle layer routing module to determine the best transmission path, and then starting data transmission; The middle layer includes a weighted dynamic clustering module and a multi-path routing module; the weighted dynamic clustering module is deployed in all network nodes, which includes an index calculation module for calculating various indexes of node participation in cluster head election according to network topology information, and a dynamic weight distribution module based on entropy method for adaptively adjusting the index weight according to the real-time state of the network; The multi-path routing module realizes efficient establishment of inter-cluster routing and aggregation of path information through the inter-cluster multi-path routing mechanism by expanding fields in the RREQ packet; at the same time, the destination cluster head selects the optimal path as the main path and at most two suboptimal paths as backup paths according to the path weight index; The above-mentioned middle layer also includes a cluster maintenance module, which ensures the stability and continuity of the network under the condition of dynamic change and failure of nodes through the HELLO-ACK interaction mechanism and the backup cluster head notification mechanism; the HELLO-ACK interaction mechanism adopts a differentiated strategy to realize efficient and reliable data transmission according to the dynamic change characteristics of the network topology; in a high dynamic network environment, a reliable transmission scheme based on redundant coding is adopted; in the case of a relatively stable network topology, a multi-path concurrent transmission mechanism based on load balancing is enabled, and the network state is sensed in real time to achieve dynamic balance between reliability and transmission efficiency; The bottom layer is a communication module responsible for basic data transmission and interacting with neighbor nodes.

10. A readable storage medium storing a computer program based on the clustering multi-path routing method for high-mobility mobile ad hoc networks according to claim 1, wherein the computer program is executable by a processor to implement the steps of any one of the clustering multi-path routing methods for high-mobility mobile ad hoc networks; the readable storage medium comprises a physical storage medium and a virtual storage medium; wherein, The virtual storage medium is an electrical carrier signal or a telecommunications signal or a software distribution medium.

11. The device for implementing the clustering multi-path routing method for high-mobility ad hoc networks according to claim 1, wherein: The physical entity includes a plurality of unmanned flight platforms with the same data forwarding capability, without ground infrastructure and centralized control center; the device module of each node includes a memory, a processor, and a computer program stored on the memory and executable on the processor; the device includes a processor, a memory, a network interface and a database connected through a system bus. The virtual storage medium is an electrical carrier signal or a telecommunications signal or a software distribution medium. The physical entity includes a plurality of unmanned flight platforms with the same data forwarding capability, without ground infrastructure and centralized control center; the device module of each node includes a memory, a processor, and a computer program stored on the memory and executable on the processor; the device includes a processor, a memory, a network interface and a database connected through a system bus.

Images