A method for inferring a topology of a drone swarm network

Abstract

The application discloses a UAV cluster network topology inference method, which takes power as a variable, takes network robustness and total energy consumption as a target function, and utilizes a non-dominated sorting genetic algorithm NSGA-II to perform double-target optimization to find optimal node power configuration, and then calculates the communication ranges of all nodes according to a power signal propagation range calculation formula, so that the network topology of the UAV cluster is determined. The method has certain correctness, and creates a prerequisite for implementing UAV cluster countermeasures based on network topology key node attacks.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) technology, specifically relating to a method for inferring the topology of a UAV swarm network. Background Technology

[0002] Due to their small size, low cost, and high maneuverability, drones are effectively applied in various fields. A drone swarm consists of several to hundreds of small drones that work together to execute a target mission. The malicious use of drones is extremely dangerous; to prevent drone swarms from being used for malicious actions, it is necessary to design an effective drone swarm countermeasure system.

[0003] Drone countermeasure systems typically employ a two-step process: first, detecting the target drone, and then eliminating the malicious drone based on the detected state information. Existing countermeasure systems primarily detect and locate the target drone to obtain its model and location information. For example, the paper "UAV Detection and Localization Based on Multi-dimensional Signal Features" in the IEEE Sensors Journal utilizes a circular array antenna to collect channel state information of the drone's radio frequency signals, extracting spatial features such as azimuth (AOA) and elevation (AOE) to achieve 3D positioning. For countering a single drone, the system only needs to detect and locate it; once its position is known, offensive or defensive actions can be taken. However, drone swarms contain numerous drones, and current technology has limited detection range. The time available for the system to counteract a detected drone is very short, insufficient for targeted strikes against each individual drone. Drone swarms interact through a network, so a feasible approach is to attack only the key nodes within the swarm, disrupting the network topology's connectivity and thus the swarm's collaborative capabilities. A necessary prerequisite for this method is the ability to perceive and infer the network topology of the drone swarm.

[0004] Network topology inference generally refers to inferring the communication relationships of a target network based on prior information, such as inter-node interaction signals, node locations, and signal transmission power, thereby deducing the target network's topology. Methods based on inter-node interaction information are mostly based on the logic that a communication packet sent by a sender will induce potential neighboring nodes to send an acknowledgment packet. For example, the paper "Topologytracking of dynamic UAV wireless networks" in the *Chinese Journal of Aeronautics* uses this logic to infer the wireless network topology of a drone swarm. However, for large-scale drone swarms, acquiring their communication signals is not easy, and even after acquiring the signals, blind source separation techniques are needed to separate the signal sources, a time-consuming process unsuitable for drone swarm countermeasures scenarios with high real-time requirements. Another method uses the signal transmission power of nodes to calculate their communication coverage range, and then determines a network topology based on node location information. For example, in the IEEE conference paper "Research on non-cooperative topology inference method based on node location information," the authors determine whether a communication link exists between two nodes based on node location and critical transmission distance. This method is fast. In current practices, the communication coverage of nodes is mostly determined based on the maximum signal transmission power of the drone. However, nodes in a drone swarm may adjust their signal transmission power during flight, thereby adjusting their communication coverage. Therefore, this method is not precise. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a method for inferring the network topology of a drone swarm. This method uses power as a variable and network robustness and total energy consumption as objective functions. It employs the Non-Dominated Sorting Genetic Algorithm (NSGA-II) for bi-objective optimization to find the optimal node power configuration. Then, based on the formula for calculating the propagation range of power signals, the communication range of each node is calculated, thereby determining the network topology of the drone swarm. This method has a certain degree of accuracy, creating a prerequisite for implementing drone swarm countermeasures based on attacks targeting key nodes in the network topology.

[0006] The technical solution adopted by this invention to solve its technical problem includes the following steps:

[0007] Step 1: Obtain the maximum signal transmission power and location of the drones through drone swarm detection and positioning technology;

[0008] Step 2: The drone swarm has N nodes. Given the known positions of all drone nodes L = (l1, l2, ..., l...),... N ) and the maximum signal transmission power Pmax of all drones = (p 1max ,p 2max ,...,p Nmax Then, using the power variation range as a variable, the network topology that meets the robustness and energy consumption indicators is optimized.

[0009] Step 2-1: Quantify robustness metrics: Network robustness is measured in two ways: first, the network's robustness to random attacks, and second, the network's robustness to targeted attacks; robustness is measured by the size of the largest connected component of the network after removing a number of nodes.

[0010] Equation (1) is the random robustness index:

[0011]

[0012] Where S r (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after randomly removing Q×N nodes; m represents the current attack round, M represents the total attack rounds, and Q represents the proportion of nodes removed to the total number of nodes.

[0013] In the original network, nodes 1, 2, ..., N are randomly removed, meaning N attacks are performed. After each attack, S needs to be calculated. r (Q), finally calculate S for N attacks. r (Q) Average value, which is counted as one round of attack; to prevent fluctuations in the random robustness index, the average value is calculated after M rounds of such attacks.

[0014] The robustness metrics of a network to targeted attacks are as follows:

[0015]

[0016] Where S t (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after removing QN nodes based on degree sorting.

[0017] First, sort the original network nodes in descending order of degree. When removing a node, remove the node with the highest degree first. Targeted attacks on nodes are not random, so the result of each round of attacks is the same, and there is no need to perform multiple rounds of attacks.

[0018] Drone swarm networks are robust to both random and targeted attacks, so the following metrics are used to measure network robustness:

[0019] R = Rr +R t (3)

[0020] Step 2-2: Quantifying Energy Consumption Indicators: Let the observation window time be t, then the energy consumption of a single node is e. i =t*p i Total energy consumption is Right now

[0021]

[0022] Where, p i This represents the power of the i-th drone;

[0023] Step 3: Use the non-dominated sorting genetic algorithm NSGA-II to perform bi-objective optimization on the two indicators of equations (3) and (4). The specific steps are as follows:

[0024] Step 3-1: During the optimization process, the power P = (p1, p2, ..., p N ) as a variable, where p i ∈(0,p imax Without considering path loss, the distance at which the signal can be reliably received under a specific power is estimated according to equation (5), which is to calculate the communication coverage radius of the UAV node under a specific power:

[0025] d i (km)=10(p i -32.44-20lg(f)) / 20 (5)

[0026] Where p i The unit is dBm; f is the communication frequency, and the unit is MHz;

[0027] As can be seen from equation (5), when the communication frequency f is determined, the power p of node i is... i The larger the size, the wider the communication coverage.

[0028] Step 3-2: Obtain the communication coverage radius D of all nodes from equation (5) = (d1, d2, ..., d N Then, based on the positions of all nodes L = (l1, l2, ..., l...), NThe distances between different nodes are calculated, and a topology of the UAV swarm network is established based on the following principles: if the distance between two UAVs is within each other's communication range, then there is an edge between the two UAVs; if either of them is outside the communication range of the other, there is no edge; if the power changes, the topology will also change, and the robustness index of the topology will also change. Therefore, the change of power will cause changes in the robustness index and energy consumption index. Thus, power is used as a variable, and NSGA-II is used to perform bi-objective optimization of topology robustness and energy consumption.

[0029] Step 3-3: Define two objective functions: Since in bi-objective optimization, the smaller the objective value, the better the individual, and the larger the objective value of the robustness index (3), the higher the robustness, the two objective functions are determined as follows:

[0030] H1 = 1 / R = 1 / (R) r +R t (6)

[0031]

[0032] Step 4: Optimize to obtain the final drone swarm network topology.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. This invention does not require prior acquisition of communication signals between drone nodes, overcoming the time-consuming problem of data acquisition and processing in passive network traffic observation methods.

[0035] 2. The method of the present invention takes into account that when UAV nodes are in flight, they will adjust their power to maintain the robustness of the entire cluster network and save energy, which will lead to dynamic changes in the communication coverage, rather than just the communication distance that can be reached based on location information and maximum communication power, thus having a high degree of accuracy. Attached Figure Description

[0036] Figure 1 (a) and Figure 1 (b) Drone swarm scene diagrams with 16 nodes and 43 nodes respectively.

[0037] Figure 2 (a) and Figure 2 (b) Pareto fronts for network topology inference using Algorithm 1 for drone swarms of 16 and 43 nodes, respectively.

[0038] Figure 3 (a) and Figure 3 (b) The network topologies of the drone clusters with 16 nodes and 43 nodes, respectively, were inferred using Algorithm 1.

[0039] Figure 4 (a) and Figure 4 (b) Network topologies of 16-node and 43-node drone clusters, respectively, inferred using explicit network topology inference methods. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] To overcome the shortcomings of methods based on node location information and communication coverage that fail to consider dynamic power changes, this invention proposes a method for inferring the network topology of a UAV swarm. This method uses power as a variable and network robustness and total energy consumption as objective functions. It employs the Non-Dominated Sorting Genetic Algorithm (NSGA-II) for bi-objective optimization to find the optimal node power configuration. Then, based on the formula for calculating the propagation range of power signals, the communication range of each node is calculated, thereby determining the network topology of the UAV swarm. This method possesses a certain degree of accuracy, creating a prerequisite for implementing UAV swarm countermeasures based on attacks targeting key nodes in the network topology.

[0042] To achieve the above-mentioned objectives, the technical solution proposed by this invention is as follows:

[0043] S1: Assuming that the maximum signal transmission power and location of drones can be obtained through drone swarm detection and positioning technology, the network topology of the drone swarm is inferred based on the following principles: Drones in the swarm continuously adjust their power during flight to adjust their communication coverage, thus maintaining a certain level of robustness in the overall network topology. However, due to power consumption limitations, the power cannot be directly increased to the maximum. Furthermore, if drones are close together, power consumption can be reduced to save energy. Therefore, the network topology of a drone swarm must maintain network robustness while keeping communication power consumption at a reasonable level.

[0044] S2: Based on this principle, the present invention deduces the network topology of a drone swarm using the following method: The number of nodes in the drone swarm is N, and the positions of all drone nodes are known to be L = (l1, l2, ..., l...). N ) and the maximum signal transmission power Pmax of all drones = (p 1max ,p 2max ,...,p Nmax Then, using the power variation range as a variable, an optimal network topology that satisfies the principles of robustness and energy conservation is found. This network topology serves as the inferred network topology for the drone swarm.

[0045] S3: Quantitative Robustness Metrics: Network robustness is often measured from two aspects: first, the network's robustness against random attacks, and second, the network's robustness against targeted attacks. Robustness is measured by the size of the largest connected component in the network after removing several nodes. Equation (1) is a random robustness metric:

[0046]

[0047] Where S r (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after randomly removing QN nodes. In the original network, 1, 2, ..., N nodes are randomly removed, meaning N attacks are performed. After each attack, S needs to be calculated. r (Q). Finally, calculate S for N attacks. r (Q) The average value, which is counted as one round of attack. To prevent fluctuations in random robustness metrics, the average value is calculated after M rounds of such attacks. The network's robustness metrics against targeted attacks are as follows:

[0048]

[0049] Where S t (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after removing QN nodes sorted by degree. First, the original network nodes are sorted in descending order of degree; when removing nodes, the node with the highest degree is removed first. Targeted attacks on nodes do not possess randomness, therefore the result of each round of attack is the same, eliminating the need for multiple rounds. Drone swarm networks generally exhibit robustness to both random and targeted attacks; therefore, the following metrics are used to measure network robustness:

[0050] R = R r +R t (3)

[0051] S4: Quantitative Energy Consumption Index: Let the observation window time be t, then the energy consumption of a single node is e. i =t*p i Total energy consumption is Right now

[0052]

[0053] S5: The non-dominated sorting genetic algorithm NSGA-II is used to perform bi-objective optimization on the two indicators of equations (3) and (4). The specific steps are as follows:

[0054] S5-1: During the optimization process, the power P = (p1, p2, ..., p N ) as a variable, where p i ∈(0,p imaxWithout considering path loss, the distance at which the signal can be reliably received under a specific power is estimated according to equation (5), which is to calculate the communication coverage radius of the UAV node under a specific power:

[0055] d i (km)=10(p i -32.44-20lg(f)) / 20 (5)

[0056] Where p i The unit is dBm (decibels), and f is the communication frequency, with the unit being MHz. From equation (5), it can be seen that when the communication frequency f is determined, the power p of node i... i The larger the value, the greater the communication coverage.

[0057] S5-2: From equation (5), the communication coverage radius D of all nodes is obtained as D = (d1, d2, ..., d N Then, based on the positions of all nodes L = (l1, l2, ..., l...), N The distances between different nodes are calculated, allowing the establishment of a drone swarm network topology based on the following principles: if two drones are both within each other's communication range, then an edge exists between them; if either drone is outside the other's communication range, no edge exists. Power changes will alter the topology, consequently affecting its robustness metrics. Therefore, power changes will cause changes in the two metrics defined in the previous section: robustness and energy consumption. Thus, using power as a variable, NSGA-II is employed for bi-objective optimization of topology robustness and energy consumption.

[0058] S5-3: Define two objective functions: Since in bi-objective optimization, the smaller the objective value, the better the individual, and the larger the objective value of the robustness index (3), the higher the robustness, the two objective functions are determined as follows:

[0059] H1 = 1 / R = 1 / (R) r +R t (6)

[0060]

[0061] S5-4: The network topology inference algorithm based on NSGA-II bi-objective optimization is described in Algorithm 1:

[0062]

[0063] Specific implementation examples:

[0065] To achieve more effective cyberspace situational awareness of drone swarms, it is necessary to quickly and effectively infer the network topology of the drone swarm. Based on the technical solution, this is implemented in the following manner:

[0066] The layout is as shown in the attached document. Figure 1 The two drone cluster scene diagrams in (a) and (b) have 16 nodes and 43 nodes respectively. A communication network protocol is added to each drone node, and the location information and power information in the scene are read.

[0067] Based on the node location and power information, the parameters of Algorithm 1 are initially configured, and Algorithm 1 is used to implicitly infer the communication network topology of the UAV swarm. (See attached diagram.) Figure 2 (a) and (b) show the Pareto fronts generated when speculating on the topologies of two UAV swarm networks. The individual selected in the last step of Algorithm 1 is the first individual in the Pareto front. Each individual entity has a corresponding power configuration that optimizes network robustness and energy consumption while keeping both relatively balanced.

[0068] After identifying the non-dominant individuals, the communication power configuration of each node is determined. Based on the calculation formula (5) for the communication coverage radius of a UAV node under a specific power, the communication coverage of each node is determined. Then, based on the location information of the nodes in the UAV swarm obtained during the detection and positioning phase, the distance between all node pairs is calculated to determine the connectivity between nodes. If two nodes are within each other's communication coverage, there is a connection between them; otherwise, there is no connection. The UAV swarm network topologies with node sizes of 16 and 43, inferred using Algorithm 1, are shown below. Figure 3 As shown in (a) and (b), * represents a node and lines represent connections between nodes.

[0069] • Verifying Correctness: In a drone swarm, the AODV routing protocol is configured for each node, and CBR traffic is added at the application layer. Then, the network topology of the drone swarm is explicitly inferred by tracing back the RREP messages of the AODV protocol. This method infers the information transmission path from the destination node to the source node, but not the complete topology. Therefore, correctness is verified by checking the overlap between the edges in the explicitly inferred topology and the implicitly inferred topology. After comparative observation, [the following is a continuation of the previous sentence]. Figure 4The edges between most nodes are a subset of the results implicitly inferred by Algorithm 1, meaning that most edges overlap with the implicit inference results. To further verify the implicit inference results, implicit and explicit topology inferences were performed on more drone swarms of different sizes. The correctness was verified by observing the overlap between the implicitly inferred edges and the explicit inference results. The proportion of overlapping edges in the explicit topology inferences was calculated as the accuracy rate. Table 1 shows the accuracy rate calculation results for drone swarms of different sizes with 16, 30, 43, and 55 nodes. For drone swarms of different sizes, the accuracy rate of network topology inference is above 70%, further demonstrating that the algorithm has a certain degree of correctness.

[0070] In summary, this network topology inference method can guarantee a certain degree of accuracy and is applicable to drone swarms of different sizes.

[0071] applicability.

[0072] Table 1. Network Topology Inference Accuracy Rate

[0073]

Claims

1. A method for topology inference in unmanned aerial vehicle (UAV) swarm networks, characterized in that, Includes the following steps: Step 1: Obtain the maximum signal transmission power and location of the drones through drone swarm detection and positioning technology; Step 2: The drone swarm has N nodes. Given the known positions of all drone nodes L = (l1, l2, ..., l...),... N ) and the maximum signal transmission power Pmax of all drones = (p 1max ,p 2max ,...,p Nmax Then, using the power variation range as a variable, the network topology that meets the robustness and energy consumption indicators is optimized. Step 2-1: Quantify robustness metrics: Network robustness is measured in two ways: first, the network's robustness to random attacks, and second, the network's robustness to targeted attacks; robustness is measured by the size of the largest connected component of the network after removing a number of nodes. Equation (1) is the random robustness index: Where S r (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after randomly removing Q×N nodes; m represents the current attack round, M represents the total attack rounds, and Q represents the proportion of nodes removed to the total number of nodes. In the original network, nodes 1, 2, ..., N are randomly removed, meaning N attacks are performed. After each attack, S needs to be calculated. r (Q), finally calculate S for N attacks. r (Q) Average value, which is counted as one round of attack; to prevent fluctuations in the random robustness index, the average value is calculated after M rounds of such attacks. The robustness metrics of a network to targeted attacks are as follows: Where S t (Q) represents the proportion of nodes in the largest connected component to the total number of nodes after removing QN nodes based on degree sorting. First, sort the original network nodes in descending order of degree. When removing a node, remove the node with the highest degree first. Targeted attacks on nodes are not random, so the result of each round of attacks is the same, and there is no need to perform multiple rounds of attacks. Drone swarm networks are robust to both random and targeted attacks, so the following metrics are used to measure network robustness: R=R r +R t (3) Step 2-2: Quantifying Energy Consumption Indicators: Let the observation window time be t, then the energy consumption of a single node is e. i =t*p i Total energy consumption is Right now Where, p i This represents the power of the i-th drone; Step 3: Use the non-dominated sorting genetic algorithm NSGA-II to perform bi-objective optimization on the two indicators of equations (3) and (4). The specific steps are as follows: Step 3-1: During the optimization process, the power P = (p1, p2, ..., p N ) as a variable, where p i ∈(0,p imax Without considering path loss, the distance at which the signal can be reliably received under a specific power is estimated according to equation (5), which is to calculate the communication coverage radius of the UAV node under a specific power: d i (km)=10(p i -32.44-20lg(f)) / 20 (5) Where p i The unit is dBm; f is the communication frequency, and the unit is MHz; As can be seen from equation (5), when the communication frequency f is determined, the power p of node i is... i The larger the size, the wider the communication coverage. Step 3-2: Obtain the communication coverage radius D of all nodes from equation (5) = (d1, d2, ..., d N Then, based on the positions of all nodes L = (l1, l2, ..., l...), N The distances between different nodes are calculated, and a topology of the UAV swarm network is established based on the following principles: if the distance between two UAVs is within each other's communication range, then there is an edge between the two UAVs; if either of them is outside the communication range of the other, there is no edge; if the power changes, the topology will also change, and the robustness index of the topology will also change. Therefore, the change of power will cause changes in the robustness index and energy consumption index. Thus, power is used as a variable, and NSGA-II is used to perform bi-objective optimization of topology robustness and energy consumption. Step 3-3: Define two objective functions: Since in bi-objective optimization, the smaller the objective value, the better the individual, and the larger the objective value of the robustness index (3), the higher the robustness, the two objective functions are determined as follows: H1=1 / R=1 / (R r +R t ) (6) Step 4: Optimize to obtain the final drone swarm network topology.

Images